华为计算微信公众号
昇腾AI开发者公众号
华为计算微博
华为计算今日头条
求助,通过mindie:2.0.RC1-300I-Duo-py311-openeuler24.03-lts镜像使用300IDuo 4卡8芯推理Qwen3-30B-A3B时,推理服务一直起不来。日志如下面所示,一直卡在这。且期间服务器CPU内存持续占用,会把CPU内存打满,导致服务器宕机。
感觉加载模型这一环节有什么问题,但是定位不到根本原因。
有大佬遇过类似问题吗?
2025-05-23 15:07:16.394 14548 LLM log default format: [yyyy-mm-dd hh:mm:ss.uuuuuu] [processid] [threadid] [llm] [loglevel] [file:line] [status code] msg [2025-05-23 15:07:16.394] [13041] [281464972439776] [llm] [INFO] [post_processing_manager.cpp:161] Get post processing manager [2025-05-23 15:07:16.395] [13041] [281464972439776] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281463798739168 [2025-05-23 15:07:16.395] [13041] [281464972439776] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472020246752 [2025-05-23 15:07:16.395] [13041] [281464972439776] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472011792608 [2025-05-23 15:07:16.395] [13041] [281464972439776] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472003338464 [2025-05-23 15:07:16.395] [13041] [281464972439776] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281471994884320 [2025-05-23 15:07:16.395] [13041] [281464972439776] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281471986430176 [2025-05-23 15:07:16.395] [13041] [281464972439776] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281471970435296 [2025-05-23 15:07:16.395] [13041] [281464972439776] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281471889699040 2025-05-23 15:07:17.291 14551 LLM log default format: [yyyy-mm-dd hh:mm:ss.uuuuuu] [processid] [threadid] [llm] [loglevel] [file:line] [status code] msg [2025-05-23 15:07:17.291] [13011] [281465486045408] [llm] [INFO] [post_processing_manager.cpp:161] Get post processing manager [2025-05-23 15:07:17.291] [13011] [281465486045408] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281464318193888 [2025-05-23 15:07:17.291] [13011] [281465486045408] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472657125600 [2025-05-23 15:07:17.291] [13011] [281465486045408] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472615117024 [2025-05-23 15:07:17.291] [13011] [281465486045408] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472573108448 [2025-05-23 15:07:17.291] [13011] [281465486045408] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472398913760 [2025-05-23 15:07:17.291] [13011] [281465486045408] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472390459616 [2025-05-23 15:07:17.291] [13011] [281465486045408] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472382005472 [2025-05-23 15:07:17.291] [13011] [281465486045408] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472373551328 2025-05-23 15:07:17.919 14552 LLM log default format: [yyyy-mm-dd hh:mm:ss.uuuuuu] [processid] [threadid] [llm] [loglevel] [file:line] [status code] msg [2025-05-23 15:07:17.920] [13066] [281464923025632] [llm] [INFO] [post_processing_manager.cpp:161] Get post processing manager [2025-05-23 15:07:17.920] [13066] [281464923025632] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281463621742816 [2025-05-23 15:07:17.920] [13066] [281464923025632] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281471755481312 [2025-05-23 15:07:17.920] [13066] [281464923025632] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281471747027168 [2025-05-23 15:07:17.920] [13066] [281464923025632] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281471738573024 [2025-05-23 15:07:17.920] [13066] [281464923025632] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281471730118880 [2025-05-23 15:07:17.920] [13066] [281464923025632] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281471721664736 [2025-05-23 15:07:17.920] [13066] [281464923025632] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281471713210592 [2025-05-23 15:07:17.920] [13066] [281464923025632] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281471704756448 2025-05-23 15:07:18.081 14549 LLM log default format: [yyyy-mm-dd hh:mm:ss.uuuuuu] [processid] [threadid] [llm] [loglevel] [file:line] [status code] msg [2025-05-23 15:07:18.081] [13013] [281465282687200] [llm] [INFO] [post_processing_manager.cpp:161] Get post processing manager [2025-05-23 15:07:18.082] [13013] [281465282687200] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281464727662816 [2025-05-23 15:07:18.082] [13013] [281465282687200] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472256504032 [2025-05-23 15:07:18.082] [13013] [281465282687200] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472248049888 [2025-05-23 15:07:18.082] [13013] [281465282687200] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472239595744 [2025-05-23 15:07:18.082] [13013] [281465282687200] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472091025632 [2025-05-23 15:07:18.082] [13013] [281465282687200] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472082571488 [2025-05-23 15:07:18.082] [13013] [281465282687200] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472074117344 [2025-05-23 15:07:18.082] [13013] [281465282687200] [llm] [INFO] [thread_pool.cpp:51] Create pthread id: 281472065663200 [2025-05-23 15:07:18,809] [13041] [281464972439776] [llmmodels] [INFO] [flash_causal_lm.py-339] : <<<<<<<after transdata k_caches[0].shape=torch.Size([1, 8, 128, 16]) [2025-05-23 15:07:19,962] [13011] [281465486045408] [llmmodels] [INFO] [flash_causal_lm.py-339] : <<<<<<<after transdata k_caches[0].shape=torch.Size([1, 8, 128, 16]) [2025-05-23 15:07:20,636] [13066] [281464923025632] [llmmodels] [INFO] [flash_causal_lm.py-339] : <<<<<<<after transdata k_caches[0].shape=torch.Size([1, 8, 128, 16]) [2025-05-23 15:07:20,924] [13013] [281465282687200] [llmmodels] [INFO] [flash_causal_lm.py-339] : <<<<<<<after transdata k_caches[0].shape=torch.Size([1, 8, 128, 16])
我要发帖子
求助,通过mindie:2.0.RC1-300I-Duo-py311-openeuler24.03-lts镜像使用300IDuo 4卡8芯推理Qwen3-30B-A3B时,推理服务一直起不来。日志如下面所示,一直卡在这。且期间服务器CPU内存持续占用,会把CPU内存打满,导致服务器宕机。
感觉加载模型这一环节有什么问题,但是定位不到根本原因。
有大佬遇过类似问题吗?