########## m0 / mmlu ########## /usr/bin/python -m lm_eval --model hf --model_args pretrained=Qwen/Qwen2.5-7B-Instruct,dtype=bfloat16 --tasks mmlu --num_fewshot 5 --batch_size auto --seed 0 --output_path /root/steering-resistance/results/full_7b/capability/m0/mmlu --apply_chat_template --fewshot_as_multiturn --limit 15 /usr/local/lib/python3.11/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.2.3) or chardet (6.0.0.post1)/charset_normalizer (3.3.2) doesn't match a supported version! warnings.warn( 2026-07-24:06:59:32 WARNING [config.evaluate_config:287] --limit SHOULD ONLY BE USED FOR TESTING. REAL METRICS SHOULD NOT BE COMPUTED USING LIMIT. 2026-07-24:06:59:40 INFO [_cli.run:388] Selected Tasks: ['mmlu'] 2026-07-24:06:59:42 INFO [evaluator:214] Setting random seed to 0 | Setting numpy seed to 0 | Setting torch manual seed to 0 | Setting fewshot manual seed to 0 2026-07-24:06:59:42 INFO [evaluator:239] Initializing hf model, with arguments: {'pretrained': 'Qwen/Qwen2.5-7B-Instruct', 'dtype': 'bfloat16'} 2026-07-24:06:59:45 INFO [models.huggingface:286] Using device 'cuda:0' 2026-07-24:06:59:47 INFO [models.huggingface:579] Model parallel was set to False, max memory was not set, and device map was set to {'': 'cuda:0'} Loading weights: 0%| | 0/339 [00:00", line 198, in _run_module_as_main File "", line 88, in _run_code File "/usr/local/lib/python3.11/dist-packages/lm_eval/__main__.py", line 14, in cli_evaluate() File "/usr/local/lib/python3.11/dist-packages/lm_eval/__main__.py", line 10, in cli_evaluate parser.execute(args) File "/usr/local/lib/python3.11/dist-packages/lm_eval/_cli/harness.py", line 60, in execute args.func(args) File "/usr/local/lib/python3.11/dist-packages/lm_eval/_cli/run.py", line 391, in _execute results = simple_evaluate( ^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/utils.py", line 575, in _wrapper return fn(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/evaluator.py", line 358, in simple_evaluate results = evaluate( ^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/utils.py", line 575, in _wrapper return fn(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/evaluator.py", line 596, in evaluate resps = getattr(lm, reqtype)(cloned_reqs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/models/huggingface.py", line 1227, in loglikelihood return super().loglikelihood(requests, disable_tqdm=disable_tqdm) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/api/model.py", line 446, in loglikelihood return self._loglikelihood_tokens(new_reqs, disable_tqdm=disable_tqdm) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/models/huggingface.py", line 1507, in _loglikelihood_tokens self._model_call(batched_inps, **call_kwargs), ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/models/huggingface.py", line 1154, in _model_call return self.model(inps).logits ^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/torch/nn/modules/module.py", line 1778, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/torch/nn/modules/module.py", line 1789, in _call_impl return forward_call(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/transformers/utils/generic.py", line 911, in wrapper output = func(self, *args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/transformers/models/qwen2/modeling_qwen2.py", line 475, in forward logits = self.lm_head(hidden_states[:, slice_indices, :]) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/torch/nn/modules/module.py", line 1778, in _wrapped_call_impl return self._call_impl(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/torch/nn/modules/module.py", line 1789, in _call_impl return forward_call(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/torch/nn/modules/linear.py", line 134, in forward return F.linear(input, self.weight, self.bias) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 3.34 GiB. GPU 0 has a total capacity of 23.58 GiB of which 2.15 GiB is free. Process 1398907 has 21.42 GiB memory in use. Of the allocated memory 18.36 GiB is allocated by PyTorch, and 2.76 GiB 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://docs.pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf) Running loglikelihood requests: 0%| | 16/3420 [00:06<22:48, 2.49it/s] ########## m0 / gsm8k_cot ########## /usr/bin/python -m lm_eval --model hf --model_args pretrained=Qwen/Qwen2.5-7B-Instruct,dtype=bfloat16 --tasks gsm8k_cot --num_fewshot 8 --batch_size 4 --seed 0 --output_path /root/steering-resistance/results/full_7b/capability/m0/gsm8k_cot --gen_kwargs do_sample=False --apply_chat_template --fewshot_as_multiturn --limit 200 /usr/local/lib/python3.11/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.2.3) or chardet (6.0.0.post1)/charset_normalizer (3.3.2) doesn't match a supported version! warnings.warn( 2026-07-24:07:07:13 WARNING [config.evaluate_config:287] --limit SHOULD ONLY BE USED FOR TESTING. REAL METRICS SHOULD NOT BE COMPUTED USING LIMIT. 2026-07-24:07:07:21 INFO [_cli.run:388] Selected Tasks: ['gsm8k_cot'] 2026-07-24:07:07:23 INFO [evaluator:214] Setting random seed to 0 | Setting numpy seed to 0 | Setting torch manual seed to 0 | Setting fewshot manual seed to 0 2026-07-24:07:07:23 WARNING [evaluator:226] generation_kwargs: {'do_sample': False} specified through cli, these settings will update set parameters in yaml tasks. Ensure 'do_sample=True' for non-greedy decoding! 2026-07-24:07:07:23 INFO [evaluator:239] Initializing hf model, with arguments: {'pretrained': 'Qwen/Qwen2.5-7B-Instruct', 'dtype': 'bfloat16'} 2026-07-24:07:07:26 INFO [models.huggingface:286] Using device 'cuda:0' 2026-07-24:07:07:27 INFO [models.huggingface:579] Model parallel was set to False, max memory was not set, and device map was set to {'': 'cuda:0'} Loading weights: 0%| | 0/339 [00:00', '<|im_end|>']} 2026-07-24:07:07:35 WARNING [evaluator:333] Overwriting default num_fewshot of gsm8k_cot from 8 to 8 2026-07-24:07:07:35 INFO [api.task:312] Building contexts for gsm8k_cot on rank 0... 0%| | 0/200 [00:00", line 198, in _run_module_as_main File "", line 88, in _run_code File "/usr/local/lib/python3.11/dist-packages/lm_eval/__main__.py", line 14, in cli_evaluate() File "/usr/local/lib/python3.11/dist-packages/lm_eval/__main__.py", line 10, in cli_evaluate parser.execute(args) File "/usr/local/lib/python3.11/dist-packages/lm_eval/_cli/harness.py", line 60, in execute args.func(args) File "/usr/local/lib/python3.11/dist-packages/lm_eval/_cli/run.py", line 391, in _execute results = simple_evaluate( ^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/utils.py", line 575, in _wrapper return fn(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/evaluator.py", line 358, in simple_evaluate results = evaluate( ^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/utils.py", line 575, in _wrapper return fn(*args, **kwargs) ^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/evaluator.py", line 596, in evaluate resps = getattr(lm, reqtype)(cloned_reqs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/models/huggingface.py", line 1227, in loglikelihood return super().loglikelihood(requests, disable_tqdm=disable_tqdm) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/api/model.py", line 446, in loglikelihood return self._loglikelihood_tokens(new_reqs, disable_tqdm=disable_tqdm) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/lm_eval/models/huggingface.py", line 1506, in _loglikelihood_tokens multi_logits = F.log_softmax( ^^^^^^^^^^^^^^ File "/usr/local/lib/python3.11/dist-packages/torch/nn/functional.py", line 2315, in log_softmax ret = input.log_softmax(dim) ^^^^^^^^^^^^^^^^^^^^^^ torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 3.34 GiB. GPU 0 has a total capacity of 23.58 GiB of which 1.10 GiB is free. Process 1401096 has 22.47 GiB memory in use. Of the allocated memory 21.12 GiB is allocated by PyTorch, and 1.05 GiB 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://docs.pytorch.org/docs/stable/notes/cuda.html#optimizing-memory-usage-with-pytorch-cuda-alloc-conf) Running loglikelihood requests: 0%| | 16/3420 [00:06<23:42, 2.39it/s] ########## m1 / gsm8k_cot ########## /usr/bin/python -m lm_eval --model hf --model_args pretrained=Qwen/Qwen2.5-7B-Instruct,dtype=bfloat16,peft=/root/steering-resistance/results/full_7b/m1_resist_adapter --tasks gsm8k_cot --num_fewshot 8 --batch_size 4 --seed 0 --output_path /root/steering-resistance/results/full_7b/capability/m1/gsm8k_cot --gen_kwargs do_sample=False --apply_chat_template --fewshot_as_multiturn --limit 200 /usr/local/lib/python3.11/dist-packages/requests/__init__.py:113: RequestsDependencyWarning: urllib3 (2.2.3) or chardet (6.0.0.post1)/charset_normalizer (3.3.2) doesn't match a supported version! warnings.warn( 2026-07-24:07:15:31 WARNING [config.evaluate_config:287] --limit SHOULD ONLY BE USED FOR TESTING. REAL METRICS SHOULD NOT BE COMPUTED USING LIMIT. 2026-07-24:07:15:40 INFO [_cli.run:388] Selected Tasks: ['gsm8k_cot'] 2026-07-24:07:15:41 INFO [evaluator:214] Setting random seed to 0 | Setting numpy seed to 0 | Setting torch manual seed to 0 | Setting fewshot manual seed to 0 2026-07-24:07:15:41 WARNING [evaluator:226] generation_kwargs: {'do_sample': False} specified through cli, these settings will update set parameters in yaml tasks. Ensure 'do_sample=True' for non-greedy decoding! 2026-07-24:07:15:41 INFO [evaluator:239] Initializing hf model, with arguments: {'pretrained': 'Qwen/Qwen2.5-7B-Instruct', 'dtype': 'bfloat16', 'peft': '/root/steering-resistance/results/full_7b/m1_resist_adapter'} 2026-07-24:07:15:44 INFO [models.huggingface:286] Using device 'cuda:0' 2026-07-24:07:15:46 INFO [models.huggingface:579] Model parallel was set to False, max memory was not set, and device map was set to {'': 'cuda:0'} Loading weights: 0%| | 0/339 [00:00', '<|im_end|>']} 2026-07-24:07:15:53 WARNING [evaluator:333] Overwriting default num_fewshot of gsm8k_cot from 8 to 8 2026-07-24:07:15:53 INFO [api.task:312] Building contexts for gsm8k_cot on rank 0... 0%| | 0/200 [00:00