Qwen/Qwen-72B_eval_request_False_float16_Original.json

{
    "model": "Qwen/Qwen-72B",
    "base_model": "",
    "revision": "main",
    "private": false,
    "precision": "float16",
    "params": 72.288,
    "architectures": "QWenLMHeadModel",
    "weight_type": "Original",
    "status": "FAILED",
    "submitted_time": "2024-04-18T23:08:40Z",
    "model_type": "🟢 : pretrained",
    "source": "leaderboard",
    "job_id": 615,
    "job_start_time": "2024-05-16T07-23-30.896151",
    "main_language": "Chinese",
    "error_msg": "CUDA out of memory. Tried to allocate 384.00 MiB. GPU 0 has a total capacty of 79.35 GiB of which 132.19 MiB is free. Process 589465 has 79.21 GiB memory in use. Of the allocated memory 78.71 GiB is allocated by PyTorch, and 12.14 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF",
    "traceback": "Traceback (most recent call last):\n  File \"/workspace/repos/llm_leaderboard/llm_leaderboard_eval_bot/evaluate_llms.py\", line 207, in wait_download_and_run_request\n    run_request(\n  File \"/workspace/repos/llm_leaderboard/llm_leaderboard_eval_bot/evaluate_llms.py\", line 69, in run_request\n    results = run_eval_on_model(\n              ^^^^^^^^^^^^^^^^^^\n  File \"/workspace/repos/llm_leaderboard/llm_leaderboard_eval_bot/run_eval.py\", line 60, in run_eval_on_model\n    result = evaluate(\n             ^^^^^^^^^\n  File \"/workspace/repos/llm_leaderboard/llm_leaderboard_eval_bot/lm_eval_util.py\", line 145, in evaluate\n    results = evaluator.simple_evaluate(\n              ^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/workspace/repos/llm_leaderboard/lm-evaluation-harness-pt/lm_eval/utils.py\", line 419, in _wrapper\n    return fn(*args, **kwargs)\n           ^^^^^^^^^^^^^^^^^^^\n  File \"/workspace/repos/llm_leaderboard/lm-evaluation-harness-pt/lm_eval/evaluator.py\", line 100, in simple_evaluate\n    lm = lm_eval.api.registry.get_model(model).create_from_arg_string(\n         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/workspace/repos/llm_leaderboard/lm-evaluation-harness-pt/lm_eval/api/model.py\", line 134, in create_from_arg_string\n    return cls(**args, **args2)\n           ^^^^^^^^^^^^^^^^^^^^\n  File \"/workspace/repos/llm_leaderboard/lm-evaluation-harness-pt/lm_eval/models/huggingface.py\", line 297, in __init__\n    self._create_model(\n  File \"/workspace/repos/llm_leaderboard/lm-evaluation-harness-pt/lm_eval/models/huggingface.py\", line 608, in _create_model\n    self._model = self.AUTO_MODEL_CLASS.from_pretrained(\n                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/workspace/repos/llm_leaderboard/transformers/src/transformers/models/auto/auto_factory.py\", line 558, in from_pretrained\n    return model_class.from_pretrained(\n           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/workspace/repos/llm_leaderboard/transformers/src/transformers/modeling_utils.py\", line 3682, in from_pretrained\n    ) = cls._load_pretrained_model(\n        ^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/workspace/repos/llm_leaderboard/transformers/src/transformers/modeling_utils.py\", line 4109, in _load_pretrained_model\n    new_error_msgs, offload_index, state_dict_index = _load_state_dict_into_meta_model(\n                                                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File \"/workspace/repos/llm_leaderboard/transformers/src/transformers/modeling_utils.py\", line 887, in _load_state_dict_into_meta_model\n    set_module_tensor_to_device(model, param_name, param_device, **set_module_kwargs)\n  File \"/root/miniconda3/envs/torch21/lib/python3.11/site-packages/accelerate/utils/modeling.py\", line 399, in set_module_tensor_to_device\n    new_value = value.to(device)\n                ^^^^^^^^^^^^^^^^\ntorch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 384.00 MiB. GPU 0 has a total capacty of 79.35 GiB of which 132.19 MiB is free. Process 589465 has 79.21 GiB memory in use. Of the allocated memory 78.71 GiB is allocated by PyTorch, and 12.14 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF\n"
}