From af0d95be0aeedfd135b3929f0377e53ef9a581f9 Mon Sep 17 00:00:00 2001 From: jinminxi104 Date: Mon, 9 Dec 2024 11:39:39 +0800 Subject: [PATCH 1/3] Update dlinfer-ascend version in runtime_ascend.txt (#2865) --- requirements/runtime_ascend.txt | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/requirements/runtime_ascend.txt b/requirements/runtime_ascend.txt index 05d74bbe72..c5d44cc995 100644 --- a/requirements/runtime_ascend.txt +++ b/requirements/runtime_ascend.txt @@ -1,5 +1,5 @@ accelerate>=0.29.3 -dlinfer-ascend>=0.1.2 +dlinfer-ascend>=0.1.3 einops fastapi fire From 14b64c769247ae082cc541233dde59f65747a714 Mon Sep 17 00:00:00 2001 From: Lyu Han Date: Mon, 9 Dec 2024 20:07:44 +0800 Subject: [PATCH 2/3] bump version to v0.6.4 (#2864) --- docs/en/get_started/installation.md | 2 +- docs/zh_cn/get_started/installation.md | 2 +- lmdeploy/version.py | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) diff --git a/docs/en/get_started/installation.md b/docs/en/get_started/installation.md index b3e8bb8abd..c00111c2ab 100644 --- a/docs/en/get_started/installation.md +++ b/docs/en/get_started/installation.md @@ -23,7 +23,7 @@ pip install lmdeploy The default prebuilt package is compiled on **CUDA 12**. If CUDA 11+ (>=11.3) is required, you can install lmdeploy by: ```shell -export LMDEPLOY_VERSION=0.6.3 +export LMDEPLOY_VERSION=0.6.4 export PYTHON_VERSION=38 pip install https://github.com/InternLM/lmdeploy/releases/download/v${LMDEPLOY_VERSION}/lmdeploy-${LMDEPLOY_VERSION}+cu118-cp${PYTHON_VERSION}-cp${PYTHON_VERSION}-manylinux2014_x86_64.whl --extra-index-url https://download.pytorch.org/whl/cu118 ``` diff --git a/docs/zh_cn/get_started/installation.md b/docs/zh_cn/get_started/installation.md index 12562c51d5..0213fa6d15 100644 --- a/docs/zh_cn/get_started/installation.md +++ b/docs/zh_cn/get_started/installation.md @@ -23,7 +23,7 @@ pip install lmdeploy 默认的预构建包是在 **CUDA 12** 上编译的。如果需要 CUDA 11+ (>=11.3)，你可以使用以下命令安装 lmdeploy： ```shell -export LMDEPLOY_VERSION=0.6.3 +export LMDEPLOY_VERSION=0.6.4 export PYTHON_VERSION=38 pip install https://github.com/InternLM/lmdeploy/releases/download/v${LMDEPLOY_VERSION}/lmdeploy-${LMDEPLOY_VERSION}+cu118-cp${PYTHON_VERSION}-cp${PYTHON_VERSION}-manylinux2014_x86_64.whl --extra-index-url https://download.pytorch.org/whl/cu118 ``` diff --git a/lmdeploy/version.py b/lmdeploy/version.py index d9f4307a78..f705fcb332 100644 --- a/lmdeploy/version.py +++ b/lmdeploy/version.py @@ -1,7 +1,7 @@ # Copyright (c) OpenMMLab. All rights reserved. from typing import Tuple -__version__ = '0.6.3' +__version__ = '0.6.4' short_version = __version__ From 47fa7cf9baf8c3f68e1a2ab9c89c91d2d62bc841 Mon Sep 17 00:00:00 2001 From: Galaxy-Husky <598756381@qq.com> Date: Mon, 9 Dec 2024 21:43:07 +0800 Subject: [PATCH 3/3] Support for loading lora adapter weights in safetensors format (#2860) Co-authored-by: Ping --- lmdeploy/pytorch/models/patch.py | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/lmdeploy/pytorch/models/patch.py b/lmdeploy/pytorch/models/patch.py index 9da1b9f4ea..a7fe4431ed 100644 --- a/lmdeploy/pytorch/models/patch.py +++ b/lmdeploy/pytorch/models/patch.py @@ -8,6 +8,7 @@ import torch from transformers.configuration_utils import PretrainedConfig +from transformers.modeling_utils import load_state_dict from lmdeploy.utils import get_logger @@ -295,7 +296,9 @@ def add_adapters(model: torch.nn.Module, for name, path in adapters.items(): adapter_id = adapter_id_map[name] checkpoint_path = f'{path}/adapter_model.bin' - state_dict = torch.load(checkpoint_path, map_location=device) + if not osp.exists(checkpoint_path): + checkpoint_path = f'{path}/adapter_model.safetensors' + state_dict = load_state_dict(checkpoint_path, map_location=device) if hasattr(model, 'load_lora_weights'): model.load_lora_weights(state_dict.items(), adapter_id=adapter_id)