2025-02-18 14:33:03 - WARNING - __main__ - Process rank: 0, device: cuda:0, n_gpu: 1 distributed training: True, 16-bits training: False
2025-02-18 14:33:03 - INFO - __main__ - Model parameters ModelConfig(model_name_or_path='Qwen/Qwen2.5-1.5B-Instruct', model_revision='main', torch_dtype='bfloat16', trust_remote_code=False, attn_implementation='flash_attention_2', use_peft=False, lora_r=16, lora_alpha=32, lora_dropout=0.05, lora_target_modules=None, lora_modules_to_save=None, lora_task_type='CAUSAL_LM', use_rslora=False, load_in_8bit=False, load_in_4bit=False, bnb_4bit_quant_type='nf4', use_bnb_nested_quant=False)
2025-02-18 14:33:03 - INFO - __main__ - Script parameters GRPOScriptArguments(dataset_name='AI-MO/NuminaMath-TIR', dataset_config=None, dataset_train_split='train', dataset_test_split='test', gradient_checkpointing_use_reentrant=False, ignore_bias_buffers=False, reward_funcs=['accuracy', 'format'], cosine_min_value_wrong=0.0, cosine_max_value_wrong=-0.5, cosine_min_value_correct=0.5, cosine_max_value_correct=1.0, cosine_max_len=1000, repetition_n_grams=3, repetition_max_penalty=-1.0)
2025-02-18 14:33:03 - INFO - __main__ - Training parameters GRPOConfig(
_n_gpu=1,
accelerator_config={'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None, 'use_configured_state': False},
adafactor=False,
adam_beta1=0.9,
adam_beta2=0.999,
adam_epsilon=1e-08,
auto_find_batch_size=False,
average_tokens_across_devices=False,
batch_eval_metrics=False,
benchmarks=[],
beta=0.04,
bf16=True,
bf16_full_eval=False,
callbacks=[],
data_seed=None,
dataloader_drop_last=False,
dataloader_num_workers=0,
dataloader_persistent_workers=False,
dataloader_pin_memory=True,
dataloader_prefetch_factor=None,
ddp_backend=None,
ddp_broadcast_buffers=None,
ddp_bucket_cap_mb=None,
ddp_find_unused_parameters=None,
ddp_timeout=1800,
debug=[],
deepspeed=None,
disable_tqdm=False,
dispatch_batches=None,
do_eval=True,
do_predict=False,
do_train=False,
ds3_gather_for_generation=True,
eval_accumulation_steps=None,
eval_delay=0,
eval_do_concat_batches=True,
eval_on_start=False,
eval_steps=100,
eval_strategy=IntervalStrategy.STEPS,
eval_use_gather_object=False,
evaluation_strategy=None,
fp16=False,
fp16_backend=auto,
fp16_full_eval=False,
fp16_opt_level=O1,
fsdp=[],
fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False},
fsdp_min_num_params=0,
fsdp_transformer_layer_cls_to_wrap=None,
full_determinism=False,
gradient_accumulation_steps=16,
gradient_checkpointing=True,
gradient_checkpointing_kwargs={'use_reentrant': False},
greater_is_better=None,
group_by_length=False,
half_precision_backend=auto,
hub_always_push=False,
hub_model_id=Qwen2.5-1.5B-Open-R1-GRPO,
hub_model_revision=main,
hub_private_repo=None,
hub_strategy=HubStrategy.EVERY_SAVE,
hub_token=<HUB_TOKEN>,
ignore_data_skip=False,
include_for_metrics=[],
include_inputs_for_metrics=False,
include_num_input_tokens_seen=False,
include_tokens_per_second=False,
jit_mode_eval=False,
label_names=None,
label_smoothing_factor=0.0,
learning_rate=2e-05,
length_column_name=length,
load_best_model_at_end=False,
local_rank=0,
log_completions=False,
log_level=info,
log_level_replica=warning,
log_on_each_node=True,
logging_dir=output/Qwen2.5-1.5B-Open-R1-GRPO/runs/Feb18_14-33-03_smilelab-a6000,
logging_first_step=False,
logging_nan_inf_filter=True,
logging_steps=5,
logging_strategy=IntervalStrategy.STEPS,
lr_scheduler_kwargs={},
lr_scheduler_type=SchedulerType.COSINE,
max_completion_length=1024,
max_grad_norm=1.0,
max_prompt_length=512,
max_steps=-1,
metric_for_best_model=None,
model_init_kwargs=None,
mp_parameters=,
neftune_noise_alpha=None,
no_cuda=False,
num_generations=3,
num_train_epochs=1,
optim=OptimizerNames.ADAMW_TORCH,
optim_args=None,
optim_target_modules=None,
output_dir=output/Qwen2.5-1.5B-Open-R1-GRPO,
overwrite_hub_revision=False,
overwrite_output_dir=True,
past_index=-1,
per_device_eval_batch_size=32,
per_device_train_batch_size=16,
prediction_loss_only=False,
push_to_hub=True,
push_to_hub_model_id=None,
push_to_hub_organization=None,
push_to_hub_revision=False,
push_to_hub_token=<PUSH_TO_HUB_TOKEN>,
ray_scope=last,
ref_model_mixup_alpha=0.9,
ref_model_sync_steps=64,
remove_unused_columns=False,
report_to=['wandb'],
restore_callback_states_from_checkpoint=False,
resume_from_checkpoint=None,
reward_weights=None,
run_name=output/Qwen2.5-1.5B-Open-R1-GRPO,
save_on_each_node=False,
save_only_model=False,
save_safetensors=True,
save_steps=500,
save_strategy=SaveStrategy.NO,
save_total_limit=None,
seed=42,
skip_memory_metrics=True,
split_batches=None,
sync_ref_model=False,
system_prompt=None,
temperature=0.9,
tf32=None,
torch_compile=False,
torch_compile_backend=None,
torch_compile_mode=None,
torch_empty_cache_steps=None,
torchdynamo=None,
tpu_metrics_debug=False,
tpu_num_cores=None,
use_cpu=False,
use_ipex=False,
use_legacy_prediction_loop=False,
use_liger_kernel=False,
use_mps_device=False,
use_vllm=True,
vllm_device=auto,
vllm_dtype=auto,
vllm_gpu_memory_utilization=0.7,
vllm_max_model_len=None,
wandb_entity=None,
wandb_project=None,
warmup_ratio=0.1,
warmup_steps=0,
weight_decay=0.0,
)
2025-02-18 14:33:03 - WARNING - __main__ - Process rank: 1, device: cuda:1, n_gpu: 1 distributed training: True, 16-bits training: False
2025-02-18 14:33:03 - WARNING - __main__ - Process rank: 2, device: cuda:2, n_gpu: 1 distributed training: True, 16-bits training: False
2025-02-18 14:33:04 - INFO - datasets.builder - Overwrite dataset info from restored data version if exists.
2025-02-18 14:33:04 - INFO - datasets.info - Loading Dataset info from /home/mingyuan/.cache/huggingface/datasets/AI-MO___numina_math-tir/default/0.0.0/77a91d7b7a1a98ac4b1beb7d86c09d156b935dcd
2025-02-18 14:33:04 - INFO - datasets.builder - Found cached dataset numina_math-tir (/home/mingyuan/.cache/huggingface/datasets/AI-MO___numina_math-tir/default/0.0.0/77a91d7b7a1a98ac4b1beb7d86c09d156b935dcd)
2025-02-18 14:33:04 - INFO - datasets.info - Loading Dataset info from /home/mingyuan/.cache/huggingface/datasets/AI-MO___numina_math-tir/default/0.0.0/77a91d7b7a1a98ac4b1beb7d86c09d156b935dcd
2025-02-18 14:33:04 - INFO - datasets.arrow_dataset - Loading cached processed dataset at /home/mingyuan/.cache/huggingface/datasets/AI-MO___numina_math-tir/default/0.0.0/77a91d7b7a1a98ac4b1beb7d86c09d156b935dcd/cache-42e22e57818641cd.arrow
2025-02-18 14:33:04 - INFO - datasets.arrow_dataset - Loading cached processed dataset at /home/mingyuan/.cache/huggingface/datasets/AI-MO___numina_math-tir/default/0.0.0/77a91d7b7a1a98ac4b1beb7d86c09d156b935dcd/cache-be70039808739fd6.arrow
2025-02-18 14:33:04 - INFO - __main__ - *** Initializing model kwargs ***
2025-02-18 14:33:31 - INFO - __main__ - *** Train ***
2025-02-18 14:39:05 - WARNING - __main__ - Process rank: 0, device: cuda:0, n_gpu: 1 distributed training: True, 16-bits training: False
2025-02-18 14:39:05 - INFO - __main__ - Model parameters ModelConfig(model_name_or_path='Qwen/Qwen2.5-1.5B-Instruct', model_revision='main', torch_dtype='bfloat16', trust_remote_code=False, attn_implementation='flash_attention_2', use_peft=False, lora_r=16, lora_alpha=32, lora_dropout=0.05, lora_target_modules=None, lora_modules_to_save=None, lora_task_type='CAUSAL_LM', use_rslora=False, load_in_8bit=False, load_in_4bit=False, bnb_4bit_quant_type='nf4', use_bnb_nested_quant=False)
2025-02-18 14:39:05 - INFO - __main__ - Script parameters GRPOScriptArguments(dataset_name='AI-MO/NuminaMath-TIR', dataset_config=None, dataset_train_split='train', dataset_test_split='test', gradient_checkpointing_use_reentrant=False, ignore_bias_buffers=False, reward_funcs=['accuracy', 'format'], cosine_min_value_wrong=0.0, cosine_max_value_wrong=-0.5, cosine_min_value_correct=0.5, cosine_max_value_correct=1.0, cosine_max_len=1000, repetition_n_grams=3, repetition_max_penalty=-1.0)
2025-02-18 14:39:05 - INFO - __main__ - Training parameters GRPOConfig(
_n_gpu=1,
accelerator_config={'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None, 'use_configured_state': False},
adafactor=False,
adam_beta1=0.9,
adam_beta2=0.999,
adam_epsilon=1e-08,
auto_find_batch_size=False,
average_tokens_across_devices=False,
batch_eval_metrics=False,
benchmarks=[],
beta=0.04,
bf16=True,
bf16_full_eval=False,
callbacks=[],
data_seed=None,
dataloader_drop_last=False,
dataloader_num_workers=0,
dataloader_persistent_workers=False,
dataloader_pin_memory=True,
dataloader_prefetch_factor=None,
ddp_backend=None,
ddp_broadcast_buffers=None,
ddp_bucket_cap_mb=None,
ddp_find_unused_parameters=None,
ddp_timeout=1800,
debug=[],
deepspeed=None,
disable_tqdm=False,
dispatch_batches=None,
do_eval=True,
do_predict=False,
do_train=False,
ds3_gather_for_generation=True,
eval_accumulation_steps=None,
eval_delay=0,
eval_do_concat_batches=True,
eval_on_start=False,
eval_steps=100,
eval_strategy=IntervalStrategy.STEPS,
eval_use_gather_object=False,
evaluation_strategy=None,
fp16=False,
fp16_backend=auto,
fp16_full_eval=False,
fp16_opt_level=O1,
fsdp=[],
fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False},
fsdp_min_num_params=0,
fsdp_transformer_layer_cls_to_wrap=None,
full_determinism=False,
gradient_accumulation_steps=16,
gradient_checkpointing=True,
gradient_checkpointing_kwargs={'use_reentrant': False},
greater_is_better=None,
group_by_length=False,
half_precision_backend=auto,
hub_always_push=False,
hub_model_id=Qwen2.5-1.5B-Open-R1-GRPO,
hub_model_revision=main,
hub_private_repo=None,
hub_strategy=HubStrategy.EVERY_SAVE,
hub_token=<HUB_TOKEN>,
ignore_data_skip=False,
include_for_metrics=[],
include_inputs_for_metrics=False,
include_num_input_tokens_seen=False,
include_tokens_per_second=False,
jit_mode_eval=False,
label_names=None,
label_smoothing_factor=0.0,
learning_rate=2e-05,
length_column_name=length,
load_best_model_at_end=False,
local_rank=0,
log_completions=False,
log_level=info,
log_level_replica=warning,
log_on_each_node=True,
logging_dir=output/Qwen2.5-1.5B-Open-R1-GRPO/runs/Feb18_14-39-05_smilelab-a6000,
logging_first_step=False,
logging_nan_inf_filter=True,
logging_steps=5,
logging_strategy=IntervalStrategy.STEPS,
lr_scheduler_kwargs={},
lr_scheduler_type=SchedulerType.COSINE,
max_completion_length=1024,
max_grad_norm=1.0,
max_prompt_length=512,
max_steps=-1,
metric_for_best_model=None,
model_init_kwargs=None,
mp_parameters=,
neftune_noise_alpha=None,
no_cuda=False,
num_generations=2,
num_train_epochs=1,
optim=OptimizerNames.ADAMW_TORCH,
optim_args=None,
optim_target_modules=None,
output_dir=output/Qwen2.5-1.5B-Open-R1-GRPO,
overwrite_hub_revision=False,
overwrite_output_dir=True,
past_index=-1,
per_device_eval_batch_size=32,
per_device_train_batch_size=16,
prediction_loss_only=False,
push_to_hub=True,
push_to_hub_model_id=None,
push_to_hub_organization=None,
push_to_hub_revision=False,
push_to_hub_token=<PUSH_TO_HUB_TOKEN>,
ray_scope=last,
ref_model_mixup_alpha=0.9,
ref_model_sync_steps=64,
remove_unused_columns=False,
report_to=['wandb'],
restore_callback_states_from_checkpoint=False,
resume_from_checkpoint=None,
reward_weights=None,
run_name=output/Qwen2.5-1.5B-Open-R1-GRPO,
save_on_each_node=False,
save_only_model=False,
save_safetensors=True,
save_steps=500,
save_strategy=SaveStrategy.NO,
save_total_limit=None,
seed=42,
skip_memory_metrics=True,
split_batches=None,
sync_ref_model=False,
system_prompt=None,
temperature=0.9,
tf32=None,
torch_compile=False,
torch_compile_backend=None,
torch_compile_mode=None,
torch_empty_cache_steps=None,
torchdynamo=None,
tpu_metrics_debug=False,
tpu_num_cores=None,
use_cpu=False,
use_ipex=False,
use_legacy_prediction_loop=False,
use_liger_kernel=False,
use_mps_device=False,
use_vllm=True,
vllm_device=auto,
vllm_dtype=auto,
vllm_gpu_memory_utilization=0.7,
vllm_max_model_len=None,
wandb_entity=None,
wandb_project=None,
warmup_ratio=0.1,
warmup_steps=0,
weight_decay=0.0,
)
2025-02-18 14:39:05 - WARNING - __main__ - Process rank: 1, device: cuda:1, n_gpu: 1 distributed training: True, 16-bits training: False
2025-02-18 14:39:07 - INFO - datasets.builder - Overwrite dataset info from restored data version if exists.
2025-02-18 14:39:07 - INFO - datasets.info - Loading Dataset info from /home/mingyuan/.cache/huggingface/datasets/AI-MO___numina_math-tir/default/0.0.0/77a91d7b7a1a98ac4b1beb7d86c09d156b935dcd
2025-02-18 14:39:07 - INFO - datasets.builder - Found cached dataset numina_math-tir (/home/mingyuan/.cache/huggingface/datasets/AI-MO___numina_math-tir/default/0.0.0/77a91d7b7a1a98ac4b1beb7d86c09d156b935dcd)
2025-02-18 14:39:07 - INFO - datasets.info - Loading Dataset info from /home/mingyuan/.cache/huggingface/datasets/AI-MO___numina_math-tir/default/0.0.0/77a91d7b7a1a98ac4b1beb7d86c09d156b935dcd
2025-02-18 14:39:07 - INFO - datasets.arrow_dataset - Loading cached processed dataset at /home/mingyuan/.cache/huggingface/datasets/AI-MO___numina_math-tir/default/0.0.0/77a91d7b7a1a98ac4b1beb7d86c09d156b935dcd/cache-42e22e57818641cd.arrow
2025-02-18 14:39:07 - INFO - datasets.arrow_dataset - Loading cached processed dataset at /home/mingyuan/.cache/huggingface/datasets/AI-MO___numina_math-tir/default/0.0.0/77a91d7b7a1a98ac4b1beb7d86c09d156b935dcd/cache-be70039808739fd6.arrow
2025-02-18 14:39:07 - INFO - __main__ - *** Initializing model kwargs ***
2025-02-18 14:39:33 - INFO - __main__ - *** Train ***
2025-02-19 03:54:32 - INFO - __main__ - *** Save model ***