Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
Menu
Open sidebar
ModelZoo
Deepseek-V3.1_vllm
Commits
5bdb0380
Commit
5bdb0380
authored
Sep 17, 2025
by
chenych
Browse files
Update README
parent
034c3360
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
8 additions
and
8 deletions
+8
-8
README.md
README.md
+8
-8
No files found.
README.md
View file @
5bdb0380
# Deepseek-V3.1
## 论文
`DeepSeek-V3 Technical Report`
-
https://arxiv.org/abs/2412.19437
暂无
## 模型结构
DeepSeek-V3.1 是一个支持思考模式和非思考模式的混合模型。与之前的版本相比,此次升级在多个方面都有所改进:
...
...
@@ -18,7 +17,7 @@ DeepSeek-V3.1 是在 DeepSeek-V3.1-Base 的基础上进行后置训练的。Deep
## 环境配置
### 硬件需求
DCU型号:BW
200
,节点数量:4台,卡数:32 张。
DCU型号:BW,节点数量:4台,卡数:32 张。
`-v 路径`
、
`docker_name`
和
`imageID`
根据实际情况修改
### Docker(方法一)
...
...
@@ -65,7 +64,7 @@ python ./infer/fp8_cast_bf16.py --input-fp8-hf-path /path/to/fp8_weights --outpu
### vllm推理方法
#### server 多机
样例模型:
[
deepseek-ai/
DeepSeek-V3.1
](
https://huggingface.co/deepseek-ai/DeepSeek-V3
)
样例模型:
[
DeepSeek-V3.1
](
https://huggingface.co/deepseek-ai/DeepSeek-V3
.1
)
1.
加入环境变量
> 请注意:
...
...
@@ -74,7 +73,9 @@ python ./infer/fp8_cast_bf16.py --input-fp8-hf-path /path/to/fp8_weights --outpu
> VLLM_HOST_IP:节点本地通信口ip,尽量选择IB网卡的IP,**避免出现rccl超时问题**
>
> NCCL_SOCKET_IFNAME和GLOO_SOCKET_IFNAME:节点本地通信网口ip对应的名称
>
> 通信口和ip查询方法:ifconfig
>
> IB口状态查询:ibstat !!!一定要active激活状态才可用,各个节点要保持统一
<div
align=
center
>
...
...
@@ -163,12 +164,11 @@ DCU与GPU精度一致,推理框架:vllm。
`制造,金融,教育`
## 预训练权重
-
[
deepseek-ai/
DeepSeek-V3.1
](
https://huggingface.co/deepseek-ai/DeepSeek-V3
)
-
[
deepseek-ai/
DeepSeek-V3.1-Base
](
https://h
f-mirror
.co
m
/deepseek-ai/DeepSeek-V3.1-Base
)
-
[
DeepSeek-V3.1
](
https://huggingface.co/deepseek-ai/DeepSeek-V3
.1
)
-
[
DeepSeek-V3.1-Base
](
https://h
uggingface
.co/deepseek-ai/DeepSeek-V3.1-Base
)
## 源码仓库及问题反馈
-
https://developer.sourcefind.cn/codes/modelzoo/deepseek-v3.1_vllm
## 参考资料
-
https://huggingface.co/deepseek-ai
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment