Instructions to use arcadia-impact/dispatch-models with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use arcadia-impact/dispatch-models with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
rollouts: coin-direct-run2/coin-direct-phase768/rollouts/raw_rollouts.rank-0.jsonl provenance
Browse files
rollouts/coin-direct-run2/coin-direct-phase768/rollouts/raw_rollouts.rank-0.jsonl.meta.json
ADDED
|
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"source_repo": "arcadia-impact/scimt-dispatch-rlvr-gemma4-26b-v1-runs",
|
| 3 |
+
"source_path": "coin-direct-run2/coin-direct-phase768/rollouts/raw_rollouts.rank-0.jsonl",
|
| 4 |
+
"declared_bytes": 6867643203,
|
| 5 |
+
"streamed_bytes": 6947819090,
|
| 6 |
+
"sha256_streamed": "ce34ba89e7fae2e1c8ae6d5ec11b0817d6c3f851efd501a7fe5e9bc64403b23d",
|
| 7 |
+
"sha256_expected": "ce34ba89e7fae2e1c8ae6d5ec11b0817d6c3f851efd501a7fe5e9bc64403b23d",
|
| 8 |
+
"sha256_match": true,
|
| 9 |
+
"rows": 20032,
|
| 10 |
+
"dropped_fields": [
|
| 11 |
+
"completion_ids",
|
| 12 |
+
"log_extra",
|
| 13 |
+
"log_metric",
|
| 14 |
+
"trainer_state"
|
| 15 |
+
],
|
| 16 |
+
"trainer_state": "TrainerState(epoch=0.041666666666666664, global_step=32, max_steps=768, logging_steps=1, eval_steps=500, save_steps=17, train_batch_size=4, num_train_epochs=1, num_input_tokens_seen=2277455, total_flos=0.0, log_history=[{'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 349.0, 'completions/max_terminated_length': 349.0, 'completions/mean_length': 57.390625, 'completions/mean_terminated_length': 57.390625, 'completions/min_length': 17.0, 'completions/min_terminated_length': 17.0, 'entropy': 0.08352478453889489, 'epoch': 0.0013020833333333333, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.016480544582009315, 'learning_rate': 1e-05, 'loss': 0.0013222710695117712, 'num_tokens': 63257.0, 'reward': 0.53125, 'reward/selected_zero_std_group_fraction': 0.0, 'reward/zero_std_group_fraction': 0.5, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.78125, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.21875, 'reward_components/parser_valid': 0.78125, 'reward_components/reward': 0.53125, 'reward_components/runs_correct': 1.125, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.65625, 'reward_std': 0.5029674172401428, 'rewards/reward_func/mean': 0.53125, 'rewards/reward_func/std': 0.5029674172401428, 'sampling/importance_sampling_ratio/max': 2.093050003051758, 'sampling/importance_sampling_ratio/mean': 0.6828386783599854, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 5.9159393310546875, 'sampling/sampling_logp_difference/mean': 0.0515105314552784, 'step': 1, 'step_time': 19.43506328598596}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 264.0, 'completions/max_terminated_length': 264.0, 'completions/mean_length': 38.109375, 'completions/mean_terminated_length': 38.109375, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.12295123049989343, 'epoch': 0.0026041666666666665, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.04708399623632431, 'learning_rate': 1e-05, 'loss': -0.0025684507563710213, 'num_tokens': 143072.0, 'reward': 0.765625, 'reward/selected_zero_std_group_fraction': 0.125, 'reward/zero_std_group_fraction': 0.5625, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.8125, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.1875, 'reward_components/parser_valid': 0.8125, 'reward_components/reward': 0.765625, 'reward_components/runs_correct': 1.015625, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.765625, 'reward_std': 0.42695629596710205, 'rewards/reward_func/mean': 0.765625, 'rewards/reward_func/std': 0.42695629596710205, 'sampling/importance_sampling_ratio/max': 1.7880834341049194, 'sampling/importance_sampling_ratio/mean': 0.829391598701477, 'sampling/importance_sampling_ratio/min': 0.0022463519126176834, 'sampling/sampling_logp_difference/max': 4.353402137756348, 'sampling/sampling_logp_difference/mean': 0.03568439558148384, 'step': 2, 'step_time': 18.90319765289314}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 155.0, 'completions/max_terminated_length': 155.0, 'completions/mean_length': 40.015625, 'completions/mean_terminated_length': 40.015625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.13309571892023087, 'epoch': 0.00390625, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.06005037575960159, 'learning_rate': 1e-05, 'loss': 0.002461154479533434, 'num_tokens': 202785.0, 'reward': 0.734375, 'reward/selected_zero_std_group_fraction': 0.08333333333333333, 'reward/zero_std_group_fraction': 0.5416666666666666, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.78125, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.78125, 'reward_components/reward': 0.734375, 'reward_components/runs_correct': 1.203125, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.734375, 'reward_std': 0.44515693187713623, 'rewards/reward_func/mean': 0.734375, 'rewards/reward_func/std': 0.44515693187713623, 'sampling/importance_sampling_ratio/max': 1.8478593826293945, 'sampling/importance_sampling_ratio/mean': 0.5296120643615723, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.872286319732666, 'sampling/sampling_logp_difference/mean': 0.0692758858203888, 'step': 3, 'step_time': 13.432064907159656}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 239.0, 'completions/max_terminated_length': 239.0, 'completions/mean_length': 39.6875, 'completions/mean_terminated_length': 39.6875, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.1007397030480206, 'epoch': 0.005208333333333333, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.028138406574726105, 'learning_rate': 1e-05, 'loss': 0.0018308290746062994, 'num_tokens': 282573.0, 'reward': 0.546875, 'reward/selected_zero_std_group_fraction': 0.1875, 'reward/zero_std_group_fraction': 0.59375, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.578125, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.015625, 'reward_components/parser_natural': 0.859375, 'reward_components/parser_unsafe': 0.421875, 'reward_components/parser_valid': 0.578125, 'reward_components/reward': 0.546875, 'reward_components/runs_correct': 0.84375, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.5625, 'reward_std': 0.501733124256134, 'rewards/reward_func/mean': 0.546875, 'rewards/reward_func/std': 0.501733124256134, 'sampling/importance_sampling_ratio/max': 1.7865898609161377, 'sampling/importance_sampling_ratio/mean': 0.6803629398345947, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 10.87292766571045, 'sampling/sampling_logp_difference/mean': 0.046786122024059296, 'step': 4, 'step_time': 17.350974324159324}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 325.0, 'completions/max_terminated_length': 325.0, 'completions/mean_length': 64.203125, 'completions/mean_terminated_length': 64.203125, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.10676396731287241, 'epoch': 0.006510416666666667, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.1702335774898529, 'learning_rate': 1e-05, 'loss': 0.01798689365386963, 'num_tokens': 351194.0, 'reward': 0.390625, 'reward/selected_zero_std_group_fraction': 0.15, 'reward/zero_std_group_fraction': 0.575, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.625, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.015625, 'reward_components/parser_natural': 0.984375, 'reward_components/parser_unsafe': 0.328125, 'reward_components/parser_valid': 0.625, 'reward_components/reward': 0.390625, 'reward_components/runs_correct': 0.828125, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.5078125, 'reward_std': 0.4917473793029785, 'rewards/reward_func/mean': 0.390625, 'rewards/reward_func/std': 0.4917473793029785, 'sampling/importance_sampling_ratio/max': 2.4725868701934814, 'sampling/importance_sampling_ratio/mean': 0.5025970935821533, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 5.562561988830566, 'sampling/sampling_logp_difference/mean': 0.05451514944434166, 'step': 5, 'step_time': 18.429810586269014}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 146.0, 'completions/max_terminated_length': 146.0, 'completions/mean_length': 37.75, 'completions/mean_terminated_length': 37.75, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.10422404948621988, 'epoch': 0.0078125, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.03020234778523445, 'learning_rate': 1e-05, 'loss': -0.0015337434597313404, 'num_tokens': 420426.0, 'reward': 0.46875, 'reward/selected_zero_std_group_fraction': 0.16666666666666666, 'reward/zero_std_group_fraction': 0.5833333333333334, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.796875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.203125, 'reward_components/parser_valid': 0.796875, 'reward_components/reward': 0.46875, 'reward_components/runs_correct': 0.859375, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.53125, 'reward_std': 0.5029674172401428, 'rewards/reward_func/mean': 0.46875, 'rewards/reward_func/std': 0.5029674172401428, 'sampling/importance_sampling_ratio/max': 2.578643560409546, 'sampling/importance_sampling_ratio/mean': 0.524068295955658, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.202038288116455, 'sampling/sampling_logp_difference/mean': 0.05009016767144203, 'step': 6, 'step_time': 14.604830794152804}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 197.0, 'completions/max_terminated_length': 197.0, 'completions/mean_length': 39.609375, 'completions/mean_terminated_length': 39.609375, 'completions/min_length': 16.0, 'completions/min_terminated_length': 16.0, 'entropy': 0.09948836034163833, 'epoch': 0.009114583333333334, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.008704844862222672, 'learning_rate': 1e-05, 'loss': -0.004645691718906164, 'num_tokens': 488561.0, 'reward': 0.734375, 'reward/selected_zero_std_group_fraction': 0.25, 'reward/zero_std_group_fraction': 0.625, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.984375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.015625, 'reward_components/parser_valid': 0.984375, 'reward_components/reward': 0.734375, 'reward_components/runs_correct': 1.21875, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.796875, 'reward_std': 0.44515693187713623, 'rewards/reward_func/mean': 0.734375, 'rewards/reward_func/std': 0.44515693187713623, 'sampling/importance_sampling_ratio/max': 2.8380684852600098, 'sampling/importance_sampling_ratio/mean': 0.9944190979003906, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 6.536219596862793, 'sampling/sampling_logp_difference/mean': 0.025410043075680733, 'step': 7, 'step_time': 14.536820802139118}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 101.0, 'completions/max_terminated_length': 101.0, 'completions/mean_length': 29.78125, 'completions/mean_terminated_length': 29.78125, 'completions/min_length': 11.0, 'completions/min_terminated_length': 11.0, 'entropy': 0.08490662928670645, 'epoch': 0.010416666666666666, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.020627930760383606, 'learning_rate': 1e-05, 'loss': 0.0013012087438255548, 'num_tokens': 561507.0, 'reward': 0.484375, 'reward/selected_zero_std_group_fraction': 0.3125, 'reward/zero_std_group_fraction': 0.65625, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.84375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.15625, 'reward_components/parser_valid': 0.84375, 'reward_components/reward': 0.484375, 'reward_components/runs_correct': 1.109375, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.6171875, 'reward_std': 0.5037065148353577, 'rewards/reward_func/mean': 0.484375, 'rewards/reward_func/std': 0.5037065148353577, 'sampling/importance_sampling_ratio/max': 1.9595078229904175, 'sampling/importance_sampling_ratio/mean': 0.6667697429656982, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 12.483682632446289, 'sampling/sampling_logp_difference/mean': 0.0844041258096695, 'step': 8, 'step_time': 13.747618639841676}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 53.0, 'completions/max_terminated_length': 53.0, 'completions/mean_length': 25.015625, 'completions/mean_terminated_length': 25.015625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.06506687146611512, 'epoch': 0.01171875, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.09078060835599899, 'learning_rate': 1e-05, 'loss': 0.004498262424021959, 'num_tokens': 643364.0, 'reward': 0.671875, 'reward/selected_zero_std_group_fraction': 0.2777777777777778, 'reward/zero_std_group_fraction': 0.6388888888888888, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.875, 'reward_components/reward': 0.671875, 'reward_components/runs_correct': 1.171875, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.734375, 'reward_std': 0.4732423722743988, 'rewards/reward_func/mean': 0.671875, 'rewards/reward_func/std': 0.4732423722743988, 'sampling/importance_sampling_ratio/max': 2.255783796310425, 'sampling/importance_sampling_ratio/mean': 0.7643908262252808, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.262850284576416, 'sampling/sampling_logp_difference/mean': 0.04843715578317642, 'step': 9, 'step_time': 14.498103346908465}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 239.0, 'completions/max_terminated_length': 239.0, 'completions/mean_length': 41.703125, 'completions/mean_terminated_length': 41.703125, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.0654769972898066, 'epoch': 0.013020833333333334, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.18836969137191772, 'learning_rate': 1e-05, 'loss': 0.0033059692941606045, 'num_tokens': 699985.0, 'reward': 0.71875, 'reward/selected_zero_std_group_fraction': 0.275, 'reward/zero_std_group_fraction': 0.6375, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.734375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.03125, 'reward_components/parser_natural': 0.96875, 'reward_components/parser_unsafe': 0.21875, 'reward_components/parser_valid': 0.734375, 'reward_components/reward': 0.71875, 'reward_components/runs_correct': 0.84375, 'reward_components/runs_total': 1.125, 'reward_components/semantic_correct': 0.71875, 'reward_std': 0.4531635046005249, 'rewards/reward_func/mean': 0.71875, 'rewards/reward_func/std': 0.4531635046005249, 'sampling/importance_sampling_ratio/max': 1.2312541007995605, 'sampling/importance_sampling_ratio/mean': 0.7441521286964417, 'sampling/importance_sampling_ratio/min': 3.263890175730921e-05, 'sampling/sampling_logp_difference/max': 6.196830749511719, 'sampling/sampling_logp_difference/mean': 0.033542830497026443, 'step': 10, 'step_time': 13.446841639932245}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 421.0, 'completions/max_terminated_length': 421.0, 'completions/mean_length': 37.8125, 'completions/mean_terminated_length': 37.8125, 'completions/min_length': 11.0, 'completions/min_terminated_length': 11.0, 'entropy': 0.09222514135763049, 'epoch': 0.014322916666666666, 'frac_reward_zero_std': 0.5, 'grad_norm': 0.021602613851428032, 'learning_rate': 1e-05, 'loss': -0.0028603768441826105, 'num_tokens': 761925.0, 'reward': 0.78125, 'reward/selected_zero_std_group_fraction': 0.25, 'reward/zero_std_group_fraction': 0.625, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.9375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.046875, 'reward_components/parser_valid': 0.9375, 'reward_components/reward': 0.78125, 'reward_components/runs_correct': 1.03125, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.796875, 'reward_std': 0.4166666865348816, 'rewards/reward_func/mean': 0.78125, 'rewards/reward_func/std': 0.4166666865348816, 'sampling/importance_sampling_ratio/max': 2.717414140701294, 'sampling/importance_sampling_ratio/mean': 0.8617135882377625, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 2.017646551132202, 'sampling/sampling_logp_difference/mean': 0.030264852568507195, 'step': 11, 'step_time': 17.05516506114509}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 210.0, 'completions/max_terminated_length': 210.0, 'completions/mean_length': 31.15625, 'completions/mean_terminated_length': 31.15625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.053772636369103566, 'epoch': 0.015625, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.010692600160837173, 'learning_rate': 1e-05, 'loss': 0.00038697264972142875, 'num_tokens': 843599.0, 'reward': 0.734375, 'reward/selected_zero_std_group_fraction': 0.2916666666666667, 'reward/zero_std_group_fraction': 0.6458333333333334, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.859375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.140625, 'reward_components/parser_valid': 0.859375, 'reward_components/reward': 0.734375, 'reward_components/runs_correct': 0.984375, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.796875, 'reward_std': 0.44515693187713623, 'rewards/reward_func/mean': 0.734375, 'rewards/reward_func/std': 0.44515693187713623, 'sampling/importance_sampling_ratio/max': 1.3127986192703247, 'sampling/importance_sampling_ratio/mean': 0.8477168679237366, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 3.2836036682128906, 'sampling/sampling_logp_difference/mean': 0.023977721109986305, 'step': 12, 'step_time': 17.140284939086996}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 83.0, 'completions/max_terminated_length': 83.0, 'completions/mean_length': 19.40625, 'completions/mean_terminated_length': 19.40625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.04930048482492566, 'epoch': 0.016927083333333332, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.03135985881090164, 'learning_rate': 1e-05, 'loss': -0.0024681859649717808, 'num_tokens': 898921.0, 'reward': 0.578125, 'reward/selected_zero_std_group_fraction': 0.3076923076923077, 'reward/zero_std_group_fraction': 0.6538461538461539, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.9375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.0625, 'reward_components/parser_valid': 0.9375, 'reward_components/reward': 0.578125, 'reward_components/runs_correct': 0.890625, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.640625, 'reward_std': 0.49776285886764526, 'rewards/reward_func/mean': 0.578125, 'rewards/reward_func/std': 0.49776285886764526, 'sampling/importance_sampling_ratio/max': 2.186142683029175, 'sampling/importance_sampling_ratio/mean': 0.8399711847305298, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 3.7161264419555664, 'sampling/sampling_logp_difference/mean': 0.029673369601368904, 'step': 13, 'step_time': 10.544942977023311}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 91.0, 'completions/max_terminated_length': 91.0, 'completions/mean_length': 26.375, 'completions/mean_terminated_length': 26.375, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.053402797318995, 'epoch': 0.018229166666666668, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 979265.0, 'reward': 0.875, 'reward/selected_zero_std_group_fraction': 0.35714285714285715, 'reward/zero_std_group_fraction': 0.6785714285714286, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.109375, 'reward_components/parser_valid': 0.875, 'reward_components/reward': 0.875, 'reward_components/runs_correct': 1.25, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.875, 'reward_std': 0.3333333432674408, 'rewards/reward_func/mean': 0.875, 'rewards/reward_func/std': 0.3333333432674408, 'sampling/importance_sampling_ratio/max': 1.4895262718200684, 'sampling/importance_sampling_ratio/mean': 0.732711911201477, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.138972759246826, 'sampling/sampling_logp_difference/mean': 0.04473675414919853, 'step': 14, 'step_time': 14.720296632149257}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 321.0, 'completions/max_terminated_length': 321.0, 'completions/mean_length': 68.640625, 'completions/mean_terminated_length': 68.640625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.04984137351857498, 'epoch': 0.01953125, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.016947992146015167, 'learning_rate': 1e-05, 'loss': -0.0015179781476035714, 'num_tokens': 1041578.0, 'reward': 0.40625, 'reward/selected_zero_std_group_fraction': 0.38333333333333336, 'reward/zero_std_group_fraction': 0.6916666666666667, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.671875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.328125, 'reward_components/parser_valid': 0.671875, 'reward_components/reward': 0.40625, 'reward_components/runs_correct': 0.6875, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.53125, 'reward_std': 0.49501484632492065, 'rewards/reward_func/mean': 0.40625, 'rewards/reward_func/std': 0.49501484632492065, 'sampling/importance_sampling_ratio/max': 2.992492437362671, 'sampling/importance_sampling_ratio/mean': 0.7279292941093445, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 5.012690544128418, 'sampling/sampling_logp_difference/mean': 0.033740025013685226, 'step': 15, 'step_time': 15.719212398049422}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.015625, 'completions/max_length': 512.0, 'completions/max_terminated_length': 65.0, 'completions/mean_length': 27.78125, 'completions/mean_terminated_length': 20.095239639282227, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.04243861627765, 'epoch': 0.020833333333333332, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.004803791642189026, 'learning_rate': 1e-05, 'loss': -0.001964979339390993, 'num_tokens': 1112220.0, 'reward': 0.734375, 'reward/selected_zero_std_group_fraction': 0.40625, 'reward/zero_std_group_fraction': 0.703125, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.015625, 'reward_components/format_valid': 0.984375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.015625, 'reward_components/parser_valid': 0.984375, 'reward_components/reward': 0.734375, 'reward_components/runs_correct': 1.09375, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.796875, 'reward_std': 0.44515693187713623, 'rewards/reward_func/mean': 0.734375, 'rewards/reward_func/std': 0.44515693187713623, 'sampling/importance_sampling_ratio/max': 2.5751125812530518, 'sampling/importance_sampling_ratio/mean': 0.9604474306106567, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 1.8092589378356934, 'sampling/sampling_logp_difference/mean': 0.01625172793865204, 'step': 16, 'step_time': 19.88996219309047}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 192.0, 'completions/max_terminated_length': 192.0, 'completions/mean_length': 23.078125, 'completions/mean_terminated_length': 23.078125, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.0919711219612509, 'epoch': 0.022135416666666668, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.027110660448670387, 'learning_rate': 1e-05, 'loss': 0.003748212242498994, 'num_tokens': 1170017.0, 'reward': 0.6875, 'reward/selected_zero_std_group_fraction': 0.75, 'reward/zero_std_group_fraction': 0.875, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.9375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.0625, 'reward_components/parser_valid': 0.9375, 'reward_components/reward': 0.6875, 'reward_components/runs_correct': 0.9375, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.75, 'reward_std': 0.467176616191864, 'rewards/reward_func/mean': 0.6875, 'rewards/reward_func/std': 0.467176616191864, 'sampling/importance_sampling_ratio/max': 2.922673463821411, 'sampling/importance_sampling_ratio/mean': 0.858650803565979, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.066415309906006, 'sampling/sampling_logp_difference/mean': 0.038733866065740585, 'step': 17, 'step_time': 15.507753579877317}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 261.0, 'completions/max_terminated_length': 261.0, 'completions/mean_length': 34.765625, 'completions/mean_terminated_length': 34.765625, 'completions/min_length': 13.0, 'completions/min_terminated_length': 13.0, 'entropy': 0.09812993602827191, 'epoch': 0.0234375, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.0041731176897883415, 'learning_rate': 1e-05, 'loss': -0.001037793350405991, 'num_tokens': 1235090.0, 'reward': 0.734375, 'reward/selected_zero_std_group_fraction': 0.75, 'reward/zero_std_group_fraction': 0.875, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.984375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.015625, 'reward_components/parser_valid': 0.984375, 'reward_components/reward': 0.734375, 'reward_components/runs_correct': 1.234375, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.796875, 'reward_std': 0.44515693187713623, 'rewards/reward_func/mean': 0.734375, 'rewards/reward_func/std': 0.44515693187713623, 'sampling/importance_sampling_ratio/max': 1.9239144325256348, 'sampling/importance_sampling_ratio/mean': 0.8273533582687378, 'sampling/importance_sampling_ratio/min': 8.232260029217287e-08, 'sampling/sampling_logp_difference/max': 5.125619888305664, 'sampling/sampling_logp_difference/mean': 0.03699759766459465, 'step': 18, 'step_time': 14.965205831918865}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 51.0, 'completions/max_terminated_length': 51.0, 'completions/mean_length': 22.0625, 'completions/mean_terminated_length': 22.0625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.10557907959446311, 'epoch': 0.024739583333333332, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.02539324387907982, 'learning_rate': 1e-05, 'loss': 0.0018892987864091992, 'num_tokens': 1315542.0, 'reward': 0.3125, 'reward/selected_zero_std_group_fraction': 0.6666666666666666, 'reward/zero_std_group_fraction': 0.8333333333333334, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.734375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.265625, 'reward_components/parser_valid': 0.734375, 'reward_components/reward': 0.3125, 'reward_components/runs_correct': 0.5625, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.3671875, 'reward_std': 0.467176616191864, 'rewards/reward_func/mean': 0.3125, 'rewards/reward_func/std': 0.467176616191864, 'sampling/importance_sampling_ratio/max': 2.6890580654144287, 'sampling/importance_sampling_ratio/mean': 0.7533330321311951, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.2506103515625, 'sampling/sampling_logp_difference/mean': 0.06202193722128868, 'step': 19, 'step_time': 14.341453789034858}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 40.0, 'completions/max_terminated_length': 40.0, 'completions/mean_length': 23.515625, 'completions/mean_terminated_length': 23.515625, 'completions/min_length': 10.0, 'completions/min_terminated_length': 10.0, 'entropy': 0.06278747692704201, 'epoch': 0.026041666666666668, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.01859920658171177, 'learning_rate': 1e-05, 'loss': 0.0003009004285559058, 'num_tokens': 1382839.0, 'reward': 0.8125, 'reward/selected_zero_std_group_fraction': 0.625, 'reward/zero_std_group_fraction': 0.8125, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.859375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.140625, 'reward_components/parser_valid': 0.859375, 'reward_components/reward': 0.8125, 'reward_components/runs_correct': 1.296875, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.828125, 'reward_std': 0.39339789748191833, 'rewards/reward_func/mean': 0.8125, 'rewards/reward_func/std': 0.39339789748191833, 'sampling/importance_sampling_ratio/max': 1.8579050302505493, 'sampling/importance_sampling_ratio/mean': 0.7181844711303711, 'sampling/importance_sampling_ratio/min': 3.841196758003207e-06, 'sampling/sampling_logp_difference/max': 9.394405364990234, 'sampling/sampling_logp_difference/mean': 0.08545529097318649, 'step': 20, 'step_time': 12.08698793395888}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 50.0, 'completions/max_terminated_length': 50.0, 'completions/mean_length': 22.5, 'completions/mean_terminated_length': 22.5, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.13343330193310976, 'epoch': 0.02734375, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.04029390960931778, 'learning_rate': 1e-05, 'loss': -0.0021170061081647873, 'num_tokens': 1435031.0, 'reward': 0.71875, 'reward/selected_zero_std_group_fraction': 0.6, 'reward/zero_std_group_fraction': 0.8, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.875, 'reward_components/reward': 0.71875, 'reward_components/runs_correct': 0.96875, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.71875, 'reward_std': 0.4531635046005249, 'rewards/reward_func/mean': 0.71875, 'rewards/reward_func/std': 0.4531635046005249, 'sampling/importance_sampling_ratio/max': 2.6840994358062744, 'sampling/importance_sampling_ratio/mean': 0.9261176586151123, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 5.541268825531006, 'sampling/sampling_logp_difference/mean': 0.04675472527742386, 'step': 21, 'step_time': 10.022101087146439}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 52.0, 'completions/max_terminated_length': 52.0, 'completions/mean_length': 29.0, 'completions/mean_terminated_length': 29.0, 'completions/min_length': 22.0, 'completions/min_terminated_length': 22.0, 'entropy': 0.03837113315239549, 'epoch': 0.028645833333333332, 'frac_reward_zero_std': 0.75, 'grad_norm': 0.04053347557783127, 'learning_rate': 1e-05, 'loss': 0.0029122955165803432, 'num_tokens': 1504727.0, 'reward': 0.78125, 'reward/selected_zero_std_group_fraction': 0.5833333333333334, 'reward/zero_std_group_fraction': 0.7916666666666666, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.953125, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.046875, 'reward_components/parser_valid': 0.953125, 'reward_components/reward': 0.78125, 'reward_components/runs_correct': 1.671875, 'reward_components/runs_total': 2.0, 'reward_components/semantic_correct': 0.8359375, 'reward_std': 0.4166666865348816, 'rewards/reward_func/mean': 0.78125, 'rewards/reward_func/std': 0.4166666865348816, 'sampling/importance_sampling_ratio/max': 2.6758294105529785, 'sampling/importance_sampling_ratio/mean': 0.8408677577972412, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.335387229919434, 'sampling/sampling_logp_difference/mean': 0.0384678952395916, 'step': 22, 'step_time': 12.449200432864018}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 65.0, 'completions/max_terminated_length': 65.0, 'completions/mean_length': 22.578125, 'completions/mean_terminated_length': 22.578125, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.050773302325978875, 'epoch': 0.029947916666666668, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.002441965974867344, 'learning_rate': 1e-05, 'loss': -0.0010980369988828897, 'num_tokens': 1575420.0, 'reward': 0.359375, 'reward/selected_zero_std_group_fraction': 0.6071428571428571, 'reward/zero_std_group_fraction': 0.8035714285714286, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.609375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.390625, 'reward_components/parser_valid': 0.609375, 'reward_components/reward': 0.359375, 'reward_components/runs_correct': 0.609375, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.421875, 'reward_std': 0.4836103618144989, 'rewards/reward_func/mean': 0.359375, 'rewards/reward_func/std': 0.4836103618144989, 'sampling/importance_sampling_ratio/max': 1.8754615783691406, 'sampling/importance_sampling_ratio/mean': 0.8311659097671509, 'sampling/importance_sampling_ratio/min': 0.0001726742775645107, 'sampling/sampling_logp_difference/max': 8.907974243164062, 'sampling/sampling_logp_difference/mean': 0.055529478937387466, 'step': 23, 'step_time': 12.61808819801081}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 67.0, 'completions/max_terminated_length': 67.0, 'completions/mean_length': 26.15625, 'completions/mean_terminated_length': 26.15625, 'completions/min_length': 16.0, 'completions/min_terminated_length': 16.0, 'entropy': 0.05146293598227203, 'epoch': 0.03125, 'frac_reward_zero_std': 1.0, 'grad_norm': 0.0, 'learning_rate': 1e-05, 'loss': 0.0, 'num_tokens': 1644870.0, 'reward': 0.875, 'reward/selected_zero_std_group_fraction': 0.65625, 'reward/zero_std_group_fraction': 0.828125, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 1.0, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.0, 'reward_components/parser_valid': 1.0, 'reward_components/reward': 0.875, 'reward_components/runs_correct': 1.75, 'reward_components/runs_total': 1.875, 'reward_components/semantic_correct': 0.9375, 'reward_std': 0.3333333432674408, 'rewards/reward_func/mean': 0.875, 'rewards/reward_func/std': 0.3333333432674408, 'sampling/importance_sampling_ratio/max': 2.363029718399048, 'sampling/importance_sampling_ratio/mean': 0.8737799525260925, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 6.52822732925415, 'sampling/sampling_logp_difference/mean': 0.03789537772536278, 'step': 24, 'step_time': 12.627118773874827}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 90.0, 'completions/max_terminated_length': 90.0, 'completions/mean_length': 30.375, 'completions/mean_terminated_length': 30.375, 'completions/min_length': 13.0, 'completions/min_terminated_length': 13.0, 'entropy': 0.07523092487826943, 'epoch': 0.032552083333333336, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.15136529505252838, 'learning_rate': 1e-05, 'loss': 0.006091809831559658, 'num_tokens': 1725598.0, 'reward': 0.765625, 'reward/selected_zero_std_group_fraction': 0.6111111111111112, 'reward/zero_std_group_fraction': 0.8055555555555556, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.921875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.078125, 'reward_components/parser_valid': 0.921875, 'reward_components/reward': 0.765625, 'reward_components/runs_correct': 1.625, 'reward_components/runs_total': 1.875, 'reward_components/semantic_correct': 0.84375, 'reward_std': 0.42695629596710205, 'rewards/reward_func/mean': 0.765625, 'rewards/reward_func/std': 0.42695629596710205, 'sampling/importance_sampling_ratio/max': 2.438002347946167, 'sampling/importance_sampling_ratio/mean': 0.9185109734535217, 'sampling/importance_sampling_ratio/min': 0.07001316547393799, 'sampling/sampling_logp_difference/max': 1.8553194999694824, 'sampling/sampling_logp_difference/mean': 0.016449037939310074, 'step': 25, 'step_time': 14.915525149786845}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 57.0, 'completions/max_terminated_length': 57.0, 'completions/mean_length': 24.359375, 'completions/mean_terminated_length': 24.359375, 'completions/min_length': 9.0, 'completions/min_terminated_length': 9.0, 'entropy': 0.07893249252811074, 'epoch': 0.033854166666666664, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.002201978350058198, 'learning_rate': 1e-05, 'loss': -0.0010763078462332487, 'num_tokens': 1783605.0, 'reward': 0.734375, 'reward/selected_zero_std_group_fraction': 0.625, 'reward/zero_std_group_fraction': 0.8125, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.875, 'reward_components/reward': 0.734375, 'reward_components/runs_correct': 1.109375, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.7421875, 'reward_std': 0.44515693187713623, 'rewards/reward_func/mean': 0.734375, 'rewards/reward_func/std': 0.44515693187713623, 'sampling/importance_sampling_ratio/max': 2.744183301925659, 'sampling/importance_sampling_ratio/mean': 0.8056846857070923, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.071113586425781, 'sampling/sampling_logp_difference/mean': 0.05523296073079109, 'step': 26, 'step_time': 10.67610967787914}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.015625, 'completions/max_length': 512.0, 'completions/max_terminated_length': 59.0, 'completions/mean_length': 36.9375, 'completions/mean_terminated_length': 29.396827697753906, 'completions/min_length': 15.0, 'completions/min_terminated_length': 15.0, 'entropy': 0.03149152733385563, 'epoch': 0.03515625, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.002209650818258524, 'learning_rate': 1e-05, 'loss': -0.0009371075429953635, 'num_tokens': 1861617.0, 'reward': 0.734375, 'reward/selected_zero_std_group_fraction': 0.6363636363636364, 'reward/zero_std_group_fraction': 0.8181818181818182, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.015625, 'reward_components/format_valid': 0.734375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.25, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.75, 'reward_components/parser_unsafe': 0.265625, 'reward_components/parser_valid': 0.734375, 'reward_components/reward': 0.734375, 'reward_components/runs_correct': 1.21875, 'reward_components/runs_total': 1.5, 'reward_components/semantic_correct': 0.734375, 'reward_std': 0.44515693187713623, 'rewards/reward_func/mean': 0.734375, 'rewards/reward_func/std': 0.44515693187713623, 'sampling/importance_sampling_ratio/max': 2.5305018424987793, 'sampling/importance_sampling_ratio/mean': 0.821128785610199, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 2.271317481994629, 'sampling/sampling_logp_difference/mean': 0.03500381112098694, 'step': 27, 'step_time': 21.406569782993756}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 114.0, 'completions/max_terminated_length': 114.0, 'completions/mean_length': 26.90625, 'completions/mean_terminated_length': 26.90625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.07301372266374528, 'epoch': 0.036458333333333336, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.004998071119189262, 'learning_rate': 1e-05, 'loss': -0.0018165383953601122, 'num_tokens': 1930859.0, 'reward': 0.734375, 'reward/selected_zero_std_group_fraction': 0.6458333333333334, 'reward/zero_std_group_fraction': 0.8229166666666666, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.859375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.859375, 'reward_components/reward': 0.734375, 'reward_components/runs_correct': 1.34375, 'reward_components/runs_total': 1.625, 'reward_components/semantic_correct': 0.734375, 'reward_std': 0.44515693187713623, 'rewards/reward_func/mean': 0.734375, 'rewards/reward_func/std': 0.44515693187713623, 'sampling/importance_sampling_ratio/max': 1.928209900856018, 'sampling/importance_sampling_ratio/mean': 0.8505411148071289, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 1.9567553997039795, 'sampling/sampling_logp_difference/mean': 0.02324735000729561, 'step': 28, 'step_time': 13.582729734131135}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 110.0, 'completions/max_terminated_length': 110.0, 'completions/mean_length': 38.6875, 'completions/mean_terminated_length': 38.6875, 'completions/min_length': 9.0, 'completions/min_terminated_length': 9.0, 'entropy': 0.10580025054514408, 'epoch': 0.037760416666666664, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.02657872624695301, 'learning_rate': 1e-05, 'loss': -0.004455592017620802, 'num_tokens': 2007383.0, 'reward': 0.90625, 'reward/selected_zero_std_group_fraction': 0.6538461538461539, 'reward/zero_std_group_fraction': 0.8269230769230769, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.90625, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.09375, 'reward_components/parser_valid': 0.90625, 'reward_components/reward': 0.90625, 'reward_components/runs_correct': 1.1875, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.90625, 'reward_std': 0.29378482699394226, 'rewards/reward_func/mean': 0.90625, 'rewards/reward_func/std': 0.29378482699394226, 'sampling/importance_sampling_ratio/max': 2.0058493614196777, 'sampling/importance_sampling_ratio/mean': 0.7304644584655762, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 3.1370503902435303, 'sampling/sampling_logp_difference/mean': 0.054920513182878494, 'step': 29, 'step_time': 14.815608220058493}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 163.0, 'completions/max_terminated_length': 163.0, 'completions/mean_length': 38.015625, 'completions/mean_terminated_length': 38.015625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.038902290631085634, 'epoch': 0.0390625, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.016050608828663826, 'learning_rate': 1e-05, 'loss': -0.002265714341774583, 'num_tokens': 2087448.0, 'reward': 0.578125, 'reward/selected_zero_std_group_fraction': 0.6607142857142857, 'reward/zero_std_group_fraction': 0.8303571428571429, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.125, 'reward_components/parser_valid': 0.875, 'reward_components/reward': 0.578125, 'reward_components/runs_correct': 1.203125, 'reward_components/runs_total': 1.75, 'reward_components/semantic_correct': 0.703125, 'reward_std': 0.49776285886764526, 'rewards/reward_func/mean': 0.578125, 'rewards/reward_func/std': 0.49776285886764526, 'sampling/importance_sampling_ratio/max': 2.9719433784484863, 'sampling/importance_sampling_ratio/mean': 0.9424229860305786, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 7.308625221252441, 'sampling/sampling_logp_difference/mean': 0.049986038357019424, 'step': 30, 'step_time': 16.34779839811381}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 326.0, 'completions/max_terminated_length': 326.0, 'completions/mean_length': 27.90625, 'completions/mean_terminated_length': 27.90625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.0941302219289355, 'epoch': 0.040364583333333336, 'frac_reward_zero_std': 0.875, 'grad_norm': 0.005837010685354471, 'learning_rate': 1e-05, 'loss': -0.0005864137783646584, 'num_tokens': 2152466.0, 'reward': 0.984375, 'reward/selected_zero_std_group_fraction': 0.6666666666666666, 'reward/zero_std_group_fraction': 0.8333333333333334, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.984375, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.0, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 1.0, 'reward_components/parser_unsafe': 0.015625, 'reward_components/parser_valid': 0.984375, 'reward_components/reward': 0.984375, 'reward_components/runs_correct': 1.359375, 'reward_components/runs_total': 1.375, 'reward_components/semantic_correct': 0.984375, 'reward_std': 0.125, 'rewards/reward_func/mean': 0.984375, 'rewards/reward_func/std': 0.125, 'sampling/importance_sampling_ratio/max': 2.9497570991516113, 'sampling/importance_sampling_ratio/mean': 0.8253538608551025, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 15.174276351928711, 'sampling/sampling_logp_difference/mean': 0.0585869625210762, 'step': 31, 'step_time': 16.259609793080017}, {'clip_ratio/high_max': 0.0, 'clip_ratio/high_mean': 0.0, 'clip_ratio/low_mean': 0.0, 'clip_ratio/low_min': 0.0, 'clip_ratio/region_mean': 0.0, 'completions/clipped_ratio': 0.0, 'completions/max_length': 147.0, 'completions/max_terminated_length': 147.0, 'completions/mean_length': 33.15625, 'completions/mean_terminated_length': 33.15625, 'completions/min_length': 8.0, 'completions/min_terminated_length': 8.0, 'entropy': 0.08133227378129959, 'epoch': 0.041666666666666664, 'frac_reward_zero_std': 0.625, 'grad_norm': 0.016582153737545013, 'learning_rate': 1e-05, 'loss': -0.0032648490741848946, 'num_tokens': 2210780.0, 'reward': 0.734375, 'reward/selected_zero_std_group_fraction': 0.640625, 'reward/zero_std_group_fraction': 0.8203125, 'reward_components/channel_close_count': 0.0, 'reward_components/channel_open_count': 0.0, 'reward_components/completion_truncated': 0.0, 'reward_components/format_valid': 0.96875, 'reward_components/native_boundary_valid': 1.0, 'reward_components/parser_json': 0.125, 'reward_components/parser_labelled_records': 0.0, 'reward_components/parser_natural': 0.875, 'reward_components/parser_unsafe': 0.015625, 'reward_components/parser_valid': 0.96875, 'reward_components/reward': 0.734375, 'reward_components/runs_correct': 0.984375, 'reward_components/runs_total': 1.25, 'reward_components/semantic_correct': 0.7890625, 'reward_std': 0.44515693187713623, 'rewards/reward_func/mean': 0.734375, 'rewards/reward_func/std': 0.44515693187713623, 'sampling/importance_sampling_ratio/max': 2.6860883235931396, 'sampling/importance_sampling_ratio/mean': 0.8768871426582336, 'sampling/importance_sampling_ratio/min': 0.0, 'sampling/sampling_logp_difference/max': 4.563495635986328, 'sampling/sampling_logp_difference/mean': 0.03013586811721325, 'step': 32, 'step_time': 12.216428584884852}], best_metric=None, best_global_step=None, best_model_checkpoint=None, is_local_process_zero=True, is_world_process_zero=True, is_hyper_param_search=False, trial_name=None, trial_params=None, stateful_callbacks={'TrainerControl': {'args': {'should_epoch_stop': False, 'should_evaluate': False, 'should_log': False, 'should_save': True, 'should_training_stop': True}, 'attributes': {}}})",
|
| 17 |
+
"note": "completion_ids removed; regenerate by tokenising completion_raw_text with the gemma4-26b-a4b tokenizer (verified exact on 221 records). log_extra/log_metric were bound-method reprs. Streamed rather than downloaded because the Hub's declared size for some files disagrees with their stored bytes."
|
| 18 |
+
}
|