diff --git a/add_noise_b9057b9c-a10c-4d1d-a360-4c79e5201dca.log b/add_noise_b9057b9c-a10c-4d1d-a360-4c79e5201dca.log new file mode 100644 index 0000000000000000000000000000000000000000..b1d0574b18e994e37cc87948287a85eab9c38524 --- /dev/null +++ b/add_noise_b9057b9c-a10c-4d1d-a360-4c79e5201dca.log @@ -0,0 +1,44 @@ +`torch_dtype` is deprecated! Use `dtype` instead! + +Loading checkpoint shards: 0%| | 0/4 [00:00 /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_intercode train=1980 val=20 + +Saving the dataset (0/1 shards): 0%| | 0/47700 [00:00 /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_goofspiel +[intercode_synth] fs2 progress 500/4000 elapsed=12s built=872 +[intercode_synth] fs2 progress 1000/4000 elapsed=14s built=999 +[intercode_synth] fs2 progress 1500/4000 elapsed=15s built=1091 +[intercode_synth] fs2 progress 2000/4000 elapsed=16s built=1146 +[intercode_synth] fs2 progress 2500/4000 elapsed=17s built=1187 +[intercode_synth] fs2: produced 600 grounded tasks (30 source files). + +Saving the dataset (0/1 shards): 0%| | 0/3960 [00:00 /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_intercode train=3960 val=40 +[intercode_synth] fs4 progress 500/4000 elapsed=20s built=1619 +[intercode_synth] fs4: produced 600 grounded tasks (0 source files). +[intercode_synth] DONE built=1800 tasks -> 5278 examples (dropped_unsafe=32, dropped_empty/err=318) + +Saving the dataset (0/1 shards): 0%| | 0/5225 [00:00 /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_intercode train=5225 val=53 + +Saving the dataset (0/1 shards): 0%| | 0/5225 [00:00 /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_intercode train=5225 val=53 +[generate_trajectories] game 425/8500 (5.0%) wins=360 losses=65 draws=0 win_rate=84.7% raw_total=2673 timeout=0 worker_fail=0 elapsed=31s rate=13.66games/s eta=591s +[generate_trajectories] game 850/8500 (10.0%) wins=707 losses=142 draws=0 win_rate=83.3% raw_total=5293 timeout=0 worker_fail=0 elapsed=59s rate=14.50games/s eta=527s +[generate_trajectories] game 1275/8500 (15.0%) wins=1051 losses=223 draws=0 win_rate=82.5% raw_total=7865 timeout=0 worker_fail=0 elapsed=85s rate=15.09games/s eta=479s +[generate_trajectories] game 1700/8500 (20.0%) wins=1400 losses=299 draws=0 win_rate=82.4% raw_total=10431 timeout=0 worker_fail=0 elapsed=110s rate=15.45games/s eta=440s +[generate_trajectories] game 2125/8500 (25.0%) wins=1755 losses=369 draws=0 win_rate=82.6% raw_total=13018 timeout=0 worker_fail=0 elapsed=135s rate=15.76games/s eta=405s +[generate_trajectories] game 2550/8500 (30.0%) wins=2108 losses=441 draws=0 win_rate=82.7% raw_total=15616 timeout=0 worker_fail=0 elapsed=161s rate=15.87games/s eta=375s +[generate_trajectories] game 2975/8500 (35.0%) wins=2469 losses=505 draws=0 win_rate=83.0% raw_total=18212 timeout=0 worker_fail=0 elapsed=186s rate=16.02games/s eta=345s +[generate_trajectories] game 300/6000 (5.0%) wins=255 losses=41 draws=4 win_rate=85.0% raw_total=8538 timeout=0 worker_fail=0 elapsed=204s rate=1.47games/s eta=3878s +[generate_trajectories] game 3400/8500 (40.0%) wins=2819 losses=580 draws=0 win_rate=82.9% raw_total=20743 timeout=0 worker_fail=0 elapsed=209s rate=16.26games/s eta=314s +[generate_trajectories] game 3825/8500 (45.0%) wins=3170 losses=654 draws=0 win_rate=82.9% raw_total=23295 timeout=0 worker_fail=0 elapsed=233s rate=16.40games/s eta=285s +[generate_trajectories] game 4250/8500 (50.0%) wins=3502 losses=747 draws=0 win_rate=82.4% raw_total=25757 timeout=0 worker_fail=0 elapsed=255s rate=16.70games/s eta=255s +[generate_trajectories] game 4675/8500 (55.0%) wins=3852 losses=821 draws=0 win_rate=82.4% raw_total=28347 timeout=0 worker_fail=0 elapsed=279s rate=16.75games/s eta=228s +[generate_trajectories] game 5100/8500 (60.0%) wins=4202 losses=896 draws=0 win_rate=82.4% raw_total=30874 timeout=0 worker_fail=0 elapsed=301s rate=16.92games/s eta=201s +[generate_trajectories] game 5525/8500 (65.0%) wins=4553 losses=970 draws=0 win_rate=82.4% raw_total=33465 timeout=0 worker_fail=0 elapsed=326s rate=16.94games/s eta=176s +[generate_trajectories] game 5950/8500 (70.0%) wins=4895 losses=1053 draws=0 win_rate=82.3% raw_total=35998 timeout=0 worker_fail=0 elapsed=349s rate=17.05games/s eta=150s +[generate_trajectories] game 6375/8500 (75.0%) wins=5251 losses=1122 draws=0 win_rate=82.4% raw_total=38614 timeout=0 worker_fail=0 elapsed=374s rate=17.05games/s eta=125s +[generate_trajectories] game 6800/8500 (80.0%) wins=5602 losses=1196 draws=0 win_rate=82.4% raw_total=41193 timeout=0 worker_fail=0 elapsed=397s rate=17.11games/s eta=99s +[generate_trajectories] game 600/6000 (10.0%) wins=524 losses=68 draws=8 win_rate=87.3% raw_total=17111 timeout=0 worker_fail=0 elapsed=406s rate=1.48games/s eta=3652s +[generate_trajectories] game 7225/8500 (85.0%) wins=5952 losses=1271 draws=0 win_rate=82.4% raw_total=43773 timeout=0 worker_fail=0 elapsed=421s rate=17.14games/s eta=74s +[generate_trajectories] game 7650/8500 (90.0%) wins=6309 losses=1339 draws=0 win_rate=82.5% raw_total=46391 timeout=0 worker_fail=0 elapsed=448s rate=17.09games/s eta=50s +[generate_trajectories] game 8075/8500 (95.0%) wins=6658 losses=1415 draws=0 win_rate=82.5% raw_total=48949 timeout=0 worker_fail=0 elapsed=472s rate=17.10games/s eta=25s +[generate_trajectories] game 8500/8500 (100.0%) wins=7001 losses=1497 draws=0 win_rate=82.4% raw_total=51534 timeout=0 worker_fail=0 elapsed=498s rate=17.05games/s eta=0s +[generate_trajectories] DONE collected=51532 skipped=2 filtered_by_score=0 timeout_failed=0 worker_failed=0 wins=7001 losses=1497 draws=0 no_score=2 win_rate=82.4% time_cap_hit=False total_elapsed=500s +[generate_trajectories] windows total=51532 +[generate_trajectories] stats={'total': 51532, 'avg_assistant_turns': 1.0, 'turn_distribution': {1: 51532}} + +Saving the dataset (0/1 shards): 0%| | 0/51016 [00:00 /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_clobber +[generate_trajectories] game 900/6000 (15.0%) wins=800 losses=89 draws=11 win_rate=88.9% raw_total=25706 timeout=0 worker_fail=0 elapsed=564s rate=1.60games/s eta=3194s +[generate_trajectories] TIME CAP 648s reached at 1168/6000 games; cancelled 4798 pending futures, draining in-flight +[generate_trajectories] game 1169/6000 (19.5%) wins=1044 losses=109 draws=16 win_rate=89.3% raw_total=33398 timeout=0 worker_fail=0 elapsed=648s rate=1.80games/s eta=2678s +[generate_trajectories] game 1170/6000 (19.5%) wins=1045 losses=109 draws=16 win_rate=89.3% raw_total=33428 timeout=0 worker_fail=0 elapsed=649s rate=1.80games/s eta=2677s +[generate_trajectories] game 1171/6000 (19.5%) wins=1046 losses=109 draws=16 win_rate=89.3% raw_total=33456 timeout=0 worker_fail=0 elapsed=649s rate=1.80games/s eta=2676s +[generate_trajectories] game 1172/6000 (19.5%) wins=1047 losses=109 draws=16 win_rate=89.3% raw_total=33485 timeout=0 worker_fail=0 elapsed=649s rate=1.81games/s eta=2674s +[generate_trajectories] game 1173/6000 (19.6%) wins=1048 losses=109 draws=16 win_rate=89.3% raw_total=33513 timeout=0 worker_fail=0 elapsed=651s rate=1.80games/s eta=2679s +[generate_trajectories] game 1174/6000 (19.6%) wins=1049 losses=109 draws=16 win_rate=89.4% raw_total=33541 timeout=0 worker_fail=0 elapsed=651s rate=1.80games/s eta=2677s +[generate_trajectories] game 1175/6000 (19.6%) wins=1050 losses=109 draws=16 win_rate=89.4% raw_total=33571 timeout=0 worker_fail=0 elapsed=651s rate=1.80games/s eta=2674s +[generate_trajectories] game 1176/6000 (19.6%) wins=1050 losses=109 draws=17 win_rate=89.3% raw_total=33601 timeout=0 worker_fail=0 elapsed=651s rate=1.81games/s eta=2672s +[generate_trajectories] game 1177/6000 (19.6%) wins=1051 losses=109 draws=17 win_rate=89.3% raw_total=33629 timeout=0 worker_fail=0 elapsed=652s rate=1.81games/s eta=2671s +[generate_trajectories] game 1178/6000 (19.6%) wins=1052 losses=109 draws=17 win_rate=89.3% raw_total=33657 timeout=0 worker_fail=0 elapsed=652s rate=1.81games/s eta=2669s +[generate_trajectories] game 1179/6000 (19.6%) wins=1053 losses=109 draws=17 win_rate=89.3% raw_total=33684 timeout=0 worker_fail=0 elapsed=653s rate=1.81games/s eta=2671s +[generate_trajectories] game 1180/6000 (19.7%) wins=1054 losses=109 draws=17 win_rate=89.3% raw_total=33712 timeout=0 worker_fail=0 elapsed=654s rate=1.80games/s eta=2670s +[generate_trajectories] game 1181/6000 (19.7%) wins=1054 losses=110 draws=17 win_rate=89.2% raw_total=33740 timeout=0 worker_fail=0 elapsed=654s rate=1.81games/s eta=2668s +[generate_trajectories] game 1182/6000 (19.7%) wins=1055 losses=110 draws=17 win_rate=89.3% raw_total=33770 timeout=0 worker_fail=0 elapsed=654s rate=1.81games/s eta=2665s +[generate_trajectories] game 1183/6000 (19.7%) wins=1055 losses=111 draws=17 win_rate=89.2% raw_total=33799 timeout=0 worker_fail=0 elapsed=654s rate=1.81games/s eta=2664s +[generate_trajectories] game 1184/6000 (19.7%) wins=1056 losses=111 draws=17 win_rate=89.2% raw_total=33829 timeout=0 worker_fail=0 elapsed=654s rate=1.81games/s eta=2662s +[generate_trajectories] game 1185/6000 (19.8%) wins=1057 losses=111 draws=17 win_rate=89.2% raw_total=33858 timeout=0 worker_fail=0 elapsed=655s rate=1.81games/s eta=2660s +[generate_trajectories] game 1186/6000 (19.8%) wins=1058 losses=111 draws=17 win_rate=89.2% raw_total=33886 timeout=0 worker_fail=0 elapsed=655s rate=1.81games/s eta=2659s +[generate_trajectories] game 1187/6000 (19.8%) wins=1059 losses=111 draws=17 win_rate=89.2% raw_total=33914 timeout=0 worker_fail=0 elapsed=655s rate=1.81games/s eta=2657s +[generate_trajectories] game 1188/6000 (19.8%) wins=1060 losses=111 draws=17 win_rate=89.2% raw_total=33942 timeout=0 worker_fail=0 elapsed=656s rate=1.81games/s eta=2655s +[generate_trajectories] game 1189/6000 (19.8%) wins=1061 losses=111 draws=17 win_rate=89.2% raw_total=33971 timeout=0 worker_fail=0 elapsed=656s rate=1.81games/s eta=2655s +[generate_trajectories] game 1190/6000 (19.8%) wins=1062 losses=111 draws=17 win_rate=89.2% raw_total=33999 timeout=0 worker_fail=0 elapsed=656s rate=1.81games/s eta=2653s +[generate_trajectories] game 1191/6000 (19.9%) wins=1063 losses=111 draws=17 win_rate=89.3% raw_total=34027 timeout=0 worker_fail=0 elapsed=656s rate=1.81games/s eta=2650s +[generate_trajectories] game 1192/6000 (19.9%) wins=1064 losses=111 draws=17 win_rate=89.3% raw_total=34055 timeout=0 worker_fail=0 elapsed=656s rate=1.82games/s eta=2648s +[generate_trajectories] game 1193/6000 (19.9%) wins=1065 losses=111 draws=17 win_rate=89.3% raw_total=34083 timeout=0 worker_fail=0 elapsed=656s rate=1.82games/s eta=2645s +[generate_trajectories] game 1194/6000 (19.9%) wins=1065 losses=112 draws=17 win_rate=89.2% raw_total=34112 timeout=0 worker_fail=0 elapsed=657s rate=1.82games/s eta=2643s +[generate_trajectories] game 1195/6000 (19.9%) wins=1066 losses=112 draws=17 win_rate=89.2% raw_total=34140 timeout=0 worker_fail=0 elapsed=657s rate=1.82games/s eta=2642s +[generate_trajectories] game 1196/6000 (19.9%) wins=1067 losses=112 draws=17 win_rate=89.2% raw_total=34169 timeout=0 worker_fail=0 elapsed=657s rate=1.82games/s eta=2640s +[generate_trajectories] game 1197/6000 (19.9%) wins=1068 losses=112 draws=17 win_rate=89.2% raw_total=34198 timeout=0 worker_fail=0 elapsed=658s rate=1.82games/s eta=2640s +[generate_trajectories] game 1198/6000 (20.0%) wins=1069 losses=112 draws=17 win_rate=89.2% raw_total=34226 timeout=0 worker_fail=0 elapsed=658s rate=1.82games/s eta=2637s +[generate_trajectories] game 1199/6000 (20.0%) wins=1070 losses=112 draws=17 win_rate=89.2% raw_total=34257 timeout=0 worker_fail=0 elapsed=658s rate=1.82games/s eta=2635s +[generate_trajectories] game 1200/6000 (20.0%) wins=1071 losses=112 draws=17 win_rate=89.2% raw_total=34285 timeout=0 worker_fail=0 elapsed=658s rate=1.82games/s eta=2634s +[generate_trajectories] game 1201/6000 (20.0%) wins=1072 losses=112 draws=17 win_rate=89.3% raw_total=34313 timeout=0 worker_fail=0 elapsed=660s rate=1.82games/s eta=2636s +[generate_trajectories] game 1202/6000 (20.0%) wins=1073 losses=112 draws=17 win_rate=89.3% raw_total=34341 timeout=0 worker_fail=0 elapsed=661s rate=1.82games/s eta=2639s +[generate_trajectories] DONE collected=34341 skipped=0 filtered_by_score=0 timeout_failed=0 worker_failed=0 wins=1073 losses=112 draws=17 no_score=0 win_rate=89.3% time_cap_hit=True cancelled_pending=4798 total_elapsed=661s +[generate_trajectories] windows total=34341 +[generate_trajectories] stats={'total': 34341, 'avg_assistant_turns': 1.0, 'turn_distribution': {1: 34341}} + +Saving the dataset (0/1 shards): 0%| | 0/33997 [00:00 /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_othello +/workspace/scripts/our_envs/alf_world_env.py:8: TRLExperimentalWarning: You are importing from 'trl.experimental'. APIs here are unstable and may change or be removed without notice. Silence this warning by setting environment variable TRL_EXPERIMENTAL_SILENCE=1. + from trl.experimental.openenv import generate_rollout_completions +[merge_trajectories] merging 4 datasets (target_per_env=off max_upsample=3.0): + [0] /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_othello + [1] /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_intercode + [2] /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_clobber + [3] /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_goofspiel + +Saving the dataset (0/2 shards): 0%| | 0/137938 [00:00 +[rank0]: main() +[rank0]: File "/workspace/scripts/train_sft_env.py", line 543, in main +[rank0]: trainer = SFTTrainer( +[rank0]: ^^^^^^^^^^^ +[rank0]: File "/workspace/.grpo_env/lib/python3.12/site-packages/trl/trainer/sft_trainer.py", line 883, in __init__ +[rank0]: super().__init__( +[rank0]: File "/workspace/.grpo_env/lib/python3.12/site-packages/transformers/utils/deprecation.py", line 172, in wrapped_func +[rank0]: return func(*args, **kwargs) +[rank0]: ^^^^^^^^^^^^^^^^^^^^^ +[rank0]: File "/workspace/.grpo_env/lib/python3.12/site-packages/transformers/trainer.py", line 612, in __init__ +[rank0]: self._move_model_to_device(model, args.device) +[rank0]: File "/workspace/.grpo_env/lib/python3.12/site-packages/transformers/trainer.py", line 910, in _move_model_to_device +[rank0]: model = model.to(device) +[rank0]: ^^^^^^^^^^^^^^^^ +[rank0]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1369, in to +[rank0]: return self._apply(convert) +[rank0]: ^^^^^^^^^^^^^^^^^^^^ +[rank0]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 928, in _apply +[rank0]: module._apply(fn) +[rank0]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 928, in _apply +[rank0]: module._apply(fn) +[rank0]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 928, in _apply +[rank0]: module._apply(fn) +[rank0]: [Previous line repeated 5 more times] +[rank0]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 955, in _apply +[rank0]: param_applied = fn(param) +[rank0]: ^^^^^^^^^ +[rank0]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1355, in convert +[rank0]: return t.to( +[rank0]: ^^^^^ +[rank0]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 112.00 MiB. GPU 0 has a total capacity of 15.57 GiB of which 82.19 MiB is free. Including non-PyTorch memory, this process has 15.48 GiB memory in use. Of the allocated memory 15.26 GiB is allocated by PyTorch, and 45.55 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables) +[rank3]: Traceback (most recent call last): +[rank3]: File "/workspace/scripts/train_sft_env.py", line 578, in +[rank3]: main() +[rank3]: File "/workspace/scripts/train_sft_env.py", line 543, in main +[rank3]: trainer = SFTTrainer( +[rank3]: ^^^^^^^^^^^ +[rank3]: File "/workspace/.grpo_env/lib/python3.12/site-packages/trl/trainer/sft_trainer.py", line 883, in __init__ +[rank3]: super().__init__( +[rank3]: File "/workspace/.grpo_env/lib/python3.12/site-packages/transformers/utils/deprecation.py", line 172, in wrapped_func +[rank3]: return func(*args, **kwargs) +[rank3]: ^^^^^^^^^^^^^^^^^^^^^ +[rank3]: File "/workspace/.grpo_env/lib/python3.12/site-packages/transformers/trainer.py", line 612, in __init__ +[rank3]: self._move_model_to_device(model, args.device) +[rank3]: File "/workspace/.grpo_env/lib/python3.12/site-packages/transformers/trainer.py", line 910, in _move_model_to_device +[rank3]: model = model.to(device) +[rank3]: ^^^^^^^^^^^^^^^^ +[rank3]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1369, in to +[rank3]: return self._apply(convert) +[rank3]: ^^^^^^^^^^^^^^^^^^^^ +[rank3]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 928, in _apply +[rank3]: module._apply(fn) +[rank3]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 928, in _apply +[rank3]: module._apply(fn) +[rank3]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 928, in _apply +[rank3]: module._apply(fn) +[rank3]: [Previous line repeated 5 more times] +[rank3]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 955, in _apply +[rank3]: param_applied = fn(param) +[rank3]: ^^^^^^^^^ +[rank3]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1355, in convert +[rank3]: return t.to( +[rank3]: ^^^^^ +[rank3]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 112.00 MiB. GPU 3 has a total capacity of 15.57 GiB of which 82.19 MiB is free. Including non-PyTorch memory, this process has 15.48 GiB memory in use. Of the allocated memory 15.26 GiB is allocated by PyTorch, and 45.55 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables) +[rank2]: Traceback (most recent call last): +[rank2]: File "/workspace/scripts/train_sft_env.py", line 578, in +[rank2]: main() +[rank2]: File "/workspace/scripts/train_sft_env.py", line 543, in main +[rank2]: trainer = SFTTrainer( +[rank2]: ^^^^^^^^^^^ +[rank2]: File "/workspace/.grpo_env/lib/python3.12/site-packages/trl/trainer/sft_trainer.py", line 883, in __init__ +[rank2]: super().__init__( +[rank2]: File "/workspace/.grpo_env/lib/python3.12/site-packages/transformers/utils/deprecation.py", line 172, in wrapped_func +[rank2]: return func(*args, **kwargs) +[rank2]: ^^^^^^^^^^^^^^^^^^^^^ +[rank2]: File "/workspace/.grpo_env/lib/python3.12/site-packages/transformers/trainer.py", line 612, in __init__ +[rank2]: self._move_model_to_device(model, args.device) +[rank2]: File "/workspace/.grpo_env/lib/python3.12/site-packages/transformers/trainer.py", line 910, in _move_model_to_device +[rank2]: model = model.to(device) +[rank2]: ^^^^^^^^^^^^^^^^ +[rank2]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1369, in to +[rank2]: return self._apply(convert) +[rank2]: ^^^^^^^^^^^^^^^^^^^^ +[rank2]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 928, in _apply +[rank2]: module._apply(fn) +[rank2]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 928, in _apply +[rank2]: module._apply(fn) +[rank2]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 928, in _apply +[rank2]: module._apply(fn) +[rank2]: [Previous line repeated 5 more times] +[rank2]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 955, in _apply +[rank2]: param_applied = fn(param) +[rank2]: ^^^^^^^^^ +[rank2]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1355, in convert +[rank2]: return t.to( +[rank2]: ^^^^^ +[rank2]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 112.00 MiB. GPU 2 has a total capacity of 15.57 GiB of which 82.19 MiB is free. Including non-PyTorch memory, this process has 15.48 GiB memory in use. Of the allocated memory 15.26 GiB is allocated by PyTorch, and 45.55 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables) +[rank1]: Traceback (most recent call last): +[rank1]: File "/workspace/scripts/train_sft_env.py", line 578, in +[rank1]: main() +[rank1]: File "/workspace/scripts/train_sft_env.py", line 543, in main +[rank1]: trainer = SFTTrainer( +[rank1]: ^^^^^^^^^^^ +[rank1]: File "/workspace/.grpo_env/lib/python3.12/site-packages/trl/trainer/sft_trainer.py", line 883, in __init__ +[rank1]: super().__init__( +[rank1]: File "/workspace/.grpo_env/lib/python3.12/site-packages/transformers/utils/deprecation.py", line 172, in wrapped_func +[rank1]: return func(*args, **kwargs) +[rank1]: ^^^^^^^^^^^^^^^^^^^^^ +[rank1]: File "/workspace/.grpo_env/lib/python3.12/site-packages/transformers/trainer.py", line 612, in __init__ +[rank1]: self._move_model_to_device(model, args.device) +[rank1]: File "/workspace/.grpo_env/lib/python3.12/site-packages/transformers/trainer.py", line 910, in _move_model_to_device +[rank1]: model = model.to(device) +[rank1]: ^^^^^^^^^^^^^^^^ +[rank1]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1369, in to +[rank1]: return self._apply(convert) +[rank1]: ^^^^^^^^^^^^^^^^^^^^ +[rank1]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 928, in _apply +[rank1]: module._apply(fn) +[rank1]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 928, in _apply +[rank1]: module._apply(fn) +[rank1]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 928, in _apply +[rank1]: module._apply(fn) +[rank1]: [Previous line repeated 5 more times] +[rank1]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 955, in _apply +[rank1]: param_applied = fn(param) +[rank1]: ^^^^^^^^^ +[rank1]: File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/nn/modules/module.py", line 1355, in convert +[rank1]: return t.to( +[rank1]: ^^^^^ +[rank1]: torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 112.00 MiB. GPU 1 has a total capacity of 15.57 GiB of which 82.19 MiB is free. Including non-PyTorch memory, this process has 15.48 GiB memory in use. Of the allocated memory 15.26 GiB is allocated by PyTorch, and 45.55 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables) +W0829 18:33:08.043000 18441 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 18491 closing signal SIGTERM +W0829 18:33:08.044000 18441 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 18492 closing signal SIGTERM +W0829 18:33:08.044000 18441 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 18493 closing signal SIGTERM +E0829 18:33:08.359000 18441 torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: 1) local_rank: 0 (pid: 18490) of binary: /workspace/.grpo_env/bin/python3 +Traceback (most recent call last): + File "/workspace/.grpo_env/bin/torchrun", line 10, in + sys.exit(main()) + ^^^^^^ + File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 357, in wrapper + return f(*args, **kwargs) + ^^^^^^^^^^^^^^^^^^ + File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/distributed/run.py", line 901, in main + run(args) + File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/distributed/run.py", line 892, in run + elastic_launch( + File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/distributed/launcher/api.py", line 143, in __call__ + return launch_agent(self._config, self._entrypoint, list(args)) + ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + File "/workspace/.grpo_env/lib/python3.12/site-packages/torch/distributed/launcher/api.py", line 277, in launch_agent + raise ChildFailedError( +torch.distributed.elastic.multiprocessing.errors.ChildFailedError: +============================================================ +train_sft_env.py FAILED +------------------------------------------------------------ +Failures: + +------------------------------------------------------------ +Root Cause (first observed failure): +[0]: + time : 2026-08-29_18:33:08 + host : 957a212b4cec + rank : 0 (local_rank: 0) + exitcode : 1 (pid: 18490) + error_file: + traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html +============================================================ + +Process completed with return code: 1 diff --git a/training_request_b9057b9c-a10c-4d1d-a360-4c79e5201dca.json b/training_request_b9057b9c-a10c-4d1d-a360-4c79e5201dca.json new file mode 100644 index 0000000000000000000000000000000000000000..a687fcb73e361c79c2e725c543f83820a146783a --- /dev/null +++ b/training_request_b9057b9c-a10c-4d1d-a360-4c79e5201dca.json @@ -0,0 +1,37 @@ +{ + "train_request": { + "model_name": "unsloth/Meta-Llama-3.1-8B-Instruct", + "model_path": "/cache/models/unsloth--Meta-Llama-3.1-8B-Instruct", + "task_id": "b9057b9c-a10c-4d1d-a360-4c79e5201dca", + "dataset": "/cache/datasets/b9057b9c-a10c-4d1d-a360-4c79e5201dca_train_data.json", + "hours_to_complete": 1.5, + "expected_repo_name": "tournament-tourn_1e08b0b313ccc59f", + "end_time": "2026-08-29 19:38:28", + "dataset_type": { + "environment_names": [ + "othello", + "intercode", + "clobber", + "goofspiel" + ] + }, + "submission_dir": "/app/checkpoints/b9057b9c-a10c-4d1d-a360-4c79e5201dca/tournament-tourn_1e08b0b313ccc59f", + "output_dir": "/workspace/scripts/soutputs/b9057b9c-a10c-4d1d-a360-4c79e5201dca", + "adjust_batch_size": false, + "request_path": "datasets/training_request_b9057b9c-a10c-4d1d-a360-4c79e5201dca.json", + "max_data_size": -1, + "max_steps": -1, + "wandb_log_dir": "/cache/wandb_logs", + "min_steps": 105, + "is_openai": false, + "reg_ratio": 1.0, + "find_lk_lr": true, + "checking_mode": "first_time", + "dataset_path": "/workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca", + "save_before_remaining_time": 5, + "periodic_save_steps": 200, + "checking_step": 70 + }, + "run_cmd": "torchrun --nproc_per_node=4 train_sft_env.py --request_path datasets/training_request_b9057b9c-a10c-4d1d-a360-4c79e5201dca.json --bf16 True --report_to wandb --output_dir /workspace/scripts/soutputs/b9057b9c-a10c-4d1d-a360-4c79e5201dca --num_train_epochs 1 --per_device_train_batch_size 32 --per_device_eval_batch_size 1 --gradient_accumulation_steps 1 --eval_accumulation_steps 1 --eval_strategy no --save_strategy epoch --logging_steps 5 --learning_rate 1.4555263157894739e-05 --weight_decay 0. --warmup_steps 35 --lr_scheduler_type cosine_with_min_lr --lr_scheduler_kwargs \"{\\\"min_lr_rate\\\": 0.25}\" --tf32 True --gradient_checkpointing True --optim paged_adamw_8bit --use_liger_kernel True --packing False --disable_fa False --max_length 4096 --use_lora True", + "generate_cmd": "(python -m our_envs.generate_trajectories --environment_name othello --output_path /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_othello --num_games 6000 --max_turn 64 --window_turns 12 --window_step 6 --num_workers 16 --max-gen-seconds 648 --wins-only) & (python -m our_envs.intercode_synth_gen --output_path /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_intercode --per_fs 4000 --max_per_fs 600) & (python -m our_envs.generate_trajectories --environment_name clobber --output_path /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_clobber --num_games 8500 --max_turn 70 --window_turns 12 --window_step 6 --num_workers 16 --max-gen-seconds 648 --wins-only) & (python -m our_envs.generate_trajectories --environment_name goofspiel --output_path /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_goofspiel --num_games 6000 --max_turn 26 --window_turns 10 --window_step 5 --num_workers 16 --max-gen-seconds 648 --wins-only) & wait && python -m our_envs.merge_trajectories --input_paths /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_othello /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_intercode /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_clobber /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_goofspiel --output_path /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca" +} \ No newline at end of file diff --git a/training_request_b9057b9c-a10c-4d1d-a360-4c79e5201dca_0.json b/training_request_b9057b9c-a10c-4d1d-a360-4c79e5201dca_0.json new file mode 100644 index 0000000000000000000000000000000000000000..09582206d8314e985eb1f601f6935a683deaafe8 --- /dev/null +++ b/training_request_b9057b9c-a10c-4d1d-a360-4c79e5201dca_0.json @@ -0,0 +1,37 @@ +{ + "train_request": { + "model_name": "unsloth/Meta-Llama-3.1-8B-Instruct", + "model_path": "/cache/models/unsloth--Meta-Llama-3.1-8B-Instruct", + "task_id": "b9057b9c-a10c-4d1d-a360-4c79e5201dca", + "dataset": "/cache/datasets/b9057b9c-a10c-4d1d-a360-4c79e5201dca_train_data.json", + "hours_to_complete": 1.5, + "expected_repo_name": "tournament-tourn_1e08b0b313ccc59f", + "end_time": "2026-08-29 19:38:28", + "dataset_type": { + "environment_names": [ + "othello", + "intercode", + "clobber", + "goofspiel" + ] + }, + "submission_dir": "/app/checkpoints/b9057b9c-a10c-4d1d-a360-4c79e5201dca/tournament-tourn_1e08b0b313ccc59f", + "output_dir": "/workspace/scripts/soutputs/b9057b9c-a10c-4d1d-a360-4c79e5201dca", + "adjust_batch_size": false, + "request_path": "datasets/training_request_b9057b9c-a10c-4d1d-a360-4c79e5201dca.json", + "max_data_size": -1, + "max_steps": -1, + "wandb_log_dir": "/cache/wandb_logs", + "min_steps": 105, + "is_openai": false, + "reg_ratio": 1.0, + "find_lk_lr": true, + "checking_mode": "none", + "dataset_path": "/workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca", + "save_before_remaining_time": 5, + "periodic_save_steps": 200, + "checking_step": 70 + }, + "run_cmd": "torchrun --nproc_per_node=4 train_sft_env.py --request_path datasets/training_request_b9057b9c-a10c-4d1d-a360-4c79e5201dca.json --bf16 True --report_to wandb --output_dir /workspace/scripts/soutputs/b9057b9c-a10c-4d1d-a360-4c79e5201dca --num_train_epochs 1 --per_device_train_batch_size 32 --per_device_eval_batch_size 1 --gradient_accumulation_steps 1 --eval_accumulation_steps 1 --eval_strategy no --save_strategy epoch --logging_steps 5 --learning_rate 1.4555263157894739e-05 --weight_decay 0. --warmup_steps 35 --lr_scheduler_type cosine_with_min_lr --lr_scheduler_kwargs \"{\\\"min_lr_rate\\\": 0.25}\" --tf32 True --gradient_checkpointing True --optim paged_adamw_8bit --use_liger_kernel True --packing False --disable_fa False --max_length 4096 --use_lora True", + "generate_cmd": "(python -m our_envs.generate_trajectories --environment_name othello --output_path /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_othello --num_games 6000 --max_turn 64 --window_turns 12 --window_step 6 --num_workers 16 --max-gen-seconds 648 --wins-only) & (python -m our_envs.intercode_synth_gen --output_path /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_intercode --per_fs 4000 --max_per_fs 600) & (python -m our_envs.generate_trajectories --environment_name clobber --output_path /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_clobber --num_games 8500 --max_turn 70 --window_turns 12 --window_step 6 --num_workers 16 --max-gen-seconds 648 --wins-only) & (python -m our_envs.generate_trajectories --environment_name goofspiel --output_path /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_goofspiel --num_games 6000 --max_turn 26 --window_turns 10 --window_step 5 --num_workers 16 --max-gen-seconds 648 --wins-only) & wait && python -m our_envs.merge_trajectories --input_paths /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_othello /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_intercode /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_clobber /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca_goofspiel --output_path /workspace/scripts/datasets/sft_env_b9057b9c-a10c-4d1d-a360-4c79e5201dca" +} \ No newline at end of file