From 8996a0421a0197b680563cc32c055e520c797b9c Mon Sep 17 00:00:00 2001 From: RunningLeon Date: Wed, 18 Oct 2023 14:31:01 +0800 Subject: [PATCH 01/11] update reqs --- requirements.txt | 20 +++----------------- requirements/build.txt | 2 ++ requirements/optional.txt | 5 +++++ requirements/runtime.txt | 9 +++++++++ requirements/test.txt | 2 ++ setup.py | 8 +++++++- 6 files changed, 28 insertions(+), 18 deletions(-) create mode 100644 requirements/build.txt create mode 100644 requirements/optional.txt create mode 100644 requirements/runtime.txt create mode 100644 requirements/test.txt diff --git a/requirements.txt b/requirements.txt index 861623c040..176d0edefa 100644 --- a/requirements.txt +++ b/requirements.txt @@ -1,17 +1,3 @@ -accelerate -datasets -fastapi -fire -gradio -mmengine -numpy -pybind11 -safetensors -sentencepiece -setuptools -shortuuid -tiktoken -torch -transformers>=4.33.0 -tritonclient[all] -uvicorn +-r requirements/build.txt +-r requirements/runtime.txt +-r requirements/optional.txt diff --git a/requirements/build.txt b/requirements/build.txt new file mode 100644 index 0000000000..b4430ae374 --- /dev/null +++ b/requirements/build.txt @@ -0,0 +1,2 @@ +pybind11 +setuptools diff --git a/requirements/optional.txt b/requirements/optional.txt new file mode 100644 index 0000000000..e4d306d96d --- /dev/null +++ b/requirements/optional.txt @@ -0,0 +1,5 @@ +accelerate +datasets +fastapi +shortuuid +uvicorn diff --git a/requirements/runtime.txt b/requirements/runtime.txt new file mode 100644 index 0000000000..765cf9d7dc --- /dev/null +++ b/requirements/runtime.txt @@ -0,0 +1,9 @@ +fire +gradio +mmengine +numpy +safetensors +sentencepiece +tiktoken +torch +transformers>=4.33.0 diff --git a/requirements/test.txt b/requirements/test.txt new file mode 100644 index 0000000000..7093b61a35 --- /dev/null +++ b/requirements/test.txt @@ -0,0 +1,2 @@ +coverage +pytest diff --git a/setup.py b/setup.py index df36118c23..2f384548ca 100644 --- a/setup.py +++ b/setup.py @@ -134,7 +134,13 @@ def gen_packages_items(): 'lmdeploy': lmdeploy_package_data, }, include_package_data=True, - install_requires=parse_requirements('requirements.txt'), + setup_requires=parse_requirements('requirements/build.txt'), + tests_require=parse_requirements('requirements/test.txt'), + install_requires=parse_requirements('requirements/runtime.txt'), + extras_require={ + 'all': parse_requirements('requirements.txt'), + 'optional': parse_requirements('requirements/optional.txt'), + }, has_ext_modules=check_ext_modules, classifiers=[ 'Programming Language :: Python :: 3.8', From 8cd2c5c573f6d04e8e66c99c5976b053345d1c19 Mon Sep 17 00:00:00 2001 From: RunningLeon Date: Wed, 18 Oct 2023 15:14:33 +0800 Subject: [PATCH 02/11] update docs --- README.md | 3 ++- README_zh-CN.md | 3 ++- docs/en/faq.md | 4 ++-- docs/en/supported_models/codellama.md | 2 +- docs/en/w4a16.md | 2 +- docs/zh_cn/faq.md | 4 ++-- docs/zh_cn/supported_models/codellama.md | 2 +- docs/zh_cn/w4a16.md | 2 +- requirements.txt | 3 ++- requirements/lite.txt | 2 ++ requirements/{optional.txt => serve.txt} | 2 -- 11 files changed, 16 insertions(+), 13 deletions(-) create mode 100644 requirements/lite.txt rename requirements/{optional.txt => serve.txt} (56%) diff --git a/README.md b/README.md index c65cff7e5f..24ef5fc0e3 100644 --- a/README.md +++ b/README.md @@ -100,7 +100,7 @@ And the request throughput of TurboMind is 30% higher than vLLM. Install lmdeploy with pip ( python 3.8+) or [from source](./docs/en/build.md) ```shell -pip install lmdeploy +pip install lmdeploy[all] ``` ### Deploy InternLM @@ -182,6 +182,7 @@ bash workspace/service_docker_up.sh Then, you can communicate with the inference server by command line, ```shell +python3 -m pip install tritonclient[grpc] lmdeploy serve triton_client {server_ip_addresss}:33337 ``` diff --git a/README_zh-CN.md b/README_zh-CN.md index 84f860ef3d..65cfb47767 100644 --- a/README_zh-CN.md +++ b/README_zh-CN.md @@ -101,7 +101,7 @@ TurboMind 的 output token throughput 超过 2000 token/s, 整体比 DeepSpeed 使用 pip ( python 3.8+) 安装 LMDeploy，或者[源码安装](./docs/zh_cn/build.md) ```shell -pip install lmdeploy +pip install lmdeploy[all] ``` ### 部署 InternLM @@ -182,6 +182,7 @@ bash workspace/service_docker_up.sh 你可以通过命令行方式与推理服务进行对话： ```shell +python3 -m pip install tritonclient[grpc] lmdeploy serve triton_client {server_ip_addresss}:33337 ``` diff --git a/docs/en/faq.md b/docs/en/faq.md index 636da5c947..6f109997cc 100644 --- a/docs/en/faq.md +++ b/docs/en/faq.md @@ -17,7 +17,7 @@ It may have been caused by the following reasons. 1. You haven't installed lmdeploy's precompiled package. `_turbomind` is the pybind package of c++ turbomind, which involves compilation. It is recommended that you install the precompiled one. ```shell -pip install lmdeploy +pip install lmdeploy[all] ``` 2. If you have installed it and still encounter this issue, it is probably because you are executing turbomind-related command in the root directory of lmdeploy source code. Switching to another directory will fix it @@ -26,7 +26,7 @@ pip install lmdeploy ### libnccl.so.2 not found -Make sure you have install lmdeploy (>=v0.0.5) through `pip install lmdeploy`. +Make sure you have install lmdeploy (>=v0.0.5) through `pip install lmdeploy[all]`. If the issue still exists after lmdeploy installation, add the path of `libnccl.so.2` to environment variable LD_LIBRARY_PATH. diff --git a/docs/en/supported_models/codellama.md b/docs/en/supported_models/codellama.md index 886dc5922c..72a1644eba 100644 --- a/docs/en/supported_models/codellama.md +++ b/docs/en/supported_models/codellama.md @@ -26,7 +26,7 @@ Based on the above table, download the model that meets your requirements. Execu ```shell # install lmdeploy -python3 -m pip install lmdeploy +python3 -m pip install lmdeploy[all] # convert weight layout lmdeploy convert codellama /the/path/of/codellama/model diff --git a/docs/en/w4a16.md b/docs/en/w4a16.md index dc70c23c2c..7b44ece9e1 100644 --- a/docs/en/w4a16.md +++ b/docs/en/w4a16.md @@ -5,7 +5,7 @@ LMDeploy supports LLM model inference of 4-bit weight, with the minimum requirem Before proceeding with the inference, please ensure that lmdeploy is installed. ```shell -pip install lmdeploy +pip install lmdeploy[all] ``` ## 4-bit LLM model Inference diff --git a/docs/zh_cn/faq.md b/docs/zh_cn/faq.md index 5f3bf0b117..c86bfc1841 100644 --- a/docs/zh_cn/faq.md +++ b/docs/zh_cn/faq.md @@ -17,7 +17,7 @@ pip install --upgrade mmengine 1. 您没有安装 lmdeploy 的预编译包。`_turbomind`是 turbomind c++ 的 pybind部分，涉及到编译。推荐您直接安装预编译包。 ``` -pip install lmdeploy +pip install lmdeploy[all] ``` 2. 如果已经安装了，还是出现这个问题，请检查下执行目录。不要在 lmdeploy 的源码根目录下执行 python -m lmdeploy.turbomind.\*下的package，换到其他目录下执行。 @@ -26,7 +26,7 @@ pip install lmdeploy ### libnccl.so.2 not found -确保通过 `pip install lmdeploy` 安装了 lmdeploy (>=v0.0.5)。 +确保通过 `pip install lmdeploy[all]` 安装了 lmdeploy (>=v0.0.5)。 如果安装之后，问题还存在，那么就把`libnccl.so.2`的路径加入到环境变量 LD_LIBRARY_PATH 中。 diff --git a/docs/zh_cn/supported_models/codellama.md b/docs/zh_cn/supported_models/codellama.md index a2abd2f4a9..47db94b62c 100644 --- a/docs/zh_cn/supported_models/codellama.md +++ b/docs/zh_cn/supported_models/codellama.md @@ -26,7 +26,7 @@ ```shell # 安装 lmdeploy -python3 -m pip install lmdeploy +python3 -m pip install lmdeploy[all] # 转模型格式 lmdeploy convert codellama /path/of/codellama/model diff --git a/docs/zh_cn/w4a16.md b/docs/zh_cn/w4a16.md index 46f5c58a91..259847da16 100644 --- a/docs/zh_cn/w4a16.md +++ b/docs/zh_cn/w4a16.md @@ -5,7 +5,7 @@ LMDeploy 支持 4bit 权重模型的推理，**对 NVIDIA 显卡的最低要求 在推理之前，请确保安装了 lmdeploy ```shell -pip install lmdeploy +pip install lmdeploy[all] ``` ## 4bit 权重模型推理 diff --git a/requirements.txt b/requirements.txt index 176d0edefa..91d38808f1 100644 --- a/requirements.txt +++ b/requirements.txt @@ -1,3 +1,4 @@ -r requirements/build.txt -r requirements/runtime.txt --r requirements/optional.txt +-r requirements/lite.txt +-r requirements/serve.txt diff --git a/requirements/lite.txt b/requirements/lite.txt new file mode 100644 index 0000000000..1c2319c73d --- /dev/null +++ b/requirements/lite.txt @@ -0,0 +1,2 @@ +accelerate +datasets diff --git a/requirements/optional.txt b/requirements/serve.txt similarity index 56% rename from requirements/optional.txt rename to requirements/serve.txt index e4d306d96d..6a58dfaba5 100644 --- a/requirements/optional.txt +++ b/requirements/serve.txt @@ -1,5 +1,3 @@ -accelerate -datasets fastapi shortuuid uvicorn From 0e35d07de611aac1dbe466b653f89727b366a1b6 Mon Sep 17 00:00:00 2001 From: RunningLeon Date: Wed, 25 Oct 2023 14:19:56 +0800 Subject: [PATCH 03/11] resolve comments --- requirements/runtime.txt | 1 - requirements/serve.txt | 1 + 2 files changed, 1 insertion(+), 1 deletion(-) diff --git a/requirements/runtime.txt b/requirements/runtime.txt index 765cf9d7dc..c65dceae70 100644 --- a/requirements/runtime.txt +++ b/requirements/runtime.txt @@ -1,5 +1,4 @@ fire -gradio mmengine numpy safetensors diff --git a/requirements/serve.txt b/requirements/serve.txt index 6a58dfaba5..71bf5f4469 100644 --- a/requirements/serve.txt +++ b/requirements/serve.txt @@ -1,3 +1,4 @@ fastapi +gradio shortuuid uvicorn From b8267bbbbce7bb360211f5b572091922a7f914ac Mon Sep 17 00:00:00 2001 From: RunningLeon Date: Wed, 25 Oct 2023 14:46:41 +0800 Subject: [PATCH 04/11] upgrade pydantic --- requirements/serve.txt | 1 + 1 file changed, 1 insertion(+) diff --git a/requirements/serve.txt b/requirements/serve.txt index 71bf5f4469..6f1ce9fb42 100644 --- a/requirements/serve.txt +++ b/requirements/serve.txt @@ -1,4 +1,5 @@ fastapi gradio +pydantic>2.0.0 shortuuid uvicorn From 080d7ad1d5e0f33fd0bc7f45d2e3a8b322ede9e2 Mon Sep 17 00:00:00 2001 From: RunningLeon Date: Wed, 25 Oct 2023 16:25:31 +0800 Subject: [PATCH 05/11] fix rebase --- setup.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/setup.py b/setup.py index 2f384548ca..ff009b735f 100644 --- a/setup.py +++ b/setup.py @@ -139,7 +139,8 @@ def gen_packages_items(): install_requires=parse_requirements('requirements/runtime.txt'), extras_require={ 'all': parse_requirements('requirements.txt'), - 'optional': parse_requirements('requirements/optional.txt'), + 'lite': parse_requirements('requirements/lite.txt'), + 'serve': parse_requirements('requirements/serve.txt') }, has_ext_modules=check_ext_modules, classifiers=[ From f615bbc4bd0ba144fa7501eea082c85591cad8e5 Mon Sep 17 00:00:00 2001 From: RunningLeon Date: Thu, 26 Oct 2023 14:25:24 +0800 Subject: [PATCH 06/11] update doc --- README.md | 8 ++++++++ README_zh-CN.md | 8 ++++++++ 2 files changed, 16 insertions(+) diff --git a/README.md b/README.md index 24ef5fc0e3..2978724064 100644 --- a/README.md +++ b/README.md @@ -103,6 +103,14 @@ Install lmdeploy with pip ( python 3.8+) or [from source](./docs/en/build.md) pip install lmdeploy[all] ``` +```{note} +Users can install lmdeploy with extra required packages, eg: pip install lmdeploy[all] + +- all: Install lmdeploy with dependencies in requirements.txt +- lite: Install lmdeploy with dependencies in requirements/runtime.txt and requirements/lite.txt +- serve: Install lmdeploy with dependencies in requirements/runtime.txt and requirements/serve.txt +``` + ### Deploy InternLM #### Get InternLM model diff --git a/README_zh-CN.md b/README_zh-CN.md index 65cfb47767..e163cf9793 100644 --- a/README_zh-CN.md +++ b/README_zh-CN.md @@ -104,6 +104,14 @@ TurboMind 的 output token throughput 超过 2000 token/s, 整体比 DeepSpeed pip install lmdeploy[all] ``` +```{note} +用户在安装lmdeplou时可指定需要安装的额外依赖包，例如: pip install lmdeploy[all]会安装所有依赖包。 + +- all: 安装lmdeploy时，安装在 requirements.txt 中的所有依赖包。 +- lite: 安装lmdeploy时，安装在 requirements/runtime.txt 和 requirements/lite.txt 中的依赖包。 +- serve: 安装lmdeploy时，安装在 requirements/runtime.txt 和 requirements/lite.txt 中的依赖包。 +``` + ### 部署 InternLM #### 获取 InternLM 模型 From 94309e6dc44879a0df5087e1aa6405589d4a5900 Mon Sep 17 00:00:00 2001 From: RunningLeon Date: Thu, 26 Oct 2023 14:36:03 +0800 Subject: [PATCH 07/11] update --- README.md | 13 ++++++------- README_zh-CN.md | 13 ++++++------- 2 files changed, 12 insertions(+), 14 deletions(-) diff --git a/README.md b/README.md index 2978724064..76cfbfe7ff 100644 --- a/README.md +++ b/README.md @@ -103,13 +103,12 @@ Install lmdeploy with pip ( python 3.8+) or [from source](./docs/en/build.md) pip install lmdeploy[all] ``` -```{note} -Users can install lmdeploy with extra required packages, eg: pip install lmdeploy[all] - -- all: Install lmdeploy with dependencies in requirements.txt -- lite: Install lmdeploy with dependencies in requirements/runtime.txt and requirements/lite.txt -- serve: Install lmdeploy with dependencies in requirements/runtime.txt and requirements/serve.txt -``` +> **Note**
+> Users can install lmdeploy with extra required packages, eg: `pip install lmdeploy[all]` +> +> - `all`: Install lmdeploy with dependencies in `requirements.txt` +> - `lite`: Install lmdeploy with dependencies in `requirements/runtime.txt` and `requirements/lite.txt` +> - `serve`: Install lmdeploy with dependencies in `requirements/runtime.txt` and `requirements/serve.txt` ### Deploy InternLM diff --git a/README_zh-CN.md b/README_zh-CN.md index e163cf9793..33eaf7e01f 100644 --- a/README_zh-CN.md +++ b/README_zh-CN.md @@ -104,13 +104,12 @@ TurboMind 的 output token throughput 超过 2000 token/s, 整体比 DeepSpeed pip install lmdeploy[all] ``` -```{note} -用户在安装lmdeplou时可指定需要安装的额外依赖包，例如: pip install lmdeploy[all]会安装所有依赖包。 - -- all: 安装lmdeploy时，安装在 requirements.txt 中的所有依赖包。 -- lite: 安装lmdeploy时，安装在 requirements/runtime.txt 和 requirements/lite.txt 中的依赖包。 -- serve: 安装lmdeploy时，安装在 requirements/runtime.txt 和 requirements/lite.txt 中的依赖包。 -``` +> **Note**
+> 用户在安装lmdeplou时可指定需要安装的额外依赖包，例如: `pip install lmdeploy[all]` 会安装所有依赖包。 +> +> - `all`: 安装lmdeploy时，安装在 `requirements.txt` 中的所有依赖包。 +> - `lite`: 安装lmdeploy时，安装在 `requirements/runtime.txt` 和 `requirements/lite.txt` 中的依赖包。 +> - `serve`: 安装lmdeploy时，安装在 `requirements/runtime.txt` 和 `requirements/lite.txt` 中的依赖包。 ### 部署 InternLM From 1d0ce881296ca7cbba547e129fa69d7bf7940982 Mon Sep 17 00:00:00 2001 From: RunningLeon Date: Thu, 26 Oct 2023 14:36:55 +0800 Subject: [PATCH 08/11] update --- README_zh-CN.md | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/README_zh-CN.md b/README_zh-CN.md index 33eaf7e01f..b0a0b6602b 100644 --- a/README_zh-CN.md +++ b/README_zh-CN.md @@ -105,11 +105,11 @@ pip install lmdeploy[all] ``` > **Note**
-> 用户在安装lmdeplou时可指定需要安装的额外依赖包，例如: `pip install lmdeploy[all]` 会安装所有依赖包。 +> 用户在安装lmdeplou时可指定需要安装的额外依赖包，例如: `pip install lmdeploy[all]` 会安装所有依赖包 > -> - `all`: 安装lmdeploy时，安装在 `requirements.txt` 中的所有依赖包。 -> - `lite`: 安装lmdeploy时，安装在 `requirements/runtime.txt` 和 `requirements/lite.txt` 中的依赖包。 -> - `serve`: 安装lmdeploy时，安装在 `requirements/runtime.txt` 和 `requirements/lite.txt` 中的依赖包。 +> - `all`: 安装lmdeploy时，安装在 `requirements.txt` 中的所有依赖包 +> - `lite`: 安装lmdeploy时，安装在 `requirements/runtime.txt` 和 `requirements/lite.txt` 中的依赖包 +> - `serve`: 安装lmdeploy时，安装在 `requirements/runtime.txt` 和 `requirements/lite.txt` 中的依赖包 ### 部署 InternLM From 25c467bd226591a084436d0b4c89e71e2a0efc20 Mon Sep 17 00:00:00 2001 From: RunningLeon Date: Thu, 26 Oct 2023 14:58:53 +0800 Subject: [PATCH 09/11] update readme --- README.md | 9 +++++---- README_zh-CN.md | 8 ++++---- 2 files changed, 9 insertions(+), 8 deletions(-) diff --git a/README.md b/README.md index 76cfbfe7ff..034f5754bd 100644 --- a/README.md +++ b/README.md @@ -104,11 +104,12 @@ pip install lmdeploy[all] ``` > **Note**
-> Users can install lmdeploy with extra required packages, eg: `pip install lmdeploy[all]` +> `pip install lmdeploy` can only install the runtime required packages. If users want to run codes from modules like `lmdeploy.lite` and `lmdeploy.serve`, they need to install the extra required packages. +> For instance, running `pip install lmdeploy[lite]` would install extra dependencies for `lmdeploy.lite` module. > -> - `all`: Install lmdeploy with dependencies in `requirements.txt` -> - `lite`: Install lmdeploy with dependencies in `requirements/runtime.txt` and `requirements/lite.txt` -> - `serve`: Install lmdeploy with dependencies in `requirements/runtime.txt` and `requirements/serve.txt` +> - `all`: Install lmdeploy with all dependencies in `requirements.txt` +> - `lite`: Install lmdeploy with extra dependencies in `requirements/lite.txt` +> - `serve`: Install lmdeploy with dependencies in `requirements/serve.txt` ### Deploy InternLM diff --git a/README_zh-CN.md b/README_zh-CN.md index b0a0b6602b..0217da795d 100644 --- a/README_zh-CN.md +++ b/README_zh-CN.md @@ -105,11 +105,11 @@ pip install lmdeploy[all] ``` > **Note**
-> 用户在安装lmdeplou时可指定需要安装的额外依赖包，例如: `pip install lmdeploy[all]` 会安装所有依赖包 +> `pip install lmdeploy`默认安装runtime依赖包，使用lmdeploy的lite和serve功能时，用户需要安装额外依赖包。例如: `pip install lmdeploy[lite]` 会额外安装`lmdeploy.lite`模块的依赖包 > -> - `all`: 安装lmdeploy时，安装在 `requirements.txt` 中的所有依赖包 -> - `lite`: 安装lmdeploy时，安装在 `requirements/runtime.txt` 和 `requirements/lite.txt` 中的依赖包 -> - `serve`: 安装lmdeploy时，安装在 `requirements/runtime.txt` 和 `requirements/lite.txt` 中的依赖包 +> - `all`: 安装`lmdeploy`所有依赖包，具体可查看`requirements.txt` +> - `lite`: 额外安装`lmdeploy.lite`模块的依赖包，具体可查看`requirements/lite.txt` +> - `serve`: 额外安装`lmdeploy.serve`模块的依赖包，具体可查看`requirements/serve.txt` ### 部署 InternLM From 88e648c759084756db742b9ff2510456a78c8b76 Mon Sep 17 00:00:00 2001 From: RunningLeon Date: Tue, 7 Nov 2023 08:58:40 +0800 Subject: [PATCH 10/11] update --- README.md | 8 +++++++- README_zh-CN.md | 8 +++++++- 2 files changed, 14 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 189c5a9d55..b3ff5231bb 100644 --- a/README.md +++ b/README.md @@ -100,7 +100,7 @@ And the request throughput of TurboMind is 30% higher than vLLM. Install lmdeploy with pip ( python 3.8+) or [from source](./docs/en/build.md) ```shell -pip install lmdeploy[all] +pip install lmdeploy ``` > **Note**
@@ -148,6 +148,9 @@ lmdeploy chat turbomind ./workspace #### Serving with gradio ```shell +# install lmdeploy with extra dependencies +pip install lmdeploy[serve] + lmdeploy serve gradio ./workspace ``` @@ -158,6 +161,9 @@ lmdeploy serve gradio ./workspace Launch inference server by: ```shell +# install lmdeploy with extra dependencies +pip install lmdeploy[serve] + lmdeploy serve api_server ./workspace --instance_num 32 --tp 1 ``` diff --git a/README_zh-CN.md b/README_zh-CN.md index d3f0177fef..7ea119aadd 100644 --- a/README_zh-CN.md +++ b/README_zh-CN.md @@ -101,7 +101,7 @@ TurboMind 的 output token throughput 超过 2000 token/s, 整体比 DeepSpeed 使用 pip ( python 3.8+) 安装 LMDeploy，或者[源码安装](./docs/zh_cn/build.md) ```shell -pip install lmdeploy[all] +pip install lmdeploy ``` > **Note**
@@ -147,6 +147,9 @@ lmdeploy chat turbomind ./workspace #### 启动 gradio server ```shell +# 安装lmdeploy额外依赖 +pip install lmdeploy[serve] + lmdeploy serve gradio ./workspace ``` @@ -157,6 +160,9 @@ lmdeploy serve gradio ./workspace 使用下面的命令启动推理服务： ```shell +# 安装lmdeploy额外依赖 +pip install lmdeploy[serve] + lmdeploy serve api_server ./workspace --server_name 0.0.0.0 --server_port ${server_port} --instance_num 32 --tp 1 ``` From c4a701201e17789a93c439c6a8528d5c2a4d9ee4 Mon Sep 17 00:00:00 2001 From: RunningLeon Date: Thu, 9 Nov 2023 09:30:49 +0800 Subject: [PATCH 11/11] add flash-attn --- requirements/lite.txt | 1 + 1 file changed, 1 insertion(+) diff --git a/requirements/lite.txt b/requirements/lite.txt index 1c2319c73d..bd10933103 100644 --- a/requirements/lite.txt +++ b/requirements/lite.txt @@ -1,2 +1,3 @@ accelerate datasets +flash-attn