Why can't methods like recycling and Lookahead run with multiple GPUs? Once multiple GPUs are used, errors start to appear from the very first question.
Check model training state: False CUDA VISIBLE DEVICES: 0,1,2 ERROR question ID: 81 Traceback (most recent call last): File "<frozen runpy>", line 198, in _run_module_as_main File "<frozen runpy>", line 88, in _run_code File "/home/shengjinhao/Spec-Bench/evaluation/inference_recycling.py", line 244, in <module> run_eval( File "/home/shengjinhao/Spec-Bench/evaluation/eval.py", line 54, in run_eval get_answers_func( File "/workspace/conda/envs/specbench/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context return func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/home/shengjinhao/Spec-Bench/evaluation/eval.py", line 152, in get_model_answers steps.append(int(step)) ^^^^ UnboundLocalError: cannot access local variable 'step' where it is not associated with a value
Why can't methods like recycling and Lookahead run with multiple GPUs? Once multiple GPUs are used, errors start to appear from the very first question.
Check model training state: False CUDA VISIBLE DEVICES: 0,1,2 ERROR question ID: 81 Traceback (most recent call last): File "<frozen runpy>", line 198, in _run_module_as_main File "<frozen runpy>", line 88, in _run_code File "/home/shengjinhao/Spec-Bench/evaluation/inference_recycling.py", line 244, in <module> run_eval( File "/home/shengjinhao/Spec-Bench/evaluation/eval.py", line 54, in run_eval get_answers_func( File "/workspace/conda/envs/specbench/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context return func(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^ File "/home/shengjinhao/Spec-Bench/evaluation/eval.py", line 152, in get_model_answers steps.append(int(step)) ^^^^ UnboundLocalError: cannot access local variable 'step' where it is not associated with a value