2025-08-17 08:12:43 - eval_util - INFO - Using AWQ quantization.
2025-08-17 08:12:43 - eval_util - INFO - Quantizing activation with bitwidth 4
2025-08-17 08:12:44 - accelerate.utils.modeling - INFO - We will use 90% of the memory on device 0 for storing the model, and 10% for the buffer to avoid OOM. You can set `max_memory` in to a higher value to use more memory (at your own risk).
2025-08-17 08:16:05 - accelerate.utils.modeling - INFO - We will use 90% of the memory on device 0 for storing the model, and 10% for the buffer to avoid OOM. You can set `max_memory` in to a higher value to use more memory (at your own risk).
2025-08-17 08:16:05 - eval_util - INFO - Evaluating wikitext...
2025-08-17 08:17:14 - eval_util - INFO - Wikitext completed: 6.425759315490723
2025-08-17 08:17:14 - eval_util - INFO - Starting evaluation of task: boolq
2025-08-17 08:17:14 - eval_util - INFO - GPU Memory before boolq: 3.81GB
2025-08-17 08:17:14 - lm_eval.models.huggingface - WARNING - `pretrained` model kwarg is not of type `str`. Many other model arguments may be ignored. Please do not launch via accelerate or use `parallelize=True` if passing an existing model this way.
2025-08-17 08:17:14 - lm_eval.models.huggingface - WARNING - Passed an already-initialized model through `pretrained`, assuming single-process call to evaluate() or custom distributed integration
2025-08-17 08:17:14 - lm_eval.evaluator - INFO - Setting random seed to 0 | Setting numpy seed to 1234 | Setting torch manual seed to 1234 | Setting fewshot manual seed to 1234
2025-08-17 08:17:14 - lm_eval.evaluator - INFO - Using pre-initialized model
2025-08-17 08:17:29 - lm_eval.api.task - WARNING - [Task: boolq] metric acc is defined, but aggregation is not. using default aggregation=mean
2025-08-17 08:17:29 - lm_eval.api.task - WARNING - [Task: boolq] metric acc is defined, but higher_is_better is not. using default higher_is_better=True
2025-08-17 08:17:33 - lm_eval.api.task - INFO - Building contexts for boolq on rank 0...
2025-08-17 08:17:34 - lm_eval.evaluator - INFO - Running loglikelihood requests
2025-08-17 08:21:08 - eval_util - INFO - ✅ Task boolq completed successfully
2025-08-17 08:21:08 - eval_util - INFO - GPU Memory after boolq: 3.81GB
2025-08-17 08:21:08 - eval_util - INFO - Starting evaluation of task: hellaswag
2025-08-17 08:21:08 - eval_util - INFO - GPU Memory before hellaswag: 3.81GB
2025-08-17 08:21:08 - lm_eval.models.huggingface - WARNING - `pretrained` model kwarg is not of type `str`. Many other model arguments may be ignored. Please do not launch via accelerate or use `parallelize=True` if passing an existing model this way.
2025-08-17 08:21:08 - lm_eval.models.huggingface - WARNING - Passed an already-initialized model through `pretrained`, assuming single-process call to evaluate() or custom distributed integration
2025-08-17 08:21:08 - lm_eval.evaluator - INFO - Setting random seed to 0 | Setting numpy seed to 1234 | Setting torch manual seed to 1234 | Setting fewshot manual seed to 1234
2025-08-17 08:21:08 - lm_eval.evaluator - INFO - Using pre-initialized model
2025-08-17 08:21:26 - lm_eval.api.task - INFO - Building contexts for hellaswag on rank 0...
2025-08-17 08:21:30 - lm_eval.evaluator - INFO - Running loglikelihood requests
2025-08-17 08:55:10 - eval_util - INFO - ✅ Task hellaswag completed successfully
2025-08-17 08:55:10 - eval_util - INFO - GPU Memory after hellaswag: 3.81GB
2025-08-17 08:55:10 - eval_util - INFO - Starting evaluation of task: winogrande
2025-08-17 08:55:10 - eval_util - INFO - GPU Memory before winogrande: 3.81GB
2025-08-17 08:55:10 - lm_eval.models.huggingface - WARNING - `pretrained` model kwarg is not of type `str`. Many other model arguments may be ignored. Please do not launch via accelerate or use `parallelize=True` if passing an existing model this way.
2025-08-17 08:55:10 - lm_eval.models.huggingface - WARNING - Passed an already-initialized model through `pretrained`, assuming single-process call to evaluate() or custom distributed integration
2025-08-17 08:55:10 - lm_eval.evaluator - INFO - Setting random seed to 0 | Setting numpy seed to 1234 | Setting torch manual seed to 1234 | Setting fewshot manual seed to 1234
2025-08-17 08:55:10 - lm_eval.evaluator - INFO - Using pre-initialized model
2025-08-17 08:55:27 - lm_eval.api.task - INFO - Building contexts for winogrande on rank 0...
2025-08-17 08:55:27 - lm_eval.evaluator - INFO - Running loglikelihood requests
2025-08-17 08:57:16 - eval_util - INFO - ✅ Task winogrande completed successfully
2025-08-17 08:57:16 - eval_util - INFO - GPU Memory after winogrande: 3.81GB
2025-08-17 08:57:16 - eval_util - INFO - Starting evaluation of task: arc_easy
2025-08-17 08:57:16 - eval_util - INFO - GPU Memory before arc_easy: 3.81GB
2025-08-17 08:57:16 - lm_eval.models.huggingface - WARNING - `pretrained` model kwarg is not of type `str`. Many other model arguments may be ignored. Please do not launch via accelerate or use `parallelize=True` if passing an existing model this way.
2025-08-17 08:57:16 - lm_eval.models.huggingface - WARNING - Passed an already-initialized model through `pretrained`, assuming single-process call to evaluate() or custom distributed integration
2025-08-17 08:57:16 - lm_eval.evaluator - INFO - Setting random seed to 0 | Setting numpy seed to 1234 | Setting torch manual seed to 1234 | Setting fewshot manual seed to 1234
2025-08-17 08:57:16 - lm_eval.evaluator - INFO - Using pre-initialized model
2025-08-17 08:57:30 - lm_eval.api.task - INFO - Building contexts for arc_easy on rank 0...
2025-08-17 08:57:32 - lm_eval.evaluator - INFO - Running loglikelihood requests
2025-08-17 09:04:03 - eval_util - INFO - ✅ Task arc_easy completed successfully
2025-08-17 09:04:03 - eval_util - INFO - GPU Memory after arc_easy: 3.81GB
2025-08-17 09:04:03 - eval_util - INFO - Starting evaluation of task: arc_challenge
2025-08-17 09:04:03 - eval_util - INFO - GPU Memory before arc_challenge: 3.81GB
2025-08-17 09:04:03 - lm_eval.models.huggingface - WARNING - `pretrained` model kwarg is not of type `str`. Many other model arguments may be ignored. Please do not launch via accelerate or use `parallelize=True` if passing an existing model this way.
2025-08-17 09:04:03 - lm_eval.models.huggingface - WARNING - Passed an already-initialized model through `pretrained`, assuming single-process call to evaluate() or custom distributed integration
2025-08-17 09:04:03 - lm_eval.evaluator - INFO - Setting random seed to 0 | Setting numpy seed to 1234 | Setting torch manual seed to 1234 | Setting fewshot manual seed to 1234
2025-08-17 09:04:03 - lm_eval.evaluator - INFO - Using pre-initialized model
2025-08-17 09:04:16 - lm_eval.api.task - INFO - Building contexts for arc_challenge on rank 0...
2025-08-17 09:04:17 - lm_eval.evaluator - INFO - Running loglikelihood requests
2025-08-17 09:07:40 - eval_util - INFO - ✅ Task arc_challenge completed successfully
2025-08-17 09:07:40 - eval_util - INFO - GPU Memory after arc_challenge: 3.81GB
2025-08-17 09:07:40 - eval_util - INFO - Evaluation Summary - Successful: 6, Failed: 0
2025-08-17 09:07:40 - eval_util - INFO - result:
2025-08-17 09:07:40 - eval_util - INFO - 	{'arc_challenge': {'acc,none': 0.4189419795221843,
2025-08-17 09:07:40 - eval_util - INFO - 	                   'acc_norm,none': 0.4462457337883959,
2025-08-17 09:07:40 - eval_util - INFO - 	                   'acc_norm_stderr,none': 0.01452670554853998,
2025-08-17 09:07:40 - eval_util - INFO - 	                   'acc_stderr,none': 0.014418106953639008,
2025-08-17 09:07:40 - eval_util - INFO - 	                   'alias': 'arc_challenge'},
2025-08-17 09:07:40 - eval_util - INFO - 	 'arc_easy': {'acc,none': 0.734006734006734,
2025-08-17 09:07:40 - eval_util - INFO - 	              'acc_norm,none': 0.6973905723905723,
2025-08-17 09:07:40 - eval_util - INFO - 	              'acc_norm_stderr,none': 0.009426434542371223,
2025-08-17 09:07:40 - eval_util - INFO - 	              'acc_stderr,none': 0.009066789565615694,
2025-08-17 09:07:40 - eval_util - INFO - 	              'alias': 'arc_easy'},
2025-08-17 09:07:40 - eval_util - INFO - 	 'boolq': {'acc,none': 0.7318042813455657,
2025-08-17 09:07:40 - eval_util - INFO - 	           'acc_stderr,none': 0.007748469592030341,
2025-08-17 09:07:40 - eval_util - INFO - 	           'alias': 'boolq'},
2025-08-17 09:07:40 - eval_util - INFO - 	 'hellaswag': {'acc,none': 0.55646285600478,
2025-08-17 09:07:40 - eval_util - INFO - 	               'acc_norm,none': 0.7457677753435571,
2025-08-17 09:07:40 - eval_util - INFO - 	               'acc_norm_stderr,none': 0.004345388614519977,
2025-08-17 09:07:40 - eval_util - INFO - 	               'acc_stderr,none': 0.004957863944093127,
2025-08-17 09:07:40 - eval_util - INFO - 	               'alias': 'hellaswag'},
2025-08-17 09:07:40 - eval_util - INFO - 	 'wikitext': 6.425759315490723,
2025-08-17 09:07:40 - eval_util - INFO - 	 'winogrande': {'acc,none': 0.6692975532754538,
2025-08-17 09:07:40 - eval_util - INFO - 	                'acc_stderr,none': 0.013222435887002683,
2025-08-17 09:07:40 - eval_util - INFO - 	                'alias': 'winogrande'}}
