Wei Tao

**Describe the bug**  **Environment** pytorch2.2.2 cuda11.8 Need help!

Environment: transformer 4.40.1 torch 2.0.1+cu118 tokenizers 0.19.1 when i try torch.jit.trace the dolly-v2-3b model, it reports below, it smees dolly dosen't support torch.jit.trace, can anyone give me some isights? ![1714458130252](https://github.com/databrickslabs/dolly/assets/126441921/5b3df095-6e60-42ac-9544-267c...

How to get best performace with optimization with torch_blade

2

[My script compare Torch with Disc](https://gist.github.com/JackWeiw/a47cf88ad327fe5db5e150fe2f80d26e) However according to profiling , Disc boost litle compare to Torch, even in some dynamic range worse than torch,. Is there some i missing...

[dlinfer][ascend]set blcoksize default to 128 for 310P device

2

## Motivation Dlinfer Ascend 310P device will trigger error when using lmdeploy chat api, since the original default block size is 64. Set block size to 128 will prevent this...