GitLab

* support inference a batch of prompts

* docstring and assert

* add

* import fire in main

* wrap to speed up fire cli

* update

* update docs

* update docs

* fix

* resolve commennts

* resolve confict and add test for cli

* add build from docker section

* update

* install python package

* update

* update

* update

* fix crash

* update profile_generation.py

* format

* use self.bos_id

* remove sys_instruct

This reverts commit 4b5c2bda.

* robust incremental decode for leading space

* speed up lookup as prefix_space_tokens is shorter than no_prefix_space_tokens

* add UT and fix qwen stuff

* move tokenizer

* remove Tokenizer in init

* update deploy.py

* free runner disk

* limit cpu

* docker.yml

* keep swap

* keep swap

* [doc] Update benchmark command in w4a16.md

* Update w4a16.md

* Update w4a16.md

add pip install nvidia-ml-py

* [doc] Update w4a16.md

* fix lint error
Signed-off-by: del-zhenwu <dele.zhenwu@gmail.com>

* [doc] update model_path & prompt_tokens
Signed-off-by: del-zhenwu <dele.zhenwu@gmail.com>

---------
Signed-off-by: del-zhenwu <dele.zhenwu@gmail.com>

* add tp hint for deploy

* fix lint

* assert tp in turbomind

* fix lint

* support deploy qwen-14b-chat

* update README

* load safetensors first

* make IPv6 compatible, safe run for coroutine interrupting

* instance_id -> session_id and fix api_client.py

* update doc

* remove useless faq

* safe ip mapping

* update app.py

* remove print

* update doc

* change shared_instances_ from weakptr to sharedptr

* update

* Support CORS for openai api server

* Remove unnecessary var

* Add CORS support follow the same style with vllm

* Fix memory leak

* modern c++

* fix benchmark serving cannot use Qwen tokenizer

* update benchmark readme

* expose stop words

* support string

* fix

* remove eoa from chatbot

* remove eoa of turbomind

* fix ut

* suffix wheel and fix InternLM no system bug

* cuda allocator fix

* graceful termination

* lint and compilation fix

Fix side effect brought by supporting codellama: `sequence_start` is always true when calling `model.get_prompt` (#466)

quantilized -> quantized