john tong comments

Results 13 comments of


                                            john tong

pyfastllm 内存泄露

是基于cpu运行的吗

[Bug] triton.runtime.autotuner.OutOfResources: out of resource: shared memory, Required: 108672, Hardware limit: 101376. Reducing block sizes or `num_stages` may help.

> [lmdeploy/lmdeploy/pytorch/内核/cuda/pagedattention.py](https://github.com/InternLM/lmdeploy/blob/edcdd8e36520b8bf7dbc99feecd2d2822c4cb5ba/lmdeploy/pytorch/kernels/cuda/pagedattention.py#L35) > > 第 35 行在[EDCDD8E](/InternLM/lmdeploy/commit/edcdd8e36520b8bf7dbc99feecd2d2822c4cb5ba) > > 海卫一。config（{}， num_stages=2， num_warps=16）， > [lmdeploy/lmdeploy/pytorch/内核/cuda/pagedattention.py](https://github.com/InternLM/lmdeploy/blob/edcdd8e36520b8bf7dbc99feecd2d2822c4cb5ba/lmdeploy/pytorch/kernels/cuda/pagedattention.py#L583) > > 583 行在[EDCDD8E](/InternLM/lmdeploy/commit/edcdd8e36520b8bf7dbc99feecd2d2822c4cb5ba) > > num_stages = 2 > > 可以试试看把这些地方的 num_stages 改成...

录制状态始终显示未直播，录制不到视频文件

是抖音吗，其他直播间可以正常录制吗