torin

line annotation bug

1

when start a task and load imgs and prelabel files, if the img-label pairs' file number is large, the client would crash. i guess the program preload all annotation at...

[Feature] support s-lora in turbomind backend

4

### Motivation In downstream tasks, Lora is one of the most common way to finetune llm. The inference speed degrades awfully from [turbomind backend+merge lora ]to [pytorch backend + merge...

[Feature] Wish to optimize prefix caching support for VLM

2

### Motivation vllm has established a V1 version, which supports prefix caching in multi modality llm. As a comparable infer engine，Wish LMDeploy has comparable feature :) ### Related resources _No...