{ "schema_version": 2, "adapter_key": "grpo-difficulty-curriculum", "manifest_ref": "reproduction:grpo-difficulty-curriculum", "method": "grpo-difficulty-curriculum", "measured_locally": true, "diagnostic_only": true, "promotion_eligible": false, "not_gpu_receipt": true, "seeds": [ 42, 43, 44 ], "actual_cuda_run_count": 15, "execution_complete": true, "all_seed_arm_mechanisms_or_negative_controls_accepted": true, "whole_method_gpu_validation_passed": true, "default_three_seed_plan_completed": true, "seed_results": [ { "seed": 42, "device": "cuda", "diagnostic_only": true, "variants": { "fixed-medium": { "method": "grpo-difficulty-curriculum", "seed": 42, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1409, "truncated_responses": 0 }, "training": [ { "step": 1, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 1.0 }, { "training_row": 1, "eligible": false, "current_success": 0.125 }, { "training_row": 2, "eligible": false, "current_success": 0.0 }, { "training_row": 3, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0 ], "loss": -2.2351741790771484e-08, "gradient_l2_norm": 0.16067364811897278, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 1.2, 1.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776 ] }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 0.875 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.125 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1431, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 1 }, { "step": 2, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0 ], "loss": -7.450580596923828e-09, "gradient_l2_norm": 0.17742520570755005, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 0.2, 1.2, 0.2, 1.2, 0.2 ], "advantages": [ 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829 ] }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0 ], "loss": -0.000545695424079895, "gradient_l2_norm": 0.16647948324680328, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 1.2, 1.2, 1.2, 1.2, 0.2, 0.2 ], "advantages": [ 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, -1.2909917831420898 ] }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": false, "current_success": 1.0 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.125 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1316, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 3, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 1.0 }, { "training_row": 1, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0 ], "loss": -2.9802322387695312e-08, "gradient_l2_norm": 0.15451529622077942, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 0.2, 1.2, 1.2, 0.2, 0.2 ], "advantages": [ -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, 1.2909917831420898, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829 ] }, { "training_row": 2, "eligible": false, "current_success": 0.0 }, { "training_row": 3, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0 ], "loss": -0.0021705925464630127, "gradient_l2_norm": 0.150689035654068, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 1.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898 ] }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 0.875 }, { "training_row": 6, "eligible": false, "current_success": 0.125 }, { "training_row": 7, "eligible": false, "current_success": 0.125 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1353, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 4, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.75 }, { "training_row": 1, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0 ], "loss": -2.2351741790771484e-08, "gradient_l2_norm": 0.17810259759426117, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2 ], "advantages": [ -1.2909917831420898, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829 ] }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": false, "current_success": 0.75 }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0 ], "loss": 0.0004472807049751282, "gradient_l2_norm": 0.19432415068149567, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 1.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776 ] }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.125 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1409, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 2 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8250, "student": 70009 }, "scored_response_tokens": { "behavior": 70009, "student_rl": 13880 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "fixed-medium", "difficulty_history": [ [ 1.0, 0.125, 0.0, 0.5, 0.0, 0.875, 0.0, 0.125 ], [ 0.875, 0.375, 0.125, 0.625, 0.125, 1.0, 0.0, 0.125 ], [ 1.0, 0.375, 0.0, 0.375, 0.0, 0.875, 0.125, 0.125 ], [ 0.75, 0.625, 0.125, 0.75, 0.125, 0.5, 0.0, 0.125 ] ], "difficulty_dynamics": { "span": [ 0.25, 0.5, 0.125, 0.375, 0.125, 0.5, 0.125, 0.0 ], "total_variation": [ 0.5, 0.5, 0.375, 0.75, 0.375, 0.625, 0.25, 0.0 ], "extra_round_trip_variation": [ 0.25, 0.0, 0.25, 0.375, 0.25, 0.125, 0.125, 0.0 ] }, "initial_static_easy_to_hard_order": null, "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] }, "staged": { "method": "grpo-difficulty-curriculum", "seed": 42, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.0, "examples": 4, "generated_tokens": 1597, "truncated_responses": 1 }, "training": [ { "step": 1, "band": [ 0.625, 0.875 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 1.0 }, { "training_row": 1, "eligible": false, "current_success": 0.125 }, { "training_row": 2, "eligible": false, "current_success": 0.0 }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "loss": -3.725290298461914e-09, "gradient_l2_norm": 0.11954058706760406, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "advantages": [ -2.6457433700561523, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.125 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1508, "truncated_responses": 0 }, "transition": "new-best", "effective_updates": 1 }, { "step": 2, "band": [ 0.625, 0.875 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0 ], "loss": -7.450580596923828e-09, "gradient_l2_norm": 0.16927285492420197, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 0.2, 0.2 ], "advantages": [ 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, -1.7320468425750732 ] }, { "training_row": 1, "eligible": false, "current_success": 0.5 }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": false, "current_success": 0.25 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 0.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "loss": 0.00010900944471359253, "gradient_l2_norm": 0.15238478779792786, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "advantages": [ -1.7320468425750732, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.25 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1199, "truncated_responses": 0 }, "transition": "keep", "effective_updates": 2 }, { "step": 3, "band": [ 0.625, 0.875 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0 ], "loss": -7.450580596923828e-09, "gradient_l2_norm": 0.16443920135498047, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2, 1.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 1, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0 ], "loss": 0.0010845735669136047, "gradient_l2_norm": 0.19610214233398438, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 1.2, 1.2, 0.2, 0.2, 1.2, 1.2 ], "advantages": [ 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829 ] }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0, 0.0 ], "loss": 0.000712469220161438, "gradient_l2_norm": 0.16621491312980652, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, -1.2909917831420898, 0.7745950818061829, -1.2909917831420898 ] }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0 ], "loss": 0.0016031339764595032, "gradient_l2_norm": 0.21787437796592712, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2 ], "advantages": [ 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1521, "truncated_responses": 1 }, "transition": "advance", "effective_updates": 4 }, { "step": 4, "band": [ 0.5, 0.75 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0 ], "loss": -7.450580596923828e-09, "gradient_l2_norm": 0.19066661596298218, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 1.2, 1.2, 1.2, 0.2, 1.2 ], "advantages": [ -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244 ] }, { "training_row": 1, "eligible": false, "current_success": 0.375 }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0 ], "loss": 0.0001943632960319519, "gradient_l2_norm": 0.14625507593154907, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 0.2, 0.2, 1.2, 0.2, 1.2 ], "advantages": [ 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776 ] }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0 ], "loss": 0.0008660480380058289, "gradient_l2_norm": 0.1390981376171112, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 1.2, 1.2, 1.2, 1.2, 0.2 ], "advantages": [ 0.7745950818061829, -1.2909917831420898, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898 ] }, { "training_row": 6, "eligible": false, "current_success": 0.125 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.0, "examples": 4, "generated_tokens": 1597, "truncated_responses": 1 }, "transition": "new-best", "effective_updates": 3 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8754, "student": 74288 }, "scored_response_tokens": { "behavior": 74288, "student_rl": 19984 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "staged", "difficulty_history": [ [ 1.0, 0.125, 0.0, 0.5, 0.0, 0.875, 0.0, 0.125 ], [ 0.75, 0.5, 0.125, 0.25, 0.0, 0.75, 0.0, 0.25 ], [ 0.875, 0.625, 0.125, 0.625, 0.0, 0.75, 0.0, 0.0 ], [ 0.75, 0.375, 0.125, 0.5, 0.125, 0.625, 0.125, 0.0 ] ], "difficulty_dynamics": { "span": [ 0.25, 0.5, 0.125, 0.375, 0.125, 0.25, 0.125, 0.25 ], "total_variation": [ 0.5, 0.75, 0.125, 0.75, 0.125, 0.25, 0.125, 0.375 ], "extra_round_trip_variation": [ 0.25, 0.25, 0.0, 0.375, 0.0, 0.0, 0.0, 0.125 ] }, "initial_static_easy_to_hard_order": null, "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] }, "static": { "method": "grpo-difficulty-curriculum", "seed": 42, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1631, "truncated_responses": 0 }, "training": [ { "step": 1, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 1.0, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ] }, { "training_row": 1, "eligible": false, "current_success": 0.125 }, { "training_row": 2, "eligible": false, "current_success": 0.0 }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "loss": -3.725290298461914e-09, "gradient_l2_norm": 0.11954058706760406, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "advantages": [ -2.6457433700561523, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.125 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1508, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 1 }, { "step": 2, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.75 }, { "training_row": 1, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0 ], "loss": 7.450580596923828e-09, "gradient_l2_norm": 0.1844378262758255, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 0.2, 1.2, 0.2, 1.2, 0.2 ], "advantages": [ 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776 ] }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0 ], "loss": 0.002510286867618561, "gradient_l2_norm": 0.13175512850284576, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 1.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244 ] }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.25 } ], "validation_diagnostic": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1431, "truncated_responses": 1 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 3, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": false, "current_success": 0.125 }, { "training_row": 2, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0 ], "loss": 1.4901161193847656e-08, "gradient_l2_norm": 0.18624506890773773, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2, 1.2 ], "advantages": [ -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, 1.7320468425750732 ] }, { "training_row": 3, "eligible": false, "current_success": 0.625 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 0.875 }, { "training_row": 6, "eligible": false, "current_success": 0.125 }, { "training_row": 7, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] } ], "validation_diagnostic": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1631, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 1 }, { "step": 4, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": false, "current_success": 0.375 }, { "training_row": 2, "eligible": false, "current_success": 0.375 }, { "training_row": 3, "eligible": false, "current_success": 0.625 }, { "training_row": 4, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1631, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 0 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 9164, "student": 74639 }, "scored_response_tokens": { "behavior": 74639, "student_rl": 9616 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "static", "difficulty_history": [ [ 1.0, 0.125, 0.0, 0.5, 0.0, 0.875, 0.0, 0.125 ], [ 0.75, 0.5, 0.125, 0.25, 0.0, 0.75, 0.0, 0.25 ], [ 0.875, 0.125, 0.25, 0.625, 0.0, 0.875, 0.125, 0.0 ], [ 0.875, 0.375, 0.375, 0.625, 0.0, 0.75, 0.0, 0.0 ] ], "difficulty_dynamics": { "span": [ 0.25, 0.375, 0.375, 0.375, 0.0, 0.125, 0.125, 0.25 ], "total_variation": [ 0.375, 1.0, 0.375, 0.625, 0.0, 0.375, 0.25, 0.375 ], "extra_round_trip_variation": [ 0.125, 0.625, 0.0, 0.25, 0.0, 0.25, 0.125, 0.125 ] }, "initial_static_easy_to_hard_order": [ 0, 5, 3, 1, 7, 2, 4, 6 ], "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] }, "too-hard": { "method": "grpo-difficulty-curriculum", "seed": 42, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1406, "truncated_responses": 0 }, "training": [ { "step": 1, "band": [ 0.125, 0.375 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 1.0 }, { "training_row": 1, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": -3.725290298461914e-09, "gradient_l2_norm": 0.14079557359218597, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 2, "eligible": false, "current_success": 0.0 }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 0.875 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0 ], "loss": -0.000309772789478302, "gradient_l2_norm": 0.14203859865665436, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757 ] } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1639, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 2, "band": [ 0.125, 0.375 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.75 }, { "training_row": 1, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0 ], "loss": 0.0, "gradient_l2_norm": 0.16137294471263885, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 0.2, 1.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244 ] }, { "training_row": 2, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": 0.0012478604912757874, "gradient_l2_norm": 0.1772703230381012, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ 1.7320468425750732, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244 ] }, { "training_row": 3, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0 ], "loss": 0.0015418380498886108, "gradient_l2_norm": 0.17813342809677124, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 0.2, 1.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829 ] }, { "training_row": 4, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": -0.0006144344806671143, "gradient_l2_norm": 0.14178799092769623, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ 1.7320468425750732, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244 ] }, { "training_row": 5, "eligible": false, "current_success": 1.0 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1551, "truncated_responses": 1 }, "transition": "not-staged", "effective_updates": 4 }, { "step": 3, "band": [ 0.125, 0.375 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 1.0 }, { "training_row": 1, "eligible": false, "current_success": 0.5 }, { "training_row": 2, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0 ], "loss": -2.2351741790771484e-08, "gradient_l2_norm": 0.17610740661621094, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 0.2, 1.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244 ] }, { "training_row": 3, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0 ], "loss": 0.000734969973564148, "gradient_l2_norm": 0.16925451159477234, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 1.2, 0.2, 1.2, 0.2, 0.2 ], "advantages": [ 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829 ] }, { "training_row": 4, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0 ], "loss": 0.00034995004534721375, "gradient_l2_norm": 0.14391998946666718, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 1.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 5, "eligible": false, "current_success": 0.875 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1457, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 3 }, { "step": 4, "band": [ 0.125, 0.375 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": false, "current_success": 0.625 }, { "training_row": 2, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0 ], "loss": -7.450580596923828e-09, "gradient_l2_norm": 0.15005114674568176, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 0.625 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1406, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 1 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8791, "student": 74204 }, "scored_response_tokens": { "behavior": 74204, "student_rl": 23930 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "too-hard", "difficulty_history": [ [ 1.0, 0.125, 0.0, 0.5, 0.0, 0.875, 0.0, 0.125 ], [ 0.75, 0.25, 0.25, 0.375, 0.25, 1.0, 0.0, 0.0 ], [ 1.0, 0.5, 0.25, 0.375, 0.125, 0.875, 0.0, 0.0 ], [ 0.875, 0.625, 0.125, 0.5, 0.0, 0.625, 0.0, 0.0 ] ], "difficulty_dynamics": { "span": [ 0.25, 0.5, 0.25, 0.125, 0.25, 0.375, 0.0, 0.125 ], "total_variation": [ 0.625, 0.5, 0.375, 0.25, 0.5, 0.5, 0.0, 0.125 ], "extra_round_trip_variation": [ 0.375, 0.0, 0.125, 0.125, 0.25, 0.125, 0.0, 0.0 ] }, "initial_static_easy_to_hard_order": null, "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] }, "unfiltered": { "method": "grpo-difficulty-curriculum", "seed": 42, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1495, "truncated_responses": 0 }, "training": [ { "step": 1, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 1.0, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ] }, { "training_row": 1, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": -3.725290298461914e-09, "gradient_l2_norm": 0.14079557359218597, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 2, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 3, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0 ], "loss": 0.0005620643496513367, "gradient_l2_norm": 0.15811462700366974, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 1.2, 1.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776 ] }, { "training_row": 4, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 5, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "loss": 0.0006006360054016113, "gradient_l2_norm": 0.12308721989393234, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "advantages": [ -2.6457433700561523, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 6, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 7, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0 ], "loss": -0.0007472187280654907, "gradient_l2_norm": 0.13807804882526398, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757 ] } ], "validation_diagnostic": { "accuracy": 0.75, "examples": 4, "generated_tokens": 1354, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 4 }, { "step": 2, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0 ], "loss": 0.0, "gradient_l2_norm": 0.13981814682483673, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 0.2, 1.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523, 0.3779633343219757 ] }, { "training_row": 1, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0 ], "loss": 0.001870371401309967, "gradient_l2_norm": 0.18133896589279175, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 0.2, 1.2, 0.2, 1.2 ], "advantages": [ 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244 ] }, { "training_row": 2, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 3, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": 0.001228034496307373, "gradient_l2_norm": 0.14652015268802643, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 4, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0 ], "loss": 0.0009716004133224487, "gradient_l2_norm": 0.15301290154457092, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 5, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0 ], "loss": 0.0008800923824310303, "gradient_l2_norm": 0.15924488008022308, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 0.2, 1.2, 1.2, 1.2, 0.2 ], "advantages": [ 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898 ] }, { "training_row": 6, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 7, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1366, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 5 }, { "step": 3, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0 ], "loss": 1.862645149230957e-08, "gradient_l2_norm": 0.14576801657676697, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 1, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0 ], "loss": 0.0009983032941818237, "gradient_l2_norm": 0.1611718088388443, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 1.2, 0.2, 0.2, 1.2, 1.2 ], "advantages": [ 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776, 0.9999979734420776 ] }, { "training_row": 2, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0 ], "loss": -0.0007110275328159332, "gradient_l2_norm": 0.14535246789455414, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 1.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 3, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0, 1.0 ], "loss": -0.00023868680000305176, "gradient_l2_norm": 0.16780060529708862, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 1.2, 0.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, -1.2909917831420898, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829 ] }, { "training_row": 4, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 5, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0 ], "loss": 0.00010679289698600769, "gradient_l2_norm": 0.14936576783657074, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 0.2, 1.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523, 0.3779633343219757 ] }, { "training_row": 6, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": -0.0007155090570449829, "gradient_l2_norm": 0.12598055601119995, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 7, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1535, "truncated_responses": 1 }, "transition": "not-staged", "effective_updates": 6 }, { "step": 4, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 0.0 ], "loss": 0.0, "gradient_l2_norm": 0.21273058652877808, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 1.2, 1.2, 0.2, 1.2, 0.2 ], "advantages": [ -1.2909917831420898, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, -1.2909917831420898 ] }, { "training_row": 1, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 1.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0 ], "loss": 0.0006969347596168518, "gradient_l2_norm": 0.17742937803268433, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 0.2, 1.2, 1.2, 0.2, 0.2 ], "advantages": [ 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829 ] }, { "training_row": 2, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 3, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0 ], "loss": 0.0008446797728538513, "gradient_l2_norm": 0.16580937802791595, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 0.2, 0.2, 1.2 ], "advantages": [ 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, -1.7320468425750732, 0.5773489475250244 ] }, { "training_row": 4, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 5, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 1.0 ], "loss": -0.004366964101791382, "gradient_l2_norm": 0.3042769432067871, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.0, 1.2, 0.2, 0.2, 1.2, 1.2 ], "advantages": [ 0.9932640194892883, -0.8986674547195435, -1.277053713798523, 0.9932640194892883, -0.8986674547195435, -0.8986674547195435, 0.9932640194892883, 0.9932640194892883 ] }, { "training_row": 6, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 7, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1495, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 4 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8577, "student": 73099 }, "scored_response_tokens": { "behavior": 73099, "student_rl": 41232 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "unfiltered", "difficulty_history": [ [ 1.0, 0.125, 0.0, 0.5, 0.0, 0.875, 0.0, 0.125 ], [ 0.875, 0.75, 0.0, 0.125, 0.125, 0.625, 0.0, 0.0 ], [ 0.875, 0.5, 0.125, 0.625, 0.0, 0.875, 0.125, 0.0 ], [ 0.625, 0.375, 0.0, 0.75, 0.0, 0.5, 0.0, 0.0 ] ], "difficulty_dynamics": { "span": [ 0.375, 0.625, 0.125, 0.625, 0.125, 0.375, 0.125, 0.125 ], "total_variation": [ 0.375, 1.0, 0.25, 1.0, 0.25, 0.875, 0.25, 0.125 ], "extra_round_trip_variation": [ 0.0, 0.375, 0.125, 0.375, 0.125, 0.5, 0.125, 0.0 ] }, "initial_static_easy_to_hard_order": null, "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] } } }, { "seed": 43, "device": "cuda", "diagnostic_only": true, "variants": { "fixed-medium": { "method": "grpo-difficulty-curriculum", "seed": 43, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1489, "truncated_responses": 0 }, "training": [ { "step": 1, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.75 }, { "training_row": 1, "eligible": false, "current_success": 0.25 }, { "training_row": 2, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0 ], "loss": -7.450580596923828e-09, "gradient_l2_norm": 0.19137264788150787, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 1.2, 0.2, 1.2, 1.2, 0.2 ], "advantages": [ -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, 0.9999979734420776, -0.9999979734420776 ] }, { "training_row": 3, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0 ], "loss": 0.0009858757257461548, "gradient_l2_norm": 0.16644123196601868, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 1.2, 1.2, 0.2, 1.2, 0.2 ], "advantages": [ 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776 ] }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1522, "truncated_responses": 1 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 2, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.75 }, { "training_row": 1, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0 ], "loss": 0.0, "gradient_l2_norm": 0.17128634452819824, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 1.2, 1.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.9999979734420776, -0.9999979734420776, 0.9999979734420776, 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776 ] }, { "training_row": 2, "eligible": false, "current_success": 0.25 }, { "training_row": 3, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0 ], "loss": 0.0003101825714111328, "gradient_l2_norm": 0.13456909358501434, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 1.2, 1.2, 1.2, 1.2 ], "advantages": [ -0.9999979734420776, -0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776, 0.9999979734420776, 0.9999979734420776, 0.9999979734420776 ] }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.125 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1669, "truncated_responses": 1 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 3, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 1.0 }, { "training_row": 1, "eligible": false, "current_success": 0.125 }, { "training_row": 2, "eligible": false, "current_success": 0.25 }, { "training_row": 3, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0 ], "loss": 1.4901161193847656e-08, "gradient_l2_norm": 0.14805005490779877, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 0.2, 1.2, 0.2, 0.2, 1.2 ], "advantages": [ 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, -1.2909917831420898, -1.2909917831420898, 0.7745950818061829 ] }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0 ], "loss": 0.003036320209503174, "gradient_l2_norm": 0.6054434776306152, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.0, 1.2, 1.2, 1.2, 0.2, 0.2 ], "advantages": [ 0.7697988748550415, 0.7697988748550415, -1.539597749710083, 0.7697988748550415, 0.7697988748550415, 0.7697988748550415, -1.154698371887207, -1.154698371887207 ] }, { "training_row": 6, "eligible": false, "current_success": 0.125 }, { "training_row": 7, "eligible": false, "current_success": 0.125 } ], "validation_diagnostic": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1487, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 4, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 1.0 }, { "training_row": 1, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0 ], "loss": 2.9802322387695312e-08, "gradient_l2_norm": 0.17466998100280762, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 1.2, 0.2, 1.2, 0.2, 0.2 ], "advantages": [ 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776 ] }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0 ], "loss": -0.0004735589027404785, "gradient_l2_norm": 0.1555124968290329, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2, 1.2 ], "advantages": [ 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, 1.2909917831420898 ] }, { "training_row": 4, "eligible": false, "current_success": 0.25 }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1489, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 2 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8988, "student": 75998 }, "scored_response_tokens": { "behavior": 75998, "student_rl": 17576 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "fixed-medium", "difficulty_history": [ [ 0.75, 0.25, 0.5, 0.5, 0.125, 0.75, 0.0, 0.0 ], [ 0.75, 0.5, 0.25, 0.5, 0.0, 0.75, 0.0, 0.125 ], [ 1.0, 0.125, 0.25, 0.625, 0.0, 0.625, 0.125, 0.125 ], [ 1.0, 0.5, 0.125, 0.375, 0.25, 0.75, 0.0, 0.0 ] ], "difficulty_dynamics": { "span": [ 0.25, 0.375, 0.375, 0.25, 0.25, 0.125, 0.125, 0.125 ], "total_variation": [ 0.25, 1.0, 0.375, 0.375, 0.375, 0.25, 0.25, 0.25 ], "extra_round_trip_variation": [ 0.0, 0.625, 0.0, 0.125, 0.125, 0.125, 0.125, 0.125 ] }, "initial_static_easy_to_hard_order": null, "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] }, "staged": { "method": "grpo-difficulty-curriculum", "seed": 43, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1361, "truncated_responses": 0 }, "training": [ { "step": 1, "band": [ 0.625, 0.875 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0 ], "loss": 0.0, "gradient_l2_norm": 0.21029211580753326, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 1.2, 1.2, 1.2, 1.2, 0.2 ], "advantages": [ 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732 ] }, { "training_row": 1, "eligible": false, "current_success": 0.25 }, { "training_row": 2, "eligible": false, "current_success": 0.5 }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "loss": 0.0006538927555084229, "gradient_l2_norm": 0.13964855670928955, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "advantages": [ -1.7320468425750732, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1533, "truncated_responses": 0 }, "transition": "new-best", "effective_updates": 2 }, { "step": 2, "band": [ 0.625, 0.875 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0 ], "loss": -3.725290298461914e-09, "gradient_l2_norm": 0.1771099865436554, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 1, "eligible": false, "current_success": 0.375 }, { "training_row": 2, "eligible": false, "current_success": 0.0 }, { "training_row": 3, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0 ], "loss": 0.0010784491896629333, "gradient_l2_norm": 0.20091204345226288, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": false, "current_success": 0.5 }, { "training_row": 6, "eligible": false, "current_success": 0.125 }, { "training_row": 7, "eligible": false, "current_success": 0.125 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1375, "truncated_responses": 0 }, "transition": "keep", "effective_updates": 2 }, { "step": 3, "band": [ 0.625, 0.875 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 1.0 }, { "training_row": 1, "eligible": false, "current_success": 0.25 }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 1.0 ], "loss": 7.450580596923828e-09, "gradient_l2_norm": 0.19416001439094543, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 1.2, 0.2, 1.2, 0.2, 1.2 ], "advantages": [ 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829 ] }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0 ], "loss": -0.001194547861814499, "gradient_l2_norm": 0.1283445805311203, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1546, "truncated_responses": 0 }, "transition": "advance", "effective_updates": 2 }, { "step": 4, "band": [ 0.5, 0.75 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0 ], "loss": 7.450580596923828e-09, "gradient_l2_norm": 0.16924527287483215, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 1.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776 ] }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0 ], "loss": 0.0017559528350830078, "gradient_l2_norm": 0.1759641468524933, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 1.2, 0.2, 1.2, 1.2, 0.2 ], "advantages": [ 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898 ] }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0 ], "loss": 0.0008786693215370178, "gradient_l2_norm": 0.16480566561222076, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 1.2, 1.2, 1.2, 0.2, 1.2 ], "advantages": [ -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244 ] }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1361, "truncated_responses": 0 }, "transition": "new-best", "effective_updates": 3 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8508, "student": 75379 }, "scored_response_tokens": { "behavior": 75379, "student_rl": 18743 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "staged", "difficulty_history": [ [ 0.75, 0.25, 0.5, 0.5, 0.125, 0.75, 0.0, 0.0 ], [ 0.875, 0.375, 0.0, 0.875, 0.125, 0.5, 0.125, 0.125 ], [ 1.0, 0.25, 0.125, 0.625, 0.125, 0.875, 0.0, 0.0 ], [ 0.875, 0.5, 0.125, 0.625, 0.125, 0.75, 0.0, 0.0 ] ], "difficulty_dynamics": { "span": [ 0.25, 0.25, 0.5, 0.375, 0.0, 0.375, 0.125, 0.125 ], "total_variation": [ 0.375, 0.5, 0.625, 0.625, 0.0, 0.75, 0.25, 0.25 ], "extra_round_trip_variation": [ 0.125, 0.25, 0.125, 0.25, 0.0, 0.375, 0.125, 0.125 ] }, "initial_static_easy_to_hard_order": null, "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] }, "static": { "method": "grpo-difficulty-curriculum", "seed": 43, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1496, "truncated_responses": 0 }, "training": [ { "step": 1, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0 ], "loss": 0.0, "gradient_l2_norm": 0.21029211580753326, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 1.2, 1.2, 1.2, 1.2, 0.2 ], "advantages": [ 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732 ] }, { "training_row": 1, "eligible": false, "current_success": 0.25 }, { "training_row": 2, "eligible": false, "current_success": 0.5 }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "loss": 0.0006538927555084229, "gradient_l2_norm": 0.13964855670928955, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "advantages": [ -1.7320468425750732, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1533, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 2, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": false, "current_success": 0.375 }, { "training_row": 2, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 3, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0 ], "loss": -7.450580596923828e-09, "gradient_l2_norm": 0.1941450536251068, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": false, "current_success": 0.5 }, { "training_row": 6, "eligible": false, "current_success": 0.125 }, { "training_row": 7, "eligible": false, "current_success": 0.125 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1512, "truncated_responses": 1 }, "transition": "not-staged", "effective_updates": 1 }, { "step": 3, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 1.0 }, { "training_row": 1, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0 ], "loss": -1.4901161193847656e-08, "gradient_l2_norm": 0.16990217566490173, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 1.2, 0.2, 0.2, 1.2, 0.2, 0.2 ], "advantages": [ 1.2909917831420898, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829 ] }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": false, "current_success": 0.625 }, { "training_row": 4, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 5, "eligible": false, "current_success": 1.0 }, { "training_row": 6, "eligible": false, "current_success": 0.125 }, { "training_row": 7, "eligible": false, "current_success": 0.125 } ], "validation_diagnostic": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1538, "truncated_responses": 1 }, "transition": "not-staged", "effective_updates": 1 }, { "step": 4, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": false, "current_success": 0.625 }, { "training_row": 2, "eligible": false, "current_success": 0.375 }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0 ], "loss": -7.450580596923828e-09, "gradient_l2_norm": 0.17390115559101105, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523 ] }, { "training_row": 7, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0 ], "loss": 0.0004882924258708954, "gradient_l2_norm": 0.13167545199394226, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 1.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1496, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 2 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8907, "student": 75594 }, "scored_response_tokens": { "behavior": 75594, "student_rl": 14595 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "static", "difficulty_history": [ [ 0.75, 0.25, 0.5, 0.5, 0.125, 0.75, 0.0, 0.0 ], [ 0.875, 0.375, 0.0, 0.875, 0.125, 0.5, 0.125, 0.125 ], [ 1.0, 0.375, 0.125, 0.625, 0.0, 1.0, 0.125, 0.125 ], [ 0.875, 0.625, 0.375, 0.5, 0.0, 0.75, 0.125, 0.125 ] ], "difficulty_dynamics": { "span": [ 0.25, 0.375, 0.5, 0.375, 0.125, 0.5, 0.125, 0.125 ], "total_variation": [ 0.375, 0.375, 0.875, 0.75, 0.125, 1.0, 0.125, 0.125 ], "extra_round_trip_variation": [ 0.125, 0.0, 0.375, 0.375, 0.0, 0.5, 0.0, 0.0 ] }, "initial_static_easy_to_hard_order": [ 0, 5, 2, 3, 1, 4, 6, 7 ], "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] }, "too-hard": { "method": "grpo-difficulty-curriculum", "seed": 43, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1524, "truncated_responses": 1 }, "training": [ { "step": 1, "band": [ 0.125, 0.375 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.75 }, { "training_row": 1, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0 ], "loss": -1.4901161193847656e-08, "gradient_l2_norm": 0.16526220738887787, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 1.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244 ] }, { "training_row": 2, "eligible": false, "current_success": 0.5 }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0 ], "loss": -0.00039048120379447937, "gradient_l2_norm": 0.1660120189189911, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757 ] }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1233, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 2, "band": [ 0.125, 0.375 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": false, "current_success": 0.5 }, { "training_row": 2, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0 ], "loss": 0.0, "gradient_l2_norm": 0.1514921486377716, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 1.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 3, "eligible": false, "current_success": 0.625 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.75, "examples": 4, "generated_tokens": 1460, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 1 }, { "step": 3, "band": [ 0.125, 0.375 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 1.0 }, { "training_row": 1, "eligible": false, "current_success": 0.625 }, { "training_row": 2, "eligible": false, "current_success": 0.5 }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 1.0 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.75, "examples": 4, "generated_tokens": 1460, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 0 }, { "step": 4, "band": [ 0.125, 0.375 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.625 }, { "training_row": 1, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0 ], "loss": 0.0, "gradient_l2_norm": 0.17602385580539703, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 1.2, 0.2, 1.2, 1.2 ], "advantages": [ -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, 1.2909917831420898, 1.2909917831420898 ] }, { "training_row": 2, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": 0.0027965866029262543, "gradient_l2_norm": 0.2077626883983612, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.0, 1.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.2886742949485779, -0.8660228848457336, 2.5980687141418457, -0.2886742949485779, -0.2886742949485779, -0.2886742949485779, -0.2886742949485779, -0.2886742949485779 ] }, { "training_row": 3, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0 ], "loss": 0.0005997270345687866, "gradient_l2_norm": 0.17210617661476135, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732 ] }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": -0.0006318092346191406, "gradient_l2_norm": 0.17092061042785645, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1524, "truncated_responses": 1 }, "transition": "not-staged", "effective_updates": 4 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8533, "student": 73796 }, "scored_response_tokens": { "behavior": 73796, "student_rl": 16620 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "too-hard", "difficulty_history": [ [ 0.75, 0.25, 0.5, 0.5, 0.125, 0.75, 0.0, 0.0 ], [ 0.875, 0.5, 0.125, 0.625, 0.0, 0.75, 0.0, 0.0 ], [ 1.0, 0.625, 0.5, 0.5, 0.0, 1.0, 0.0, 0.0 ], [ 0.625, 0.375, 0.125, 0.25, 0.0, 0.75, 0.125, 0.0 ] ], "difficulty_dynamics": { "span": [ 0.375, 0.375, 0.375, 0.375, 0.125, 0.25, 0.125, 0.0 ], "total_variation": [ 0.625, 0.625, 1.125, 0.5, 0.125, 0.5, 0.125, 0.0 ], "extra_round_trip_variation": [ 0.25, 0.25, 0.75, 0.125, 0.0, 0.25, 0.0, 0.0 ] }, "initial_static_easy_to_hard_order": null, "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] }, "unfiltered": { "method": "grpo-difficulty-curriculum", "seed": 43, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1353, "truncated_responses": 0 }, "training": [ { "step": 1, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0 ], "loss": 0.0, "gradient_l2_norm": 0.21029211580753326, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 1.2, 1.2, 1.2, 1.2, 0.2 ], "advantages": [ 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732 ] }, { "training_row": 1, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0 ], "loss": 0.0016159191727638245, "gradient_l2_norm": 0.16332122683525085, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 1.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244 ] }, { "training_row": 2, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 0.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0 ], "loss": 0.0014016702771186829, "gradient_l2_norm": 0.19165532290935516, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 1.2, 0.2, 1.2, 1.2, 0.2 ], "advantages": [ -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, 0.9999979734420776, -0.9999979734420776 ] }, { "training_row": 3, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 0.0 ], "loss": 0.0017340034246444702, "gradient_l2_norm": 0.18799102306365967, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 1.2, 1.2, 0.2, 1.2, 0.2 ], "advantages": [ 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776 ] }, { "training_row": 4, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0 ], "loss": -0.0004289001226425171, "gradient_l2_norm": 0.16215303540229797, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757 ] }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "loss": 0.0003112107515335083, "gradient_l2_norm": 0.14815963804721832, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "advantages": [ -1.7320468425750732, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 6, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 7, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1356, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 6 }, { "step": 2, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0 ], "loss": 3.725290298461914e-09, "gradient_l2_norm": 0.14475379884243011, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 1, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0 ], "loss": -0.0010508820414543152, "gradient_l2_norm": 0.18245378136634827, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 0.2, 1.2, 0.2, 0.2, 0.2 ], "advantages": [ 1.2909917831420898, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829 ] }, { "training_row": 2, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0 ], "loss": 0.00011548399925231934, "gradient_l2_norm": 0.14699749648571014, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244 ] }, { "training_row": 3, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 0.0 ], "loss": 0.0007837265729904175, "gradient_l2_norm": 0.15821406245231628, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 1.2, 1.2, 1.2, 0.2, 0.2 ], "advantages": [ 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776, 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776 ] }, { "training_row": 4, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 0.0 ], "loss": -0.00015047192573547363, "gradient_l2_norm": 0.1654481440782547, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2, 0.2 ], "advantages": [ 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732 ] }, { "training_row": 6, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 7, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1424, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 5 }, { "step": 3, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 1.0, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ] }, { "training_row": 1, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": 0.0, "gradient_l2_norm": 0.1733408421278, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.5773489475250244, 1.7320468425750732, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244 ] }, { "training_row": 2, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0 ], "loss": 0.0007157251238822937, "gradient_l2_norm": 0.5526610612869263, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 0.2, 0.2, 1.2, 0.2, 0.2 ], "advantages": [ -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244 ] }, { "training_row": 3, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0 ], "loss": -0.0003365352749824524, "gradient_l2_norm": 0.15619783103466034, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 0.2, 1.2, 0.2, 0.2, 0.2 ], "advantages": [ 0.9999979734420776, 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, -0.9999979734420776 ] }, { "training_row": 4, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0 ], "loss": -0.0005653202533721924, "gradient_l2_norm": 0.15326756238937378, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 0.2, 1.2, 0.2, 1.2, 1.2 ], "advantages": [ 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 6, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0 ], "loss": 0.00017213821411132812, "gradient_l2_norm": 0.15466849505901337, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 0.2, 1.2, 0.2, 0.2, 0.2 ], "advantages": [ 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244 ] }, { "training_row": 7, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] } ], "validation_diagnostic": { "accuracy": 0.75, "examples": 4, "generated_tokens": 1600, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 5 }, { "step": 4, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 1.0 ], "loss": 7.450580596923828e-09, "gradient_l2_norm": 0.19009071588516235, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 0.2, 1.2, 0.2, 1.2, 1.2 ], "advantages": [ 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 1, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0 ], "loss": -0.00018861517310142517, "gradient_l2_norm": 0.13179272413253784, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523 ] }, { "training_row": 2, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": 0.0004029311239719391, "gradient_l2_norm": 0.14190854132175446, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 3, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0 ], "loss": -0.0009687170386314392, "gradient_l2_norm": 0.19252340495586395, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 0.2, 1.2, 0.2, 1.2, 1.2 ], "advantages": [ 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, -1.2909917831420898, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829 ] }, { "training_row": 4, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "loss": 0.0013948827981948853, "gradient_l2_norm": 0.1810312420129776, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "advantages": [ 0.5773489475250244, -1.7320468425750732, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 6, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0 ], "loss": -0.001209937036037445, "gradient_l2_norm": 0.1682882457971573, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732 ] }, { "training_row": 7, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1353, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 6 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8418, "student": 76241 }, "scored_response_tokens": { "behavior": 76241, "student_rl": 50002 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "unfiltered", "difficulty_history": [ [ 0.75, 0.25, 0.5, 0.5, 0.125, 0.75, 0.0, 0.0 ], [ 0.875, 0.375, 0.25, 0.5, 0.0, 0.75, 0.0, 0.0 ], [ 1.0, 0.25, 0.25, 0.5, 0.0, 0.75, 0.25, 0.0 ], [ 0.75, 0.125, 0.125, 0.625, 0.0, 0.75, 0.25, 0.0 ] ], "difficulty_dynamics": { "span": [ 0.25, 0.25, 0.375, 0.125, 0.125, 0.0, 0.25, 0.0 ], "total_variation": [ 0.5, 0.375, 0.375, 0.125, 0.125, 0.0, 0.25, 0.0 ], "extra_round_trip_variation": [ 0.25, 0.125, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ] }, "initial_static_easy_to_hard_order": null, "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] } } }, { "seed": 44, "device": "cuda", "diagnostic_only": true, "variants": { "fixed-medium": { "method": "grpo-difficulty-curriculum", "seed": 44, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1369, "truncated_responses": 0 }, "training": [ { "step": 1, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0 ], "loss": 1.4901161193847656e-08, "gradient_l2_norm": 0.13831821084022522, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 0.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.7745950818061829, 1.2909917831420898, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898 ] }, { "training_row": 2, "eligible": false, "current_success": 0.25 }, { "training_row": 3, "eligible": false, "current_success": 0.75 }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1456, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 1 }, { "step": 2, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0 ], "loss": 4.470348358154297e-08, "gradient_l2_norm": 0.1652008593082428, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 1.2, 0.2, 1.2, 0.2, 0.2 ], "advantages": [ 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829 ] }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": false, "current_success": 0.125 }, { "training_row": 4, "eligible": false, "current_success": 0.25 }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": false, "current_success": 0.25 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1269, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 1 }, { "step": 3, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": false, "current_success": 0.25 }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0, 1.0 ], "loss": -1.4901161193847656e-08, "gradient_l2_norm": 0.15434415638446808, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 0.2, 1.2, 0.2, 1.2, 1.2 ], "advantages": [ 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, 0.9999979734420776 ] }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 1.0 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1417, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 1 }, { "step": 4, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": false, "current_success": 0.75 }, { "training_row": 2, "eligible": false, "current_success": 0.0 }, { "training_row": 3, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 0.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0 ], "loss": -1.4901161193847656e-08, "gradient_l2_norm": 0.2216559499502182, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ -1.2909917831420898, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829 ] }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": false, "current_success": 1.0 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.25 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1369, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 1 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8212, "student": 73434 }, "scored_response_tokens": { "behavior": 73434, "student_rl": 8774 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "fixed-medium", "difficulty_history": [ [ 0.875, 0.375, 0.25, 0.75, 0.125, 0.75, 0.0, 0.0 ], [ 0.875, 0.375, 0.125, 0.125, 0.25, 0.75, 0.25, 0.0 ], [ 0.875, 0.25, 0.125, 0.5, 0.0, 1.0, 0.0, 0.0 ], [ 0.875, 0.75, 0.0, 0.625, 0.125, 1.0, 0.0, 0.25 ] ], "difficulty_dynamics": { "span": [ 0.0, 0.5, 0.25, 0.625, 0.25, 0.25, 0.25, 0.25 ], "total_variation": [ 0.0, 0.625, 0.25, 1.125, 0.5, 0.25, 0.5, 0.25 ], "extra_round_trip_variation": [ 0.0, 0.125, 0.0, 0.5, 0.25, 0.0, 0.25, 0.0 ] }, "initial_static_easy_to_hard_order": null, "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] }, "staged": { "method": "grpo-difficulty-curriculum", "seed": 44, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1383, "truncated_responses": 0 }, "training": [ { "step": 1, "band": [ 0.625, 0.875 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0 ], "loss": 0.0, "gradient_l2_norm": 0.1826382428407669, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 1, "eligible": false, "current_success": 0.375 }, { "training_row": 2, "eligible": false, "current_success": 0.25 }, { "training_row": 3, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0 ], "loss": 0.0016562342643737793, "gradient_l2_norm": 0.1566203087568283, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0 ], "loss": -0.0011830627918243408, "gradient_l2_norm": 0.13673582673072815, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 0.2, 0.2, 1.2 ], "advantages": [ 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, -1.7320468425750732, 0.5773489475250244 ] }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1383, "truncated_responses": 0 }, "transition": "new-best", "effective_updates": 3 }, { "step": 2, "band": [ 0.625, 0.875 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0 ], "loss": 0.0, "gradient_l2_norm": 0.34789571166038513, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 0.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523 ] }, { "training_row": 1, "eligible": false, "current_success": 0.375 }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0 ], "loss": 0.0020799487829208374, "gradient_l2_norm": 0.18106399476528168, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 6, "eligible": false, "current_success": 0.125 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1471, "truncated_responses": 0 }, "transition": "keep", "effective_updates": 2 }, { "step": 3, "band": [ 0.625, 0.875 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "loss": 1.1175870895385742e-08, "gradient_l2_norm": 0.18197466433048248, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "advantages": [ 0.3779633343219757, -2.6457433700561523, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 1, "eligible": false, "current_success": 0.25 }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 0.0, 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0 ], "loss": -0.000806860625743866, "gradient_l2_norm": 0.13967664539813995, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 1.2, 1.2, 1.2, 1.2, 0.2 ], "advantages": [ -1.2909917831420898, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898 ] }, { "training_row": 4, "eligible": false, "current_success": 0.25 }, { "training_row": 5, "eligible": false, "current_success": 1.0 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1570, "truncated_responses": 1 }, "transition": "advance", "effective_updates": 2 }, { "step": 4, "band": [ 0.5, 0.75 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 1.0 }, { "training_row": 1, "eligible": false, "current_success": 0.375 }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": false, "current_success": 0.375 }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": false, "current_success": 1.0 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1383, "truncated_responses": 0 }, "transition": "no-effective-updates", "effective_updates": 0 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8522, "student": 74800 }, "scored_response_tokens": { "behavior": 74800, "student_rl": 14252 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "staged", "difficulty_history": [ [ 0.875, 0.375, 0.25, 0.75, 0.125, 0.75, 0.0, 0.0 ], [ 0.875, 0.375, 0.125, 0.5, 0.125, 0.75, 0.125, 0.0 ], [ 0.875, 0.25, 0.125, 0.625, 0.25, 1.0, 0.0, 0.0 ], [ 1.0, 0.375, 0.125, 0.375, 0.125, 1.0, 0.0, 0.0 ] ], "difficulty_dynamics": { "span": [ 0.125, 0.125, 0.125, 0.375, 0.125, 0.25, 0.125, 0.0 ], "total_variation": [ 0.125, 0.25, 0.125, 0.625, 0.25, 0.25, 0.25, 0.0 ], "extra_round_trip_variation": [ 0.0, 0.125, 0.0, 0.25, 0.125, 0.0, 0.125, 0.0 ] }, "initial_static_easy_to_hard_order": null, "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] }, "static": { "method": "grpo-difficulty-curriculum", "seed": 44, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1310, "truncated_responses": 0 }, "training": [ { "step": 1, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0 ], "loss": 0.0, "gradient_l2_norm": 0.1826382428407669, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 1, "eligible": false, "current_success": 0.375 }, { "training_row": 2, "eligible": false, "current_success": 0.25 }, { "training_row": 3, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0 ], "loss": 0.0016562342643737793, "gradient_l2_norm": 0.1566203087568283, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.75, "examples": 4, "generated_tokens": 1471, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 2, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 1.0 }, { "training_row": 1, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0 ], "loss": -7.450580596923828e-09, "gradient_l2_norm": 0.15844206511974335, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 0.2, 0.2, 1.2, 0.2, 0.2 ], "advantages": [ -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244 ] }, { "training_row": 2, "eligible": false, "current_success": 0.125 }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 0.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0 ], "loss": 0.0021607503294944763, "gradient_l2_norm": 0.17329858243465424, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 1.2, 1.2, 1.2, 1.2, 0.2, 1.2 ], "advantages": [ 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244 ] }, { "training_row": 6, "eligible": false, "current_success": 0.125 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1633, "truncated_responses": 1 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 3, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": false, "current_success": 0.75 }, { "training_row": 2, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0 ], "loss": -2.60770320892334e-08, "gradient_l2_norm": 0.17747393250465393, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757 ] }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 0.0 ], "loss": 0.0005359947681427002, "gradient_l2_norm": 0.14197126030921936, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 0.2, 1.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244 ] }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.75, "examples": 4, "generated_tokens": 1563, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 4, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": false, "current_success": 0.5 }, { "training_row": 2, "eligible": false, "current_success": 0.0 }, { "training_row": 3, "eligible": false, "current_success": 0.25 }, { "training_row": 4, "eligible": false, "current_success": 0.125 }, { "training_row": 5, "eligible": false, "current_success": 1.0 }, { "training_row": 6, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 7, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0 ], "loss": -2.9802322387695312e-08, "gradient_l2_norm": 0.16588783264160156, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523 ] } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1310, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 1 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8619, "student": 72820 }, "scored_response_tokens": { "behavior": 72820, "student_rl": 15398 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "static", "difficulty_history": [ [ 0.875, 0.375, 0.25, 0.75, 0.125, 0.75, 0.0, 0.0 ], [ 1.0, 0.25, 0.125, 0.5, 0.0, 0.75, 0.125, 0.0 ], [ 0.875, 0.75, 0.125, 0.5, 0.25, 0.75, 0.0, 0.0 ], [ 0.875, 0.5, 0.0, 0.25, 0.125, 1.0, 0.0, 0.125 ] ], "difficulty_dynamics": { "span": [ 0.125, 0.5, 0.25, 0.5, 0.25, 0.25, 0.125, 0.125 ], "total_variation": [ 0.25, 0.875, 0.25, 0.5, 0.5, 0.25, 0.25, 0.125 ], "extra_round_trip_variation": [ 0.125, 0.375, 0.0, 0.0, 0.25, 0.0, 0.125, 0.0 ] }, "initial_static_easy_to_hard_order": [ 0, 3, 5, 1, 2, 4, 6, 7 ], "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] }, "too-hard": { "method": "grpo-difficulty-curriculum", "seed": 44, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1550, "truncated_responses": 1 }, "training": [ { "step": 1, "band": [ 0.125, 0.375 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0 ], "loss": 1.4901161193847656e-08, "gradient_l2_norm": 0.13831821084022522, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 0.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.7745950818061829, 1.2909917831420898, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898 ] }, { "training_row": 2, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0 ], "loss": 0.0009204447269439697, "gradient_l2_norm": 0.16590556502342224, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244 ] }, { "training_row": 3, "eligible": false, "current_success": 0.75 }, { "training_row": 4, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0 ], "loss": 0.000863734632730484, "gradient_l2_norm": 0.12472493201494217, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523 ] }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1766, "truncated_responses": 1 }, "transition": "not-staged", "effective_updates": 3 }, { "step": 2, "band": [ 0.125, 0.375 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": false, "current_success": 0.625 }, { "training_row": 2, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0 ], "loss": 0.0, "gradient_l2_norm": 0.19227898120880127, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.5773489475250244, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244 ] }, { "training_row": 3, "eligible": false, "current_success": 0.5 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 1.0 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1524, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 1 }, { "step": 3, "band": [ 0.125, 0.375 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.75 }, { "training_row": 1, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0 ], "loss": -1.4901161193847656e-08, "gradient_l2_norm": 0.16933807730674744, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 1.2, 1.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898 ] }, { "training_row": 2, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0 ], "loss": 0.0015721023082733154, "gradient_l2_norm": 0.15421536564826965, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 1.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244 ] }, { "training_row": 3, "eligible": false, "current_success": 0.625 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 1.0 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1378, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 2 }, { "step": 4, "band": [ 0.125, 0.375 ], "groups": [ { "training_row": 0, "eligible": false, "current_success": 0.875 }, { "training_row": 1, "eligible": false, "current_success": 0.625 }, { "training_row": 2, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 1.0, 0.0 ], "loss": 0.0, "gradient_l2_norm": 0.15359120070934296, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 0.2, 1.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829 ] }, { "training_row": 3, "eligible": false, "current_success": 0.625 }, { "training_row": 4, "eligible": false, "current_success": 0.0 }, { "training_row": 5, "eligible": false, "current_success": 0.75 }, { "training_row": 6, "eligible": false, "current_success": 0.0 }, { "training_row": 7, "eligible": false, "current_success": 0.0 } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1550, "truncated_responses": 1 }, "transition": "not-staged", "effective_updates": 1 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 9100, "student": 72857 }, "scored_response_tokens": { "behavior": 72857, "student_rl": 17190 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "too-hard", "difficulty_history": [ [ 0.875, 0.375, 0.25, 0.75, 0.125, 0.75, 0.0, 0.0 ], [ 0.875, 0.625, 0.25, 0.5, 0.0, 1.0, 0.0, 0.0 ], [ 0.75, 0.375, 0.25, 0.625, 0.0, 1.0, 0.0, 0.0 ], [ 0.875, 0.625, 0.375, 0.625, 0.0, 0.75, 0.0, 0.0 ] ], "difficulty_dynamics": { "span": [ 0.125, 0.25, 0.125, 0.25, 0.125, 0.25, 0.0, 0.0 ], "total_variation": [ 0.25, 0.75, 0.125, 0.375, 0.125, 0.5, 0.0, 0.0 ], "extra_round_trip_variation": [ 0.125, 0.5, 0.0, 0.125, 0.0, 0.25, 0.0, 0.0 ] }, "initial_static_easy_to_hard_order": null, "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] }, "unfiltered": { "method": "grpo-difficulty-curriculum", "seed": 44, "device": "cuda", "diagnostic_only": true, "promotion_eligible": false, "evaluation_tier": "l1_mechanism", "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "baseline": { "accuracy": 0.25, "examples": 4, "generated_tokens": 1332, "truncated_responses": 0 }, "method_metrics": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1308, "truncated_responses": 0 }, "training": [ { "step": 1, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0 ], "loss": 0.0, "gradient_l2_norm": 0.1826382428407669, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757 ] }, { "training_row": 1, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0 ], "loss": 0.001038283109664917, "gradient_l2_norm": 0.1318325251340866, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 0.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.7745950818061829, 1.2909917831420898, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898 ] }, { "training_row": 2, "eligible": true, "current_success": 0.25, "correctness_rewards": [ 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0 ], "loss": 0.0006513670086860657, "gradient_l2_norm": 0.15832626819610596, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ 1.7320468425750732, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, -0.5773489475250244, 1.7320468425750732, -0.5773489475250244 ] }, { "training_row": 3, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0 ], "loss": 0.0011982470750808716, "gradient_l2_norm": 0.15507328510284424, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 4, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0 ], "loss": 0.0003406256437301636, "gradient_l2_norm": 0.12433010339736938, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523 ] }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0, 1.0 ], "loss": -0.0005710572004318237, "gradient_l2_norm": 0.1346355676651001, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 0.2, 0.2, 1.2 ], "advantages": [ 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, -1.7320468425750732, 0.5773489475250244 ] }, { "training_row": 6, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 7, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] } ], "validation_diagnostic": { "accuracy": 0.75, "examples": 4, "generated_tokens": 1491, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 6 }, { "step": 2, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.875, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0 ], "loss": 3.725290298461914e-09, "gradient_l2_norm": 0.33505764603614807, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 0.2 ], "advantages": [ 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, 0.3779633343219757, -2.6457433700561523 ] }, { "training_row": 1, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0 ], "loss": 0.0004852190613746643, "gradient_l2_norm": 0.16745416820049286, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 0.2, 0.2, 0.2, 0.2, 1.2, 1.2, 1.2 ], "advantages": [ 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776, 0.9999979734420776, 0.9999979734420776 ] }, { "training_row": 2, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 0.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0 ], "loss": -0.00025279074907302856, "gradient_l2_norm": 0.1713051199913025, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 1.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, -0.7745950818061829 ] }, { "training_row": 3, "eligible": true, "current_success": 0.375, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 1.0 ], "loss": 0.000545307993888855, "gradient_l2_norm": 0.14962179958820343, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 1.2, 1.2, 0.2, 1.2 ], "advantages": [ -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, -0.7745950818061829, 1.2909917831420898, 1.2909917831420898, -0.7745950818061829, 1.2909917831420898 ] }, { "training_row": 4, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": -0.0001786835491657257, "gradient_l2_norm": 0.12846629321575165, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 0.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 1.0 ], "loss": 0.0003411918878555298, "gradient_l2_norm": 0.1480056792497635, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 1.2, 1.2, 1.2, 0.2, 1.2 ], "advantages": [ -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244 ] }, { "training_row": 6, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 7, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1464, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 6 }, { "step": 3, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 1.0, "correctness_rewards": [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ] }, { "training_row": 1, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 1.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0 ], "loss": -7.450580596923828e-09, "gradient_l2_norm": 0.15700997412204742, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 0.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ 0.9999979734420776, 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776 ] }, { "training_row": 2, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": 0.0030484534800052643, "gradient_l2_norm": 0.15841951966285706, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 3, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 0.0, 1.0 ], "loss": 0.00012655556201934814, "gradient_l2_norm": 0.1552085429430008, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 0.2, 1.2, 0.2, 0.2, 1.2 ], "advantages": [ 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, -1.2909917831420898, -1.2909917831420898, 0.7745950818061829 ] }, { "training_row": 4, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 1.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0, 1.0 ], "loss": 0.0005815383046865463, "gradient_l2_norm": 0.18718504905700684, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 1.2, 1.0, 0.2, 1.2, 1.2 ], "advantages": [ 0.6493079662322998, 0.6493079662322998, -1.7118117809295654, 0.6493079662322998, 0.17708398401737213, -1.7118117809295654, 0.6493079662322998, 0.6493079662322998 ] }, { "training_row": 6, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": 0.004795640707015991, "gradient_l2_norm": 2.288238525390625, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.0, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ 0.377958744764328, 0.377958744764328, -2.6457114219665527, 0.377958744764328, 0.377958744764328, 0.377958744764328, 0.377958744764328, 0.377958744764328 ] }, { "training_row": 7, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0 ], "loss": 0.001440785825252533, "gradient_l2_norm": 0.16666735708713531, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 1.2, 0.2 ], "advantages": [ -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, 2.6457433700561523, -0.3779633343219757 ] } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1568, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 6 }, { "step": 4, "band": [ 0.375, 0.625 ], "groups": [ { "training_row": 0, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 1.0, 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0 ], "loss": -2.9802322387695312e-08, "gradient_l2_norm": 0.19042925536632538, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 1.2, 0.2, 1.2, 0.2, 1.2, 0.2 ], "advantages": [ 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, -1.2909917831420898 ] }, { "training_row": 1, "eligible": true, "current_success": 0.5, "correctness_rewards": [ 1.0, 1.0, 0.0, 1.0, 0.0, 1.0, 0.0, 0.0 ], "loss": 0.000828765332698822, "gradient_l2_norm": 0.17850220203399658, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 1.2, 1.2, 0.2, 1.2, 0.2, 1.2, 0.2, 0.2 ], "advantages": [ 0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, 0.9999979734420776, -0.9999979734420776, -0.9999979734420776 ] }, { "training_row": 2, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 3, "eligible": true, "current_success": 0.625, "correctness_rewards": [ 0.0, 0.0, 1.0, 1.0, 1.0, 0.0, 1.0, 1.0 ], "loss": 0.0028143376111984253, "gradient_l2_norm": 0.22348636388778687, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 0.2, 1.2, 1.2, 1.2, 0.2, 1.2, 1.2 ], "advantages": [ -1.2909917831420898, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829, 0.7745950818061829, -1.2909917831420898, 0.7745950818061829, 0.7745950818061829 ] }, { "training_row": 4, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": -0.0008893869817256927, "gradient_l2_norm": 0.16440168023109436, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] }, { "training_row": 5, "eligible": true, "current_success": 0.75, "correctness_rewards": [ 0.0, 1.0, 1.0, 0.0, 1.0, 1.0, 1.0, 1.0 ], "loss": 0.001261599361896515, "gradient_l2_norm": 0.17630229890346527, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 1.2, 0.2, 1.2, 1.2, 1.2, 1.2 ], "advantages": [ -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, -1.7320468425750732, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244, 0.5773489475250244 ] }, { "training_row": 6, "eligible": true, "current_success": 0.0, "correctness_rewards": [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "optimizer_update_executed": false, "zero_advantage_group": true, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ] }, { "training_row": 7, "eligible": true, "current_success": 0.125, "correctness_rewards": [ 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 ], "loss": 0.0008482374250888824, "gradient_l2_norm": 0.16358359158039093, "optimizer_update_executed": true, "zero_advantage_group": false, "rewards": [ 0.2, 1.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "advantages": [ -0.3779633343219757, 2.6457433700561523, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757, -0.3779633343219757 ] } ], "validation_diagnostic": { "accuracy": 0.5, "examples": 4, "generated_tokens": 1308, "truncated_responses": 0 }, "transition": "not-staged", "effective_updates": 6 } ], "cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8471, "student": 75016 }, "scored_response_tokens": { "behavior": 75016, "student_rl": 55419 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "protocol": { "test_opened": false, "gold_in_student_generation": false, "gold_in_student_generation_scope": "evaluated_student_inference_only", "training_construction_roles_with_gold": [], "comparison_kind": "before_after_training", "train_examples": 8, "validation_examples": 4, "max_tokens": 512 }, "parameters_changed": true, "variant": "unfiltered", "difficulty_history": [ [ 0.875, 0.375, 0.25, 0.75, 0.125, 0.75, 0.0, 0.0 ], [ 0.875, 0.5, 0.375, 0.375, 0.125, 0.75, 0.0, 0.0 ], [ 1.0, 0.5, 0.125, 0.625, 0.0, 0.75, 0.0, 0.125 ], [ 0.625, 0.5, 0.0, 0.625, 0.125, 0.75, 0.0, 0.125 ] ], "difficulty_dynamics": { "span": [ 0.375, 0.125, 0.375, 0.375, 0.125, 0.0, 0.0, 0.125 ], "total_variation": [ 0.5, 0.125, 0.5, 0.625, 0.25, 0.0, 0.0, 0.125 ], "extra_round_trip_variation": [ 0.125, 0.0, 0.125, 0.25, 0.125, 0.0, 0.0, 0.0 ] }, "initial_static_easy_to_hard_order": null, "execution_status": "passed", "limitations": [ "Small public train-only GSM8K holdout diagnostic; no official test is read.", "Qwen2.5-0.5B replaces 1.5B and diagnostic cadence is scaled to each bounded step.", "Run fixed-medium/staged/too-hard/unfiltered/static separately with matched configured budgets.", "All rejected groups and calibration generations are charged; no invented nonzero advantage.", "Staged diagnostic uses the validation split, not original 128-per-band training-heldout pools." ] } } } ], "aggregate_metrics": { "fixed-medium.initial.accuracy": { "mean": 0.25, "std": 0.0, "n": 3 }, "fixed-medium.validation.accuracy": { "mean": 0.5, "std": 0.0, "n": 3 }, "staged.initial.accuracy": { "mean": 0.25, "std": 0.0, "n": 3 }, "staged.validation.accuracy": { "mean": 0.3333333333333333, "std": 0.28867513459481287, "n": 3 }, "static.initial.accuracy": { "mean": 0.25, "std": 0.0, "n": 3 }, "static.validation.accuracy": { "mean": 0.4166666666666667, "std": 0.14433756729740643, "n": 3 }, "too-hard.initial.accuracy": { "mean": 0.25, "std": 0.0, "n": 3 }, "too-hard.validation.accuracy": { "mean": 0.5, "std": 0.0, "n": 3 }, "unfiltered.initial.accuracy": { "mean": 0.25, "std": 0.0, "n": 3 }, "unfiltered.validation.accuracy": { "mean": 0.5, "std": 0.0, "n": 3 } }, "dataset": { "verified": true, "test_read": false, "source_id": "openai/grade-school-math", "source_revision": "3101c7d5072418e28b9008a6636bde82a006892c", "source_sha256": "17f347dc51477c50d4efb83959dbb7c56297aba886e5544ee2aaed3024813465", "partition_sha256": { "train": "6b8672439c8cf6f6b6e65bf5f0e515fbf4c497c9066eefb62a6a6a61ced158e3", "validation": "2cf1119ad7b2e31e825274487ef32c073e03a8ed44f08ecd3e0366c1f40ebdb7" }, "partition": "normalized-question sha256 unsigned-prefix / 2^64 < 0.1 validation" }, "checkpoint": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "verified_mirror": { "public_id": "Qwen/Qwen2.5-0.5B-Instruct", "revision": "7ae557604adf67be50417f59c2c2f167def9a775", "file_sha256": { "config.json": "18e18afcaccafade98daf13a54092927904649e1dd4eba8299ab717d5d94ff45", "generation_config.json": "e558847a8b4402616f1273797b015104dc266fe4b520056fca88823ba8f8ebe6", "merges.txt": "599bab54075088774b1733fde865d5bd747cbcc7a547c5bc12610e874e26f5e3", "model.safetensors": "fdf756fa7fcbe7404d5c60e26bff1a0c8b8aa1f72ced49e7dd0210fe288fb7fe", "tokenizer.json": "c0382117ea329cdf097041132f6d735924b697924d6f6fc3945713e96ce87539", "tokenizer_config.json": "5b5d4f65d0acd3b2d56a35b56d374a36cbc1c8fa5cf3b3febbbfabf22f359583", "vocab.json": "ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910" } } }, "execution_checks": { "42": { "fixed-medium": { "actual_optimizer_updates": 7, "eligible_groups": 7, "mathematically_equal_reward_groups_skipped": 0, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [], "all_groups_audited": 32, "invocation_elapsed_seconds": 636.1500999443233, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 42, "variant": "fixed-medium", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 633.8375765234232, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 2901858304, "actual_max_memory_reserved_bytes": 5387583488, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8250, "student": 70009 }, "scored_response_tokens": { "behavior": 70009, "student_rl": 13880 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false }, "staged": { "actual_optimizer_updates": 10, "eligible_groups": 10, "mathematically_equal_reward_groups_skipped": 0, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [], "all_groups_audited": 32, "invocation_elapsed_seconds": 673.6660227067769, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 42, "variant": "staged", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 671.18108067289, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3108669440, "actual_max_memory_reserved_bytes": 5154799616, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8754, "student": 74288 }, "scored_response_tokens": { "behavior": 74288, "student_rl": 19984 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false }, "static": { "actual_optimizer_updates": 4, "eligible_groups": 8, "mathematically_equal_reward_groups_skipped": 4, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [ { "step": 1, "training_row": 0, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 3, "training_row": 7, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 4, "training_row": 4, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 4, "training_row": 6, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false } ], "all_groups_audited": 32, "invocation_elapsed_seconds": 685.4193786792457, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 42, "variant": "static", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 683.0766994766891, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3514061824, "actual_max_memory_reserved_bytes": 5807013888, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 9164, "student": 74639 }, "scored_response_tokens": { "behavior": 74639, "student_rl": 9616 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false }, "too-hard": { "actual_optimizer_updates": 10, "eligible_groups": 10, "mathematically_equal_reward_groups_skipped": 0, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [], "all_groups_audited": 32, "invocation_elapsed_seconds": 658.0403874441981, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 42, "variant": "too-hard", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 655.53652812168, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3842614784, "actual_max_memory_reserved_bytes": 6148849664, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8791, "student": 74204 }, "scored_response_tokens": { "behavior": 74204, "student_rl": 23930 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false }, "unfiltered": { "actual_optimizer_updates": 19, "eligible_groups": 32, "mathematically_equal_reward_groups_skipped": 13, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [ { "step": 1, "training_row": 0, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 1, "training_row": 2, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 1, "training_row": 4, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 1, "training_row": 6, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 2, "training_row": 2, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 2, "training_row": 6, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 2, "training_row": 7, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 3, "training_row": 4, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 3, "training_row": 7, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 4, "training_row": 2, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 4, "training_row": 4, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 4, "training_row": 6, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 4, "training_row": 7, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false } ], "all_groups_audited": 32, "invocation_elapsed_seconds": 606.7968027666211, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 42, "variant": "unfiltered", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 604.5412887074053, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3464610816, "actual_max_memory_reserved_bytes": 5729419264, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8577, "student": 73099 }, "scored_response_tokens": { "behavior": 73099, "student_rl": 41232 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false } }, "43": { "fixed-medium": { "actual_optimizer_updates": 8, "eligible_groups": 8, "mathematically_equal_reward_groups_skipped": 0, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [], "all_groups_audited": 32, "invocation_elapsed_seconds": 677.5788688659668, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 43, "variant": "fixed-medium", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 675.4739038906991, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3551163904, "actual_max_memory_reserved_bytes": 5112856576, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8988, "student": 75998 }, "scored_response_tokens": { "behavior": 75998, "student_rl": 17576 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false }, "staged": { "actual_optimizer_updates": 9, "eligible_groups": 9, "mathematically_equal_reward_groups_skipped": 0, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [], "all_groups_audited": 32, "invocation_elapsed_seconds": 692.4588116928935, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 43, "variant": "staged", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 690.2053823918104, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3142192640, "actual_max_memory_reserved_bytes": 4758437888, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8508, "student": 75379 }, "scored_response_tokens": { "behavior": 75379, "student_rl": 18743 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false }, "static": { "actual_optimizer_updates": 6, "eligible_groups": 8, "mathematically_equal_reward_groups_skipped": 2, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [ { "step": 2, "training_row": 2, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 3, "training_row": 4, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false } ], "all_groups_audited": 32, "invocation_elapsed_seconds": 690.217955969274, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 43, "variant": "static", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 687.8983246386051, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3544874496, "actual_max_memory_reserved_bytes": 5693767680, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8907, "student": 75594 }, "scored_response_tokens": { "behavior": 75594, "student_rl": 14595 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false }, "too-hard": { "actual_optimizer_updates": 7, "eligible_groups": 7, "mathematically_equal_reward_groups_skipped": 0, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [], "all_groups_audited": 32, "invocation_elapsed_seconds": 664.7406627573073, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 43, "variant": "too-hard", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 662.2785841114819, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3273183744, "actual_max_memory_reserved_bytes": 5259657216, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8533, "student": 73796 }, "scored_response_tokens": { "behavior": 73796, "student_rl": 16620 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false }, "unfiltered": { "actual_optimizer_updates": 22, "eligible_groups": 32, "mathematically_equal_reward_groups_skipped": 10, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [ { "step": 1, "training_row": 6, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 1, "training_row": 7, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 2, "training_row": 4, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 2, "training_row": 6, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 2, "training_row": 7, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 3, "training_row": 0, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 3, "training_row": 4, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 3, "training_row": 7, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 4, "training_row": 4, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 4, "training_row": 7, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false } ], "all_groups_audited": 32, "invocation_elapsed_seconds": 677.3662807494402, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 43, "variant": "unfiltered", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 675.1541140787303, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3616701440, "actual_max_memory_reserved_bytes": 5100273664, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8418, "student": 76241 }, "scored_response_tokens": { "behavior": 76241, "student_rl": 50002 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false } }, "44": { "fixed-medium": { "actual_optimizer_updates": 4, "eligible_groups": 4, "mathematically_equal_reward_groups_skipped": 0, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [], "all_groups_audited": 32, "invocation_elapsed_seconds": 620.4139524810016, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 44, "variant": "fixed-medium", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 618.2496887072921, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3193501696, "actual_max_memory_reserved_bytes": 4833935360, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8212, "student": 73434 }, "scored_response_tokens": { "behavior": 73434, "student_rl": 8774 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false }, "staged": { "actual_optimizer_updates": 7, "eligible_groups": 7, "mathematically_equal_reward_groups_skipped": 0, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [], "all_groups_audited": 32, "invocation_elapsed_seconds": 661.6682027466595, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 44, "variant": "staged", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 659.177813436836, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3108116480, "actual_max_memory_reserved_bytes": 4536139776, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8522, "student": 74800 }, "scored_response_tokens": { "behavior": 74800, "student_rl": 14252 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false }, "static": { "actual_optimizer_updates": 7, "eligible_groups": 8, "mathematically_equal_reward_groups_skipped": 1, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [ { "step": 4, "training_row": 6, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false } ], "all_groups_audited": 32, "invocation_elapsed_seconds": 645.960937846452, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 44, "variant": "static", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 643.5252490639687, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3480293888, "actual_max_memory_reserved_bytes": 5156896768, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8619, "student": 72820 }, "scored_response_tokens": { "behavior": 72820, "student_rl": 15398 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false }, "too-hard": { "actual_optimizer_updates": 7, "eligible_groups": 7, "mathematically_equal_reward_groups_skipped": 0, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [], "all_groups_audited": 32, "invocation_elapsed_seconds": 664.2123421393335, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 44, "variant": "too-hard", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 661.9004598706961, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3770255360, "actual_max_memory_reserved_bytes": 5511315456, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 9100, "student": 72857 }, "scored_response_tokens": { "behavior": 72857, "student_rl": 17190 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false }, "unfiltered": { "actual_optimizer_updates": 24, "eligible_groups": 32, "mathematically_equal_reward_groups_skipped": 8, "equal_reward_optimizer_updates": 0, "actual_zero_group_ledgers": [ { "step": 1, "training_row": 6, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 1, "training_row": 7, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 2, "training_row": 6, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 2, "training_row": 7, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 3, "training_row": 0, "rewards": [ 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2, 1.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 3, "training_row": 4, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 4, "training_row": 2, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false }, { "step": 4, "training_row": 6, "rewards": [ 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2 ], "optimizer_update_executed": false, "zero_advantage_group": true, "advantages_recorded": null, "advantages_field_present": false } ], "all_groups_audited": 32, "invocation_elapsed_seconds": 673.5393040217459, "configured_wall_cap_seconds": 900, "peak_memory": { "core_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "core_runner_sha256": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c", "core_import_binding": "prepend frozen core root/src; refuse mismatched preloaded package", "seed": 44, "variant": "unfiltered", "execution_config_canonical_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "artifact_kind": "observational_cuda_memory_not_gpu_receipt", "not_gpu_receipt": true, "core_exit_code": 0, "actual_elapsed_seconds": 671.0687175318599, "no_monkeypatch": true, "training_arguments_forwarded_unchanged": true, "measurement_scope": "whole frozen-core invocation after reset; per-process Torch allocator, not device-global NVML peak", "peak_memory_available": true, "actual_max_memory_allocated_bytes": 3770255360, "actual_max_memory_reserved_bytes": 5515509760, "accelerator": { "vendor": "NVIDIA", "model": "NVIDIA A30" } }, "measured_cost": { "generation_calls": { "evaluation": 24, "student": 32 }, "generated_tokens": { "evaluation": 8471, "student": 75016 }, "scored_response_tokens": { "behavior": 75016, "student_rl": 55419 }, "frozen_encoder_forward_calls": 0, "frozen_encoder_input_tokens_including_prompt": 0, "scope": "entire invocation, including training controls, profiling and evaluation", "includes_rejected_and_profiling_generations": true }, "execution_complete": true, "mechanism_passed": true, "negative_control_completed": false } } }, "evaluation_protocol": { "tier": "l1_mechanism", "seeds": [ 42, 43, 44 ], "formal_comparison": false, "claim_policy": "single/few-seed smoke result; do not claim a stable improvement", "promotion_eligible": false, "test_used_for_selection": false, "within_A30_same_seed_arms_paired": true, "paired_with_old_A100_or_51b": false, "variants_are_independent_seeds": false, "concurrent_latency_comparison_eligible": false }, "provenance": { "code_commit": "b90862754bf51b1beeb5add3f7dc2a51ca00ace4", "executed_source_sha256": { "scripts/prepare_soar_catalog_qa.py": "483a4eb3d9932d31046888c165c06590e67c6df5575a5b2afd4e302145f445e7", "scripts/run_oct13_post_training.py": "916b3b9a0eaeeaf30bad811376fa05305011754c922bb368a0f39833aad07ffa", "src/auto_research/oct12_public_sources.py": "4e1150cbe4ea975819c244267180f853c83c4ad4c2c27f9e0ce95cc16793e093", "src/auto_research/post_training/oct10_checkpoint.py": "247f1f97a3ca701277f241ec81d99296f0ea82e1db8f68006fcd7bfdc661cd74", "src/auto_research/post_training/oct13_common.py": "d19401266a420afdc64867f145f8384fd649dc1991e4b74ea1f7ee8b8d91bde5", "src/auto_research/post_training/proximal_checkpoint.py": "5307a54204720e5407d2cd0435b654ae0e594f30c85cdf07d55d4e90612a1397", "src/auto_research/post_training/response_policy.py": "e380a5b4d1d53974b973b72b7d9ad6328129dc8c9bae4d233f6c1cbdf8508bf7", "src/auto_research/public_holdout.py": "033d9c1fd112accf5ac79573b99d9f31cffc646bcab6911cf3e49459fa003b46", "src/auto_research/reproductions/gppo_framework/__init__.py": "7fe8cab42fef9f32b9653dfb5f54f03182de0753de2c9e5b359d124d7cd763d8", "src/auto_research/reproductions/gppo_framework/adapter.py": "a7f4ee792a8ded3daeeb183cbca5c99de63c1b7d3cb195f5bedd5289cb0e0260", "src/auto_research/reproductions/gppo_framework/objective.py": "0f8fbb2ad397b04a70466dde4e939f4d094ad7e7df07cfbba59f4a6ca5d05dd0", "src/auto_research/reproductions/gppo_framework/paper.yaml": "a023ee98bbd078077b2708122973d27a98085848be816ce461b7d4291aca8e1f", "src/auto_research/reproductions/grpo_difficulty_curriculum/__init__.py": "d09d8d7e0e6f5bcec9979241eed06a1d2a6ac8f3fb7a25f49e0df55be0071acb", "src/auto_research/reproductions/grpo_difficulty_curriculum/adapter.py": "bbd3e964dc035ea4bbd999a581ea07eccabd986b6456774b21a4c24092cbdfbe", "src/auto_research/reproductions/grpo_difficulty_curriculum/algorithm.py": "4131e39279d30c0748b14f611ab657b3fea76147a7c164aea2883c62bde75785", "src/auto_research/reproductions/grpo_difficulty_curriculum/checkpoint.py": "d03a7d2db0a19b469fe5429ddce634b02d8cbd0eeed8d09bef0afc51c1d7ac70", "src/auto_research/reproductions/grpo_difficulty_curriculum/paper.yaml": "c417d2269314b15e2dfef4ba52dfb24771b2e1270f808fe5be47a54e9d0c1bb1", "src/auto_research/reproductions/magicbeanstalk/__init__.py": "8dd207ea37a89dfa34f87b02eb4a2fd26472895678d1362a7e5b59dad5cda616", "src/auto_research/reproductions/magicbeanstalk/adapter.py": "319b6503de342feee49fb6446b7bbdeb117f7dc0dd7b79ed1c3bd34cce1ce73f", "src/auto_research/reproductions/magicbeanstalk/algorithm.py": "2fb98f7137f336414ae6e07069fe34584d0e1cbc4bd594683a67e11ffc982aba", "src/auto_research/reproductions/magicbeanstalk/checkpoint.py": "78416567c61f23505f579805b845a40a3098c311e57fb42c7ac8d5ebc67390dd", "src/auto_research/reproductions/magicbeanstalk/paper.yaml": "e9c5619918b4d72d776008aba3fa95d50ba6ebe99afab729ba9e6dfe82fac06e", "src/auto_research/reproductions/progressive_contraction/__init__.py": "e50fe46d888e0a0b11b7d4d6d8f3224ae5940394f8ba8396fc5afb1cd38a4d53", "src/auto_research/reproductions/progressive_contraction/adapter.py": "06f23b5c795f7222f76dee9beaa8164f460eb5183e18c6d374bff011737235dd", "src/auto_research/reproductions/progressive_contraction/algorithm.py": "ba077b8c52a8d10e268a4e5c85770b9cebcb12415f9b2aaf7b9500e1303ca4b1", "src/auto_research/reproductions/progressive_contraction/checkpoint.py": "67fa44814d613d3838716fea68476217d608af14f23a051e32a916b72538e7ec", "src/auto_research/reproductions/progressive_contraction/paper.yaml": "faf5243c71a87407601b982469d98dc4d48ddf842da56c210ede077c2b27c1c1", "src/auto_research/reproductions/soar/__init__.py": "3d87d1189d7037e32c1e2cb0c249d13609c5af3d7062a19904590ba7923c799a", "src/auto_research/reproductions/soar/adapter.py": "8fbbe8f8fe54b9f014ccc26a75b01e72b73b95e2b03cc379c5b5db4b63c9b31e", "src/auto_research/reproductions/soar/algorithm.py": "9b7068d9774d1c93800c512a04b9ba4d0ee98fc13d308e5ac0970299c8bb2485", "src/auto_research/reproductions/soar/checkpoint.py": "91901d9eb9f81faa41dda170ff1f4d2283f56cd17c7df0572348356c7b7eb76e", "src/auto_research/reproductions/soar/paper.yaml": "2f756cab9f7fefd4613bb69971fd8252c2b9ad451fa5c5eff5d03fa063d91521", "src/auto_research/reproductions/state_opd/__init__.py": "8db07736f0336ed7f39502ae2953bdae76658942542cacbe4b6b2c1fd0abe094", "src/auto_research/reproductions/state_opd/adapter.py": "2f8bf4773e8f40c0eabe777fbd1e1aa6d372a7b05501404c3e3f7b2d54987524", "src/auto_research/reproductions/state_opd/algorithm.py": "dc7fe3eae619ff69ec51c47826f89471f00be564ea4e6730e12726a060ba7e2f", "src/auto_research/reproductions/state_opd/checkpoint.py": "95755e0604f563a5c96f92e9de9069befe3e1274d31c224a43ca20d71f756b09", "src/auto_research/reproductions/state_opd/environment.py": "0e9564f9ec540985eb010b1e70f46938acc1bd946041b518d6579253762080e1", "src/auto_research/reproductions/state_opd/paper.yaml": "86c75255f1c416611a181c26359e9f5d9865caaea5266ca7b13dd8a183b906cc", "src/auto_research/reproductions/synthself/__init__.py": "474bb75e10b95afe6faec1f3631beab2f17f9a43013d3d29e79b7d4b36e06031", "src/auto_research/reproductions/synthself/adapter.py": "db4c08e1ccc2acfc28d19fc72acfa39b0a0dce096bbc430bafe9b21dbb72cddd", "src/auto_research/reproductions/synthself/algorithm.py": "1e7ea1ff65d3ba0ece1af889022c19f7147f51d10fb8196ab8f9fa2e18604b95", "src/auto_research/reproductions/synthself/checkpoint.py": "dbcf9f222094c197e0392f82191a0ae6551c6573ee54a9eee8c7d37553472981", "src/auto_research/reproductions/synthself/paper.yaml": "d6d31d755561626c2b667f4fc2b990dee8360fa52f92193b5c6a6e93b1f6ca71", "src/auto_research/reproductions/tapo_trajectory/__init__.py": "a0dd8d4825b91dd076b71bdd3a3eccb0638ba4bf0240fe9da76d61a7290f69c6", "src/auto_research/reproductions/tapo_trajectory/adapter.py": "c01b75bf7853546caa2d517dd5f9c9faf739537b95207b8f1fd28adcdd5aafa7", "src/auto_research/reproductions/tapo_trajectory/objective.py": "49ab02ac758251acde90091a82b22576449d294a1f25cc3144ff1addd1132ae2", "src/auto_research/reproductions/tapo_trajectory/paper.yaml": "493e59ec67f96ee37ffecf6694dd5c4d8555849b8f756ccb63644da90bbb48d6", "src/auto_research/reproductions/upturn/__init__.py": "1e9793e9ea85e94144f7eac8fa141567d7f964438fa628788b668358154e196e", "src/auto_research/reproductions/upturn/adapter.py": "72e5da3cb70856944d0c9a25376f73d4a1d9f16213efd43e19b3acf2a584afc2", "src/auto_research/reproductions/upturn/algorithm.py": "421abd06cc492c39cf72e9b15c7f85c01e85fae0a7c26fb3053fbd19bc4897a4", "src/auto_research/reproductions/upturn/checkpoint.py": "fbcb7c830cf53d3130cfc06621f3178516aa72c7a939f4f765c3c7864b383787", "src/auto_research/reproductions/upturn/paper.yaml": "6eca098f6d945a30732f54cb04f2ee984b165a3c4f06cb65d49879d7edd28696", "scripts/diagnose_oct14_post_training.py": "2a15d720a4f84bd1e415a4f0653c589c9716d526d4842acd5df728fd3575ca5b", "scripts/diagnose_oct14_soar_format.py": "02b375e4359b8b8031246853a7e30e52b0cf7e0a94cf563299758260953f7ded" }, "artifact_path": "docs/reproductions/2610.13669-grpo-difficulty-curriculum/metrics/public-a30-complete-five-arms-seeds42-44-b9086275.json", "execution_config_sha256": "96a8984e40ec68ca01186a33a150dfea89c100c7a74d8e613f8f905f98d9b3f9", "dataset_fingerprint": "69efc634f51a589335fcdea85238bf3218653a94de2ba2604a64056960547dab", "raw_results_sha256": { "42.unfiltered": "eb2620796f9897761d9fdee5c5618e0bd41127066890a2491ea3c016ecf839d1", "42.static": "d22403f608483bbfb0778a6c9e11e22d8b8e9e0efb6134e6bf57590550728670", "42.fixed-medium": "bfe4bdae3c65fb7145f15d6465005214745694ea2d0cb54b0ca5a43970e59723", "42.staged": "1759a9d1f8af5119b868fa16b2467f2dada662cedeaf991fe1c0cbd7f27d6cda", "42.too-hard": "1508e40c26ae7e170be357aa2586a15b7979444da016db2abc8d36aa65a52fba", "43.unfiltered": "91551fdeebc05f36c25d3415151b31f84a88f2dc9048131d34864e3dfa82206c", "43.static": "b541c20db2c11c39cd354108b5c7c0b68afdbabcc47967e1a92f72af11286a38", "43.fixed-medium": "93ef5dc2d7e724d3e54eab1b72349af55041a3ac5324475e8fcc421268417add", "43.staged": "40d289e30a4d6c3dc9d505e3d1b0e000e54315b20958e347cebc4696439b8187", "43.too-hard": "8ef0d91ba2dbf8f6f51b9e0e237333b7a5c4de7c392a1cdd7f098e5bd528774e", "44.unfiltered": "8feae863af6560ef29f70f14541e2766d03ae4c397d1f9a4a158e4451b7eadbd", "44.static": "69a614861d169277bad3b6c34b2286d98f342c9c51c685123ef0004a38506132", "44.fixed-medium": "c0c09458e3cac25b95f67263240c6355b9ec44210d338a1ed1f5d843ca33235a", "44.staged": "9d75d1054d367bb518e2382cfbff5e5d0bcc7fbe7dde066d93f93342072b2681", "44.too-hard": "6575fa47b9fe466e29386049a4bc4fd51ba53cfe097a29dc389d99841144b2ce" }, "observer_commit": "92c2081869dcfbcec89b85580721d9bdb0dd136c", "observer_source_sha256": { "scripts/observe_oct14_cuda_memory.py": "e106b23ff74d4d7ffc4572fddd77780ef1811eb2f33cb1967d8f86776874d36c" } }, "limitations": [ "Zero and rejected groups are fully audited, not dropped from denominators; no old-source pseudo updates are reused.", "Five variants are paired method/control arms, not five independent seeds; only actually completed seeds enter descriptive metrics.", "Matched configured budget is not matched realized updates/token cost; concurrency timing is not a throughput comparison.", "Every first-seed and remaining-seed arm is same-device/source/config; old6a/51b cohorts are not reused or pooled.", "Three seed groups remain L1 diagnostics, not a stable gain, original paper benchmark or production A/B.", "Negative staged and any other lower validation metrics are preserved; too-hard is not forced to have zero updates." ] }