Closed yxm0513 closed 6 months ago
A GPU with more than 24GB of Cuda memory is required. If you feel that the cuda memory is not enough, you can try using frameworks like vllm, deepseed, etc. to run the Qwen model.
I have not tested the model on a CPU, so I'm not sure if there would be any impact. I suggest carefully checking the output files and the generated content from the model to determine if the model has generated appropriate content.
If you need experiment results, please contact the email address s1583050085@gmail.com.
1 GPU Memery issue (CUDA out of memory.)
Output saved at ./output/test_top2_Qwen_7B_Chat/ContinueWriting_qwen7b.json!
0%| | 0/5 [00:00<?, ?it/s]2024-04-03 14:51:50.480 | WARNING | src.llms.base:safe_request:43 - OutOfMemoryError('CUDA out of memory. Tried to allocate 104.00 MiB (GPU 1; 7.92 GiB total capacity; 7.74 GiB already allocated; 7.69 MiB free; 7.75 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF')
2024-04-03 14:51:50.498 | WARNING | src.llms.base:safe_request:43 - OutOfMemoryError('CUDA out of memory. Tried to allocate 16.00 MiB (GPU 1; 7.92 GiB total capacity; 7.74 GiB already allocated; 7.69 MiB free; 7.75 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF')
2 swith to CPU, does ZeroDivisionError impacts final results?
100%|██████████| 5/5 [07:06<00:00, 85.39s/it] Output saved at ./output/test_top2_Qwen_7B_Chat/Summary_qwen7b.json! 100%|██████████| 5/5 [30:58<00:00, 371.78s/it] Output saved at ./output/test_top2_Qwen_7B_Chat/ContinueWriting_qwen7b.json! 0%| | 0/5 [00:00<?, ?it/s]2024-04-03 11:19:09.803 | WARNING | src.metric.common:wrapper:19 - ZeroDivisionError('float division by zero') 2024-04-03 11:19:09.803 | WARNING | evaluator:process_data_point:94 - TypeError('cannot unpack non-iterable NoneType object') 80%|████████ | 4/5 [1:04:08<10:57, 657.84s/it]2024-04-03 11:19:50.945 | WARNING | src.metric.common:wrapper:19 - ZeroDivisionError('float division by zero') 2024-04-03 11:19:50.946 | WARNING | evaluator:process_data_point:94 - TypeError('cannot unpack non-iterable NoneType object') 100%|██████████| 5/5 [1:04:09<00:00, 769.96s/it] Output saved at ./output/test_top2_Qwen_7B_Chat/HalluModified_qwen7b.json! 0%| | 0/5 [00:00<?, ?it/s]2024-04-03 12:07:17.097 | WARNING | src.metric.common:wrapper:19 - ZeroDivisionError('float division by zero') 2024-04-03 12:07:17.098 | WARNING | evaluator:process_data_point:94 - TypeError('cannot unpack non-iterable NoneType object') 100%|██████████| 5/5 [47:27<00:00, 569.47s/it] Output saved at ./output/test_top2_Qwen_7B_Chat/QuestAnswer1Doc_qwen7b.json! 0%| | 0/5 [00:00<?, ?it/s]2024-04-03 12:50:40.766 | WARNING | src.metric.common:wrapper:19 - ZeroDivisionError('float division by zero') 2024-04-03 12:50:40.767 | WARNING | evaluator:process_data_point:94 - TypeError('cannot unpack non-iterable NoneType object') 100%|██████████| 5/5 [43:21<00:00, 520.36s/it] Output saved at ./output/test_top2_Qwen_7B_Chat/QuestAnswer2Docs_qwen7b.json! 20%|██ | 1/5 [37:01<2:28:04, 2221.07s/it]2024-04-03 13:34:40.221 | WARNING | src.metric.common:wrapper:19 - ZeroDivisionError('float division by zero') 2024-04-03 13:34:40.222 | WARNING | evaluator:process_data_point:94 - TypeError('cannot unpack non-iterable NoneType object') 100%|██████████| 5/5 [43:52<00:00, 526.53s/it] Output saved at ./output/test_top2_Qwen_7B_Chat/QuestAnswer3Docs_qwen7b.json!
3 do you have results for models, like qwen, gpt? where to fetech it?