RealArDor's picture
download
raw
8.01 kB
{
"input": "/data/rwang/nanojev_sonic_sft_20260920/unified/hard",
"init_checkpoint": "/tmp/nanojev_sonic_sft_20260920/initial_checkpoint",
"output_dir": "/data/rwang/nanojev_sonic_sft_20260920/experiment/hard_lr1e5",
"stage": "sft",
"loss": "ce",
"balance": "task",
"policy_pool_weights": "/data/rwang/nanojev_sonic_sft_20260920/experiment/population_weights.json",
"retention_fraction": 0.25,
"reward_samples": 32,
"steps": 600,
"head_steps": 0,
"batch_questions": 24,
"microbatch_questions": 8,
"max_microbatch_tokens": 32768,
"max_length": 8192,
"eval_every": 100,
"seed": 17,
"backbone_lr": 1e-05,
"head_lr": 0.0001,
"head_warmup_lr": 0.001,
"weight_decay": 0.01,
"precision": "bf16",
"gradient_checkpointing": true,
"disable_native_triton": true,
"td_episodes": null,
"td_n_step": 3,
"td_weight": 0.0,
"target_update_every": 25,
"validate_only": false,
"resolved_policy_pool_weights": {
"maze/policy": 0.3333333333333333,
"snake/policy": 0.3333333333333333,
"shooting/policy/basic": 0.16666666666666666,
"shooting/policy/predict_position": 0.16666666666666666
},
"schema_version": "nanojev-unified-games-v1",
"model": "Qwen/Qwen3-0.6B",
"set_head": "attention",
"resolved_model_revision": "c1899de289a04d12100db370d81485cdf75e47ca",
"initialization": "local DecisionModel warm start; fresh optimizer",
"init_weights_sha256": "38116340795de1c82369b7fe15819d92d79600a7b4dc7a3cd0d4390cb6782639",
"data_sha256": {
"/data/rwang/nanojev_sonic_sft_20260920/unified/hard/manifest.json": "e79f0c3a8559e480db3161781963b1ac7c69be83772eb889208532e4cc5ab80a",
"/data/rwang/nanojev_sonic_sft_20260920/unified/hard/train.jsonl": "bda8712f89d5db94d80f840f9620f18312c92baca2d8a6df3456218fc807acd0",
"/data/rwang/nanojev_sonic_sft_20260920/unified/hard/dev.jsonl": "605b40f56e4834b0f234f3328b1e6a8ed41174641986b229e94927a3f05f70bd",
"/data/rwang/nanojev_sonic_sft_20260920/unified/hard/calibration.jsonl": "6408aa09b9238c724c557c86b68b23f2359140911d84daac7693f9e347fa091b",
"/data/rwang/nanojev_sonic_sft_20260920/unified/hard/test.jsonl": "2f96cc4e45c67fa375980ff4f355b47716985cf5fb4a1ef034c5e65481b072c9",
"/data/rwang/nanojev_sonic_sft_20260920/unified/hard/ood.jsonl": "7a778752e0f7699bd709aa6c83bb1ffaaa205568c304909000039b40fb7ff636"
},
"implementation_sha256": "a4aedc3b7ad25d875af67943bef6dbc017799f46c4ae0a845fc200e1866fa6cd",
"continuation_policy_id": null,
"population_weights": {
"maze/policy": 0.3333333333333333,
"snake/policy": 0.3333333333333333,
"shooting/policy/basic": 0.16666666666666666,
"shooting/policy/predict_position": 0.16666666666666666
},
"policy_pool_weights_sha256": "66afd9cdb9f0479a77896671c201834db1e5889ea081f81aa068dc73359f0449",
"eligible_by_split_sampling_pool": {
"calibration/maze/policy": 188,
"calibration/shooting/policy/basic": 377,
"calibration/shooting/policy/predict_position": 1006,
"calibration/snake/policy": 137,
"dev/maze/policy": 213,
"dev/shooting/policy/basic": 384,
"dev/shooting/policy/predict_position": 958,
"dev/snake/policy": 157,
"ood/maze/policy": 202,
"ood/shooting/policy/basic": 589,
"ood/shooting/policy/predict_position": 1008,
"ood/snake/policy": 141,
"test/maze/policy": 211,
"test/shooting/policy/basic": 756,
"test/shooting/policy/predict_position": 1413,
"test/snake/policy": 116,
"train/maze/policy": 651,
"train/shooting/policy/basic": 3054,
"train/shooting/policy/predict_position": 6788,
"train/snake/policy": 400
},
"sampling": "stratified with replacement; per-cell exact population weights; train rows only",
"parameter_storage": "float32",
"forward_autocast": "bf16",
"objective": "Choice CE with explicit API or expert targets plus observed Boolean event loss",
"selection": "minimum fixed population-weighted dev CE including initial checkpoint; test only after selection",
"temperature": 1.0,
"temperature_fitted": false,
"deps": {
"torch": "2.14.0",
"transformers": "5.17.0",
"safetensors": "0.8.0"
},
"gpu": "NVIDIA A100-SXM4-80GB",
"schema_counts": {
"records": 18760,
"outcome_questions": 0,
"continuation_policy_id": null,
"questions_by_split_task_role": {
"calibration/maze/policy": 188,
"calibration/shooting/policy": 1383,
"calibration/snake/policy": 138,
"dev/maze/policy": 215,
"dev/shooting/policy": 1342,
"dev/snake/policy": 158,
"ood/maze/policy": 202,
"ood/shooting/policy": 1597,
"ood/snake/policy": 143,
"test/maze/policy": 211,
"test/shooting/policy": 2169,
"test/snake/policy": 116,
"train/maze/policy": 653,
"train/shooting/policy": 9842,
"train/snake/policy": 403
},
"eligible_by_split_task_role": {
"calibration/maze/policy": 188,
"calibration/shooting/policy": 1383,
"calibration/snake/policy": 137,
"dev/maze/policy": 213,
"dev/shooting/policy": 1342,
"dev/snake/policy": 157,
"ood/maze/policy": 202,
"ood/shooting/policy": 1597,
"ood/snake/policy": 141,
"test/maze/policy": 211,
"test/shooting/policy": 2169,
"test/snake/policy": 116,
"train/maze/policy": 651,
"train/shooting/policy": 9842,
"train/snake/policy": 400
},
"quarantined_policy_questions": 11,
"policy_questions_by_split_task_target_kind": {
"calibration/maze/api_policy_distribution": 188,
"calibration/shooting/expert_action": 1383,
"calibration/snake/api_policy_distribution": 138,
"dev/maze/api_policy_distribution": 215,
"dev/shooting/expert_action": 1342,
"dev/snake/api_policy_distribution": 158,
"ood/maze/api_policy_distribution": 202,
"ood/shooting/expert_action": 1597,
"ood/snake/api_policy_distribution": 143,
"test/maze/api_policy_distribution": 211,
"test/shooting/expert_action": 2169,
"test/snake/api_policy_distribution": 116,
"train/maze/api_policy_distribution": 653,
"train/shooting/expert_action": 9842,
"train/snake/api_policy_distribution": 403
},
"unique_episodes_by_split_task_role": {
"calibration/maze/policy": 10,
"calibration/shooting/policy": 128,
"calibration/snake/policy": 8,
"dev/maze/policy": 10,
"dev/shooting/policy": 128,
"dev/snake/policy": 8,
"ood/maze/policy": 10,
"ood/shooting/policy": 256,
"ood/snake/policy": 8,
"test/maze/policy": 10,
"test/shooting/policy": 256,
"test/snake/policy": 8,
"train/maze/policy": 32,
"train/shooting/policy": 1024,
"train/snake/policy": 28
},
"rows_missing_episode_id_by_split_task_role": {},
"outcome_episode_counts_by_split_task": {},
"outcome_state_label_counts_by_split_task": {},
"outcome_question_label_counts_by_split_task": {},
"episode_audit_note": "Repeated states/questions are not independent episodes. Single-outcome cells are retained without forced balancing. Missing episode IDs are counted explicitly and excluded from episode totals."
},
"predictive_samples_are_physical_actions": false,
"frozen_continuation_policy_is_updated_by_this_script": false,
"td_enabled": false,
"td_audit": null,
"td_implementation_sha256": null,
"td_contract": {
"active": false,
"gamma": 1,
"n_step_unit": "recorded decision transitions",
"continuation_policy_id": null,
"bootstrap_policy": "recorded frozen behavior_probs; never online greedy",
"target_microbatch_questions": 4,
"max_microbatch_tokens": 32768,
"heldout_target": "observed terminal outcome only",
"outcome_loss": "(1-td_weight)*MC_Brier + td_weight*detached_soft_TD_Brier",
"initial_target": "deepcopy of initial online model; eval mode; no gradients",
"prediction_cache": "none; only bootstrap input tokens and indices are cached",
"bootstrap_cache": null
}
}

Xet Storage Details

Size:
8.01 kB
·
Xet hash:
8499e59191167a29a39491b2fa6dbe8dcdd54c327c61ecff40f9a50e81066be4

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.