Buckets:

Mercity
/

Desicion-Model-Custom

Files

xet

Mercity/Desicion-Model-Custom / run2_v2 /logs /v2_run.log

Pranav2748

13 days ago

download

raw

15.2 kB

{"time": "2026-10-31T13:31:52Z", "event": "start", "stage": "merge_teacher"}
merged 662 teacher records
{"time": "2026-10-31T13:31:53Z", "event": "done", "stage": "merge_teacher"}
sweep 1500; dev_sweep 500; ablation 12000 (7333 with chains)
{"time": "2026-10-31T13:32:09Z", "event": "start", "stage": "speed_bench"}
{"timed_examples": 251, "timed_tokens": 260032, "timed_reasoner": 67, "warm_examples": 56, "check_microbatch_tokens": [6918, 6992, 8249]}
{"config": "R", "check_losses": [7.028905391693115, 0.5920570492744446, 0.176888570189476], "equivalent": false}
{"config": "A", "liger": false, "micro": 16384, "ckpt_above": 0, "tok_per_s": 3023.2289978067524, "seconds": 86.01134753227234, "tokens": 260032, "peak_gb": 18.874352128, "check_losses": [6.984011173248291, 0.5924540162086487, 0.17689478397369385], "grad_cosine_vs_fp32": 0.9478817496018348, "grad_err_vs_fp32": 0.33129514233342916, "loss_err_vs_fp32": 0.006387085320266368, "equivalent": true}
{"config": "B", "liger": true, "micro": 16384, "ckpt_above": 0, "tok_per_s": 3368.338140175667, "seconds": 77.19890022277832, "tokens": 260032, "peak_gb": 18.7914624, "check_losses": [6.92450475692749, 0.5925873517990112, 0.17689892649650574], "grad_cosine_vs_fp32": 0.9590832216505911, "grad_err_vs_fp32": 0.2865537126507501, "loss_err_vs_fp32": 0.014853043105261811, "equivalent": false}
{"config": "C", "liger": false, "micro": 8192, "ckpt_above": 8192, "tok_per_s": 4474.20676314094, "seconds": 58.118011474609375, "tokens": 260032, "peak_gb": 74.92635904, "check_losses": [6.984011173248291, 0.5924540162086487, 0.17689478397369385], "grad_cosine_vs_fp32": 0.9471831037158528, "grad_err_vs_fp32": 0.33392519331678217, "loss_err_vs_fp32": 0.006387085320266368, "equivalent": true}
{"config": "D", "liger": true, "micro": 8192, "ckpt_above": 8192, "tok_per_s": 5204.978060833649, "seconds": 49.958327770233154, "tokens": 260032, "peak_gb": 66.781046784, "check_losses": [6.92450475692749, 0.5925873517990112, 0.17689892649650574], "grad_cosine_vs_fp32": 0.9584863619082864, "grad_err_vs_fp32": 0.28883762488781384, "loss_err_vs_fp32": 0.014853043105261811, "equivalent": false}
CHOSEN {"chosen": "A", "flags": "", "speedup_vs_A": 1.0}
{"time": "2026-10-31T13:45:15Z", "event": "done", "stage": "speed_bench"}
{"time": "2026-10-31T13:45:46Z", "event": "chosen", "stage": "speed=default"}
{"time": "2026-10-31T13:45:46Z", "event": "start", "stage": "sweep_1e-4"}
{"examples": 1500, "tokens": 753558, "readout_from_votes": 259, "readout_eval_order": 720, "readout_gold_only": 1241}

Loading weights:   0%|          | 0/426 [00:00<?, ?it/s]
Loading weights: 100%|██████████| 426/426 [00:00<00:00, 4807.25it/s]
trainable params: 60,948,480 || all params: 4,266,699,776 || trainable%: 1.4285
[transformers] `causal_conv1d_fn` is falling back to its reference PyTorch implementation because `causal_conv1d` is not installed. This is correct but much slower; install `causal_conv1d` for the optimized kernel.
/workspace/decision-reasoner/runs/v2/train.py:215: UserWarning: Converting a tensor with requires_grad=True to a scalar may lead to unexpected behavior.
Consider using tensor.detach() first. (Triggered internally at /__w/pytorch/pytorch/torch/csrc/autograd/generated/python_variable_methods.cpp:822.)
  stats["readout_loss"] += float(loss)
{"step": 1, "of": 17, "lr": 0.0001, "grad_norm": 9.24509048461914, "examples": 89, "readout_examples": 89, "reasoner_examples": 0, "step_tokens": 46749, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3659.591147858746, "elapsed_min": 0.21290629704793293, "eta_min": 3.406500752766927, "chain_loss": null, "readout_loss": 0.9717099721619684, "readout_argmax_matches_target": 0.651685393258427, "readout_mean_top_prob": 0.6780225949340992, "readout_target_entropy": 0.08248412915442684, "letter_loss": null, "letter_top_prob": null, "mem_gb": 16.935811584}
{"step": 2, "of": 17, "lr": 9.903926402016153e-05, "grad_norm": 14.038253784179688, "examples": 138, "readout_examples": 138, "reasoner_examples": 0, "step_tokens": 47085, "chain_tokens": 0, "micro_batches": 4, "tok_per_s": 3118.749342470187, "elapsed_min": 0.5014510075251262, "eta_min": 3.760882556438446, "chain_loss": null, "readout_loss": 0.8416770423839584, "readout_argmax_matches_target": 0.6304347826086957, "readout_mean_top_prob": 0.8406794619733009, "readout_target_entropy": 0.04357187665890956, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
{"step": 3, "of": 17, "lr": 9.619397662556435e-05, "grad_norm": 6.111450672149658, "examples": 90, "readout_examples": 90, "reasoner_examples": 0, "step_tokens": 42718, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3331.2193085583876, "elapsed_min": 0.6831932862599691, "eta_min": 3.188235335879855, "chain_loss": null, "readout_loss": 1.2067683474057251, "readout_argmax_matches_target": 0.43333333333333335, "readout_mean_top_prob": 0.5096879975663291, "readout_target_entropy": 0.13333093267348078, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
{"step": 10, "of": 17, "lr": 4.0245483899193595e-05, "grad_norm": 4.178615093231201, "examples": 92, "readout_examples": 92, "reasoner_examples": 0, "step_tokens": 45015, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3433.4376575278447, "elapsed_min": 2.1637643496195476, "eta_min": 1.5146350447336832, "chain_loss": null, "readout_loss": 0.924819931914301, "readout_argmax_matches_target": 0.6304347826086957, "readout_mean_top_prob": 0.5786254320131696, "readout_target_entropy": 0.06314167380332947, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
{"dev_n": 500, "dev_acc": 0.614, "dev_nll": 0.8802512453980744, "dev_ece15": 0.09050485435128212}
{"dev_n": 500, "dev_acc": 0.614, "dev_nll": 0.8802512453980744, "dev_ece15": 0.09050485435128212, "v2_options": {"readout_pool": "all", "subset": "data/sweep_ids.txt", "min_correct": 2, "second_chain_min": 3, "rep_max": 20, "post_soft": true, "eval_order_frac": 0.5, "drop_recall": true, "strip_cw": true}, "speed_options": {"liger": false, "ckpt_above": 0, "micro_tokens": 16384, "no_probe": true}, "mode": "readout_only", "alpha": 0.5, "lr": 0.0001, "rank": 32, "lora_alpha": 64, "seed": 17, "examples": 1500, "steps_done": 17, "steps_planned": 17, "stopped_at_deadline": false, "train_tokens": 753558, "train_seconds": 213.45540022850037, "tok_per_s": 3530.283137336085, "counts": {"readout_from_votes": 259, "readout_eval_order": 720, "readout_gold_only": 1241}, "reasoner_examples": 0, "readout_examples": 1500}
{"time": "2026-10-31T13:50:11Z", "event": "done", "stage": "sweep_1e-4"}
{"time": "2026-10-31T13:50:39Z", "event": "start", "stage": "sweep_2e-4"}
{"examples": 1500, "tokens": 753558, "readout_from_votes": 259, "readout_eval_order": 720, "readout_gold_only": 1241}

Loading weights:   0%|          | 0/426 [00:00<?, ?it/s]
Loading weights: 100%|██████████| 426/426 [00:00<00:00, 5435.00it/s]
trainable params: 60,948,480 || all params: 4,266,699,776 || trainable%: 1.4285
[transformers] `causal_conv1d_fn` is falling back to its reference PyTorch implementation because `causal_conv1d` is not installed. This is correct but much slower; install `causal_conv1d` for the optimized kernel.
/workspace/decision-reasoner/runs/v2/train.py:215: UserWarning: Converting a tensor with requires_grad=True to a scalar may lead to unexpected behavior.
Consider using tensor.detach() first. (Triggered internally at /__w/pytorch/pytorch/torch/csrc/autograd/generated/python_variable_methods.cpp:822.)
  stats["readout_loss"] += float(loss)
{"step": 1, "of": 17, "lr": 0.0002, "grad_norm": 9.24509048461914, "examples": 89, "readout_examples": 89, "reasoner_examples": 0, "step_tokens": 46749, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3395.473544454682, "elapsed_min": 0.22946725686391195, "eta_min": 3.671476109822591, "chain_loss": null, "readout_loss": 0.9717099721619684, "readout_argmax_matches_target": 0.651685393258427, "readout_mean_top_prob": 0.6780225949340992, "readout_target_entropy": 0.08248412915442684, "letter_loss": null, "letter_top_prob": null, "mem_gb": 16.935811584}
{"step": 2, "of": 17, "lr": 0.00019807852804032305, "grad_norm": 14.038683891296387, "examples": 138, "readout_examples": 138, "reasoner_examples": 0, "step_tokens": 47085, "chain_tokens": 0, "micro_batches": 4, "tok_per_s": 3369.4372556560056, "elapsed_min": 0.46414278745651244, "eta_min": 3.4810709059238434, "chain_loss": null, "readout_loss": 0.8416770423839584, "readout_argmax_matches_target": 0.6304347826086957, "readout_mean_top_prob": 0.8406794619733009, "readout_target_entropy": 0.04357187665890956, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
{"step": 3, "of": 17, "lr": 0.0001923879532511287, "grad_norm": 7.012028694152832, "examples": 90, "readout_examples": 90, "reasoner_examples": 0, "step_tokens": 42718, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3520.8647460000952, "elapsed_min": 0.6463942329088846, "eta_min": 3.016506420241462, "chain_loss": null, "readout_loss": 1.3452464786668619, "readout_argmax_matches_target": 0.3111111111111111, "readout_mean_top_prob": 0.5647904626197285, "readout_target_entropy": 0.13333093267348078, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
{"step": 10, "of": 17, "lr": 8.049096779838719e-05, "grad_norm": 3.333402156829834, "examples": 92, "readout_examples": 92, "reasoner_examples": 0, "step_tokens": 45015, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3691.6241911520146, "elapsed_min": 2.012433989842733, "eta_min": 1.4087037928899129, "chain_loss": null, "readout_loss": 0.8976337474811337, "readout_argmax_matches_target": 0.6304347826086957, "readout_mean_top_prob": 0.5487607485898163, "readout_target_entropy": 0.06314167380332947, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
{"dev_n": 500, "dev_acc": 0.65, "dev_nll": 0.8279993297066539, "dev_ece15": 0.05523754885792731}
{"dev_n": 500, "dev_acc": 0.65, "dev_nll": 0.8279993297066539, "dev_ece15": 0.05523754885792731, "v2_options": {"readout_pool": "all", "subset": "data/sweep_ids.txt", "min_correct": 2, "second_chain_min": 3, "rep_max": 20, "post_soft": true, "eval_order_frac": 0.5, "drop_recall": true, "strip_cw": true}, "speed_options": {"liger": false, "ckpt_above": 0, "micro_tokens": 16384, "no_probe": true}, "mode": "readout_only", "alpha": 0.5, "lr": 0.0002, "rank": 32, "lora_alpha": 64, "seed": 17, "examples": 1500, "steps_done": 17, "steps_planned": 17, "stopped_at_deadline": false, "train_tokens": 753558, "train_seconds": 199.42422223091125, "tok_per_s": 3778.668366210113, "counts": {"readout_from_votes": 259, "readout_eval_order": 720, "readout_gold_only": 1241}, "reasoner_examples": 0, "readout_examples": 1500}
{"time": "2026-10-31T13:54:50Z", "event": "done", "stage": "sweep_2e-4"}
{"time": "2026-10-31T13:55:07Z", "event": "start", "stage": "sweep_3e-4"}
{"examples": 1500, "tokens": 753558, "readout_from_votes": 259, "readout_eval_order": 720, "readout_gold_only": 1241}

Loading weights:   0%|          | 0/426 [00:00<?, ?it/s]
Loading weights:  89%|████████▉ | 379/426 [00:00<00:00, 3762.81it/s]
Loading weights: 100%|██████████| 426/426 [00:00<00:00, 3840.21it/s]
trainable params: 60,948,480 || all params: 4,266,699,776 || trainable%: 1.4285
[transformers] `causal_conv1d_fn` is falling back to its reference PyTorch implementation because `causal_conv1d` is not installed. This is correct but much slower; install `causal_conv1d` for the optimized kernel.
/workspace/decision-reasoner/runs/v2/train.py:215: UserWarning: Converting a tensor with requires_grad=True to a scalar may lead to unexpected behavior.
Consider using tensor.detach() first. (Triggered internally at /__w/pytorch/pytorch/torch/csrc/autograd/generated/python_variable_methods.cpp:822.)
  stats["readout_loss"] += float(loss)
{"step": 1, "of": 17, "lr": 0.0003, "grad_norm": 9.24509048461914, "examples": 89, "readout_examples": 89, "reasoner_examples": 0, "step_tokens": 46749, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3110.0486094750986, "elapsed_min": 0.2505266308784485, "eta_min": 4.008426094055176, "chain_loss": null, "readout_loss": 0.9717099721619684, "readout_argmax_matches_target": 0.651685393258427, "readout_mean_top_prob": 0.6780225949340992, "readout_target_entropy": 0.08248412915442684, "letter_loss": null, "letter_top_prob": null, "mem_gb": 16.935811584}
{"step": 2, "of": 17, "lr": 0.00029711779206048454, "grad_norm": 14.03878402709961, "examples": 138, "readout_examples": 138, "reasoner_examples": 0, "step_tokens": 47085, "chain_tokens": 0, "micro_batches": 4, "tok_per_s": 3197.5437632287594, "elapsed_min": 0.4890941659609477, "eta_min": 3.6682062447071075, "chain_loss": null, "readout_loss": 0.8416770423839584, "readout_argmax_matches_target": 0.6304347826086957, "readout_mean_top_prob": 0.8406794619733009, "readout_target_entropy": 0.04357187665890956, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
{"step": 3, "of": 17, "lr": 0.000288581929876693, "grad_norm": 8.426118850708008, "examples": 90, "readout_examples": 90, "reasoner_examples": 0, "step_tokens": 42718, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3391.384143001221, "elapsed_min": 0.6710730989774069, "eta_min": 3.1316744618945647, "chain_loss": null, "readout_loss": 1.5099286515679624, "readout_argmax_matches_target": 0.2777777777777778, "readout_mean_top_prob": 0.6285712169276343, "readout_target_entropy": 0.13333093267348078, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
{"step": 10, "of": 17, "lr": 0.00012073645169758076, "grad_norm": 5.670900821685791, "examples": 92, "readout_examples": 92, "reasoner_examples": 0, "step_tokens": 45015, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3662.2643625865917, "elapsed_min": 2.028567373752594, "eta_min": 1.4199971616268159, "chain_loss": null, "readout_loss": 0.9605440597819246, "readout_argmax_matches_target": 0.5543478260869565, "readout_mean_top_prob": 0.5456273620543273, "readout_target_entropy": 0.06314167380332947, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
{"dev_n": 500, "dev_acc": 0.638, "dev_nll": 0.8704747354204301, "dev_ece15": 0.055960430085659034}
{"dev_n": 500, "dev_acc": 0.638, "dev_nll": 0.8704747354204301, "dev_ece15": 0.055960430085659034, "v2_options": {"readout_pool": "all", "subset": "data/sweep_ids.txt", "min_correct": 2, "second_chain_min": 3, "rep_max": 20, "post_soft": true, "eval_order_frac": 0.5, "drop_recall": true, "strip_cw": true}, "speed_options": {"liger": false, "ckpt_above": 0, "micro_tokens": 16384, "no_probe": true}, "mode": "readout_only", "alpha": 0.5, "lr": 0.0003, "rank": 32, "lora_alpha": 64, "seed": 17, "examples": 1500, "steps_done": 17, "steps_planned": 17, "stopped_at_deadline": false, "train_tokens": 753558, "train_seconds": 200.36477303504944, "tok_per_s": 3760.9305697073883, "counts": {"readout_from_votes": 259, "readout_eval_order": 720, "readout_gold_only": 1241}, "reasoner_examples": 0, "readout_examples": 1500}
{"time": "2026-10-31T13:59:18Z", "event": "done", "stage": "sweep_3e-4"}
RUN_V2_EXIT 143

Xet Storage Details

Size:: 15.2 kB
Xet hash:: baa005d344ac06855a44c9ee4e9a100c17259da2a1e5c31efb798df1842e8174

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

	{"time": "2026-10-31T13:31:52Z", "event": "start", "stage": "merge_teacher"}
	merged 662 teacher records
	{"time": "2026-10-31T13:31:53Z", "event": "done", "stage": "merge_teacher"}
	sweep 1500; dev_sweep 500; ablation 12000 (7333 with chains)
	{"time": "2026-10-31T13:32:09Z", "event": "start", "stage": "speed_bench"}
	{"timed_examples": 251, "timed_tokens": 260032, "timed_reasoner": 67, "warm_examples": 56, "check_microbatch_tokens": [6918, 6992, 8249]}
	{"config": "R", "check_losses": [7.028905391693115, 0.5920570492744446, 0.176888570189476], "equivalent": false}
	{"config": "A", "liger": false, "micro": 16384, "ckpt_above": 0, "tok_per_s": 3023.2289978067524, "seconds": 86.01134753227234, "tokens": 260032, "peak_gb": 18.874352128, "check_losses": [6.984011173248291, 0.5924540162086487, 0.17689478397369385], "grad_cosine_vs_fp32": 0.9478817496018348, "grad_err_vs_fp32": 0.33129514233342916, "loss_err_vs_fp32": 0.006387085320266368, "equivalent": true}
	{"config": "B", "liger": true, "micro": 16384, "ckpt_above": 0, "tok_per_s": 3368.338140175667, "seconds": 77.19890022277832, "tokens": 260032, "peak_gb": 18.7914624, "check_losses": [6.92450475692749, 0.5925873517990112, 0.17689892649650574], "grad_cosine_vs_fp32": 0.9590832216505911, "grad_err_vs_fp32": 0.2865537126507501, "loss_err_vs_fp32": 0.014853043105261811, "equivalent": false}
	{"config": "C", "liger": false, "micro": 8192, "ckpt_above": 8192, "tok_per_s": 4474.20676314094, "seconds": 58.118011474609375, "tokens": 260032, "peak_gb": 74.92635904, "check_losses": [6.984011173248291, 0.5924540162086487, 0.17689478397369385], "grad_cosine_vs_fp32": 0.9471831037158528, "grad_err_vs_fp32": 0.33392519331678217, "loss_err_vs_fp32": 0.006387085320266368, "equivalent": true}
	{"config": "D", "liger": true, "micro": 8192, "ckpt_above": 8192, "tok_per_s": 5204.978060833649, "seconds": 49.958327770233154, "tokens": 260032, "peak_gb": 66.781046784, "check_losses": [6.92450475692749, 0.5925873517990112, 0.17689892649650574], "grad_cosine_vs_fp32": 0.9584863619082864, "grad_err_vs_fp32": 0.28883762488781384, "loss_err_vs_fp32": 0.014853043105261811, "equivalent": false}
	CHOSEN {"chosen": "A", "flags": "", "speedup_vs_A": 1.0}
	{"time": "2026-10-31T13:45:15Z", "event": "done", "stage": "speed_bench"}
	{"time": "2026-10-31T13:45:46Z", "event": "chosen", "stage": "speed=default"}
	{"time": "2026-10-31T13:45:46Z", "event": "start", "stage": "sweep_1e-4"}
	{"examples": 1500, "tokens": 753558, "readout_from_votes": 259, "readout_eval_order": 720, "readout_gold_only": 1241}
	Loading weights: 0%\| \| 0/426 [00:00<?, ?it/s] Loading weights: 100%\|██████████\| 426/426 [00:00<00:00, 4807.25it/s]
	trainable params: 60,948,480 \|\| all params: 4,266,699,776 \|\| trainable%: 1.4285
	[transformers] `causal_conv1d_fn` is falling back to its reference PyTorch implementation because `causal_conv1d` is not installed. This is correct but much slower; install `causal_conv1d` for the optimized kernel.
	/workspace/decision-reasoner/runs/v2/train.py:215: UserWarning: Converting a tensor with requires_grad=True to a scalar may lead to unexpected behavior.
	Consider using tensor.detach() first. (Triggered internally at /__w/pytorch/pytorch/torch/csrc/autograd/generated/python_variable_methods.cpp:822.)
	stats["readout_loss"] += float(loss)
	{"step": 1, "of": 17, "lr": 0.0001, "grad_norm": 9.24509048461914, "examples": 89, "readout_examples": 89, "reasoner_examples": 0, "step_tokens": 46749, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3659.591147858746, "elapsed_min": 0.21290629704793293, "eta_min": 3.406500752766927, "chain_loss": null, "readout_loss": 0.9717099721619684, "readout_argmax_matches_target": 0.651685393258427, "readout_mean_top_prob": 0.6780225949340992, "readout_target_entropy": 0.08248412915442684, "letter_loss": null, "letter_top_prob": null, "mem_gb": 16.935811584}
	{"step": 2, "of": 17, "lr": 9.903926402016153e-05, "grad_norm": 14.038253784179688, "examples": 138, "readout_examples": 138, "reasoner_examples": 0, "step_tokens": 47085, "chain_tokens": 0, "micro_batches": 4, "tok_per_s": 3118.749342470187, "elapsed_min": 0.5014510075251262, "eta_min": 3.760882556438446, "chain_loss": null, "readout_loss": 0.8416770423839584, "readout_argmax_matches_target": 0.6304347826086957, "readout_mean_top_prob": 0.8406794619733009, "readout_target_entropy": 0.04357187665890956, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
	{"step": 3, "of": 17, "lr": 9.619397662556435e-05, "grad_norm": 6.111450672149658, "examples": 90, "readout_examples": 90, "reasoner_examples": 0, "step_tokens": 42718, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3331.2193085583876, "elapsed_min": 0.6831932862599691, "eta_min": 3.188235335879855, "chain_loss": null, "readout_loss": 1.2067683474057251, "readout_argmax_matches_target": 0.43333333333333335, "readout_mean_top_prob": 0.5096879975663291, "readout_target_entropy": 0.13333093267348078, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
	{"step": 10, "of": 17, "lr": 4.0245483899193595e-05, "grad_norm": 4.178615093231201, "examples": 92, "readout_examples": 92, "reasoner_examples": 0, "step_tokens": 45015, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3433.4376575278447, "elapsed_min": 2.1637643496195476, "eta_min": 1.5146350447336832, "chain_loss": null, "readout_loss": 0.924819931914301, "readout_argmax_matches_target": 0.6304347826086957, "readout_mean_top_prob": 0.5786254320131696, "readout_target_entropy": 0.06314167380332947, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
	{"dev_n": 500, "dev_acc": 0.614, "dev_nll": 0.8802512453980744, "dev_ece15": 0.09050485435128212}
	{"dev_n": 500, "dev_acc": 0.614, "dev_nll": 0.8802512453980744, "dev_ece15": 0.09050485435128212, "v2_options": {"readout_pool": "all", "subset": "data/sweep_ids.txt", "min_correct": 2, "second_chain_min": 3, "rep_max": 20, "post_soft": true, "eval_order_frac": 0.5, "drop_recall": true, "strip_cw": true}, "speed_options": {"liger": false, "ckpt_above": 0, "micro_tokens": 16384, "no_probe": true}, "mode": "readout_only", "alpha": 0.5, "lr": 0.0001, "rank": 32, "lora_alpha": 64, "seed": 17, "examples": 1500, "steps_done": 17, "steps_planned": 17, "stopped_at_deadline": false, "train_tokens": 753558, "train_seconds": 213.45540022850037, "tok_per_s": 3530.283137336085, "counts": {"readout_from_votes": 259, "readout_eval_order": 720, "readout_gold_only": 1241}, "reasoner_examples": 0, "readout_examples": 1500}
	{"time": "2026-10-31T13:50:11Z", "event": "done", "stage": "sweep_1e-4"}
	{"time": "2026-10-31T13:50:39Z", "event": "start", "stage": "sweep_2e-4"}
	{"examples": 1500, "tokens": 753558, "readout_from_votes": 259, "readout_eval_order": 720, "readout_gold_only": 1241}
	Loading weights: 0%\| \| 0/426 [00:00<?, ?it/s] Loading weights: 100%\|██████████\| 426/426 [00:00<00:00, 5435.00it/s]
	trainable params: 60,948,480 \|\| all params: 4,266,699,776 \|\| trainable%: 1.4285
	[transformers] `causal_conv1d_fn` is falling back to its reference PyTorch implementation because `causal_conv1d` is not installed. This is correct but much slower; install `causal_conv1d` for the optimized kernel.
	/workspace/decision-reasoner/runs/v2/train.py:215: UserWarning: Converting a tensor with requires_grad=True to a scalar may lead to unexpected behavior.
	Consider using tensor.detach() first. (Triggered internally at /__w/pytorch/pytorch/torch/csrc/autograd/generated/python_variable_methods.cpp:822.)
	stats["readout_loss"] += float(loss)
	{"step": 1, "of": 17, "lr": 0.0002, "grad_norm": 9.24509048461914, "examples": 89, "readout_examples": 89, "reasoner_examples": 0, "step_tokens": 46749, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3395.473544454682, "elapsed_min": 0.22946725686391195, "eta_min": 3.671476109822591, "chain_loss": null, "readout_loss": 0.9717099721619684, "readout_argmax_matches_target": 0.651685393258427, "readout_mean_top_prob": 0.6780225949340992, "readout_target_entropy": 0.08248412915442684, "letter_loss": null, "letter_top_prob": null, "mem_gb": 16.935811584}
	{"step": 2, "of": 17, "lr": 0.00019807852804032305, "grad_norm": 14.038683891296387, "examples": 138, "readout_examples": 138, "reasoner_examples": 0, "step_tokens": 47085, "chain_tokens": 0, "micro_batches": 4, "tok_per_s": 3369.4372556560056, "elapsed_min": 0.46414278745651244, "eta_min": 3.4810709059238434, "chain_loss": null, "readout_loss": 0.8416770423839584, "readout_argmax_matches_target": 0.6304347826086957, "readout_mean_top_prob": 0.8406794619733009, "readout_target_entropy": 0.04357187665890956, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
	{"step": 3, "of": 17, "lr": 0.0001923879532511287, "grad_norm": 7.012028694152832, "examples": 90, "readout_examples": 90, "reasoner_examples": 0, "step_tokens": 42718, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3520.8647460000952, "elapsed_min": 0.6463942329088846, "eta_min": 3.016506420241462, "chain_loss": null, "readout_loss": 1.3452464786668619, "readout_argmax_matches_target": 0.3111111111111111, "readout_mean_top_prob": 0.5647904626197285, "readout_target_entropy": 0.13333093267348078, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
	{"step": 10, "of": 17, "lr": 8.049096779838719e-05, "grad_norm": 3.333402156829834, "examples": 92, "readout_examples": 92, "reasoner_examples": 0, "step_tokens": 45015, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3691.6241911520146, "elapsed_min": 2.012433989842733, "eta_min": 1.4087037928899129, "chain_loss": null, "readout_loss": 0.8976337474811337, "readout_argmax_matches_target": 0.6304347826086957, "readout_mean_top_prob": 0.5487607485898163, "readout_target_entropy": 0.06314167380332947, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
	{"dev_n": 500, "dev_acc": 0.65, "dev_nll": 0.8279993297066539, "dev_ece15": 0.05523754885792731}
	{"dev_n": 500, "dev_acc": 0.65, "dev_nll": 0.8279993297066539, "dev_ece15": 0.05523754885792731, "v2_options": {"readout_pool": "all", "subset": "data/sweep_ids.txt", "min_correct": 2, "second_chain_min": 3, "rep_max": 20, "post_soft": true, "eval_order_frac": 0.5, "drop_recall": true, "strip_cw": true}, "speed_options": {"liger": false, "ckpt_above": 0, "micro_tokens": 16384, "no_probe": true}, "mode": "readout_only", "alpha": 0.5, "lr": 0.0002, "rank": 32, "lora_alpha": 64, "seed": 17, "examples": 1500, "steps_done": 17, "steps_planned": 17, "stopped_at_deadline": false, "train_tokens": 753558, "train_seconds": 199.42422223091125, "tok_per_s": 3778.668366210113, "counts": {"readout_from_votes": 259, "readout_eval_order": 720, "readout_gold_only": 1241}, "reasoner_examples": 0, "readout_examples": 1500}
	{"time": "2026-10-31T13:54:50Z", "event": "done", "stage": "sweep_2e-4"}
	{"time": "2026-10-31T13:55:07Z", "event": "start", "stage": "sweep_3e-4"}
	{"examples": 1500, "tokens": 753558, "readout_from_votes": 259, "readout_eval_order": 720, "readout_gold_only": 1241}
	Loading weights: 0%\| \| 0/426 [00:00<?, ?it/s] Loading weights: 89%\|████████▉ \| 379/426 [00:00<00:00, 3762.81it/s] Loading weights: 100%\|██████████\| 426/426 [00:00<00:00, 3840.21it/s]
	trainable params: 60,948,480 \|\| all params: 4,266,699,776 \|\| trainable%: 1.4285
	[transformers] `causal_conv1d_fn` is falling back to its reference PyTorch implementation because `causal_conv1d` is not installed. This is correct but much slower; install `causal_conv1d` for the optimized kernel.
	/workspace/decision-reasoner/runs/v2/train.py:215: UserWarning: Converting a tensor with requires_grad=True to a scalar may lead to unexpected behavior.
	Consider using tensor.detach() first. (Triggered internally at /__w/pytorch/pytorch/torch/csrc/autograd/generated/python_variable_methods.cpp:822.)
	stats["readout_loss"] += float(loss)
	{"step": 1, "of": 17, "lr": 0.0003, "grad_norm": 9.24509048461914, "examples": 89, "readout_examples": 89, "reasoner_examples": 0, "step_tokens": 46749, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3110.0486094750986, "elapsed_min": 0.2505266308784485, "eta_min": 4.008426094055176, "chain_loss": null, "readout_loss": 0.9717099721619684, "readout_argmax_matches_target": 0.651685393258427, "readout_mean_top_prob": 0.6780225949340992, "readout_target_entropy": 0.08248412915442684, "letter_loss": null, "letter_top_prob": null, "mem_gb": 16.935811584}
	{"step": 2, "of": 17, "lr": 0.00029711779206048454, "grad_norm": 14.03878402709961, "examples": 138, "readout_examples": 138, "reasoner_examples": 0, "step_tokens": 47085, "chain_tokens": 0, "micro_batches": 4, "tok_per_s": 3197.5437632287594, "elapsed_min": 0.4890941659609477, "eta_min": 3.6682062447071075, "chain_loss": null, "readout_loss": 0.8416770423839584, "readout_argmax_matches_target": 0.6304347826086957, "readout_mean_top_prob": 0.8406794619733009, "readout_target_entropy": 0.04357187665890956, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
	{"step": 3, "of": 17, "lr": 0.000288581929876693, "grad_norm": 8.426118850708008, "examples": 90, "readout_examples": 90, "reasoner_examples": 0, "step_tokens": 42718, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3391.384143001221, "elapsed_min": 0.6710730989774069, "eta_min": 3.1316744618945647, "chain_loss": null, "readout_loss": 1.5099286515679624, "readout_argmax_matches_target": 0.2777777777777778, "readout_mean_top_prob": 0.6285712169276343, "readout_target_entropy": 0.13333093267348078, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
	{"step": 10, "of": 17, "lr": 0.00012073645169758076, "grad_norm": 5.670900821685791, "examples": 92, "readout_examples": 92, "reasoner_examples": 0, "step_tokens": 45015, "chain_tokens": 0, "micro_batches": 3, "tok_per_s": 3662.2643625865917, "elapsed_min": 2.028567373752594, "eta_min": 1.4199971616268159, "chain_loss": null, "readout_loss": 0.9605440597819246, "readout_argmax_matches_target": 0.5543478260869565, "readout_mean_top_prob": 0.5456273620543273, "readout_target_entropy": 0.06314167380332947, "letter_loss": null, "letter_top_prob": null, "mem_gb": 17.477917696}
	{"dev_n": 500, "dev_acc": 0.638, "dev_nll": 0.8704747354204301, "dev_ece15": 0.055960430085659034}
	{"dev_n": 500, "dev_acc": 0.638, "dev_nll": 0.8704747354204301, "dev_ece15": 0.055960430085659034, "v2_options": {"readout_pool": "all", "subset": "data/sweep_ids.txt", "min_correct": 2, "second_chain_min": 3, "rep_max": 20, "post_soft": true, "eval_order_frac": 0.5, "drop_recall": true, "strip_cw": true}, "speed_options": {"liger": false, "ckpt_above": 0, "micro_tokens": 16384, "no_probe": true}, "mode": "readout_only", "alpha": 0.5, "lr": 0.0003, "rank": 32, "lora_alpha": 64, "seed": 17, "examples": 1500, "steps_done": 17, "steps_planned": 17, "stopped_at_deadline": false, "train_tokens": 753558, "train_seconds": 200.36477303504944, "tok_per_s": 3760.9305697073883, "counts": {"readout_from_votes": 259, "readout_eval_order": 720, "readout_gold_only": 1241}, "reasoner_examples": 0, "readout_examples": 1500}
	{"time": "2026-10-31T13:59:18Z", "event": "done", "stage": "sweep_3e-4"}
	RUN_V2_EXIT 143