docs: add GPU requirements and multi-GPU usage tips to README files

This commit is contained in:
myhloli
2025-06-20 18:47:16 +08:00
parent d29cf4e076
commit 2c702890a4
2 changed files with 14 additions and 2 deletions

View File

@@ -665,6 +665,12 @@ mineru -p <input_path> -o <output_path> -b vlm-sglang-engine
mineru-sglang-server --port 30000
```
> [!TIP]
> sglang acceleration requires a GPU with Ampere architecture or newer, and at least 24GB VRAM. If you have two 12GB or 16GB GPUs, you can use Tensor Parallelism (TP) mode:
> `mineru-sglang-server --port 30000 --tp 2`
>
> If you still encounter out-of-memory errors with two GPUs, or if you need to improve throughput or inference speed using multi-GPU parallelism, please refer to the [sglang official documentation](https://docs.sglang.ai/backend/server_arguments.html#common-launch-commands).
2. Use Client in another terminal:
```bash

View File

@@ -493,8 +493,8 @@ uv pip install -e .[core] -i https://mirrors.aliyun.com/pypi/simple
```
> [!TIP]
> Linux和macOS系统安装后自动支持cuda/mps加速Windows用户如需使用cuda加速请前往 [Pytorch官网](https://pytorch.org/get-started/locally/)
> 选择合适的cuda版本安装pytorch。
> Linux和macOS系统安装后自动支持cuda/mps加速Windows用户如需使用cuda加速
> 请前往 [Pytorch官网](https://pytorch.org/get-started/locally/) 选择合适的cuda版本安装pytorch。
#### 1.3 安装完整版(支持 sglang 加速需确保设备有Ampere及以后架构24G显存及以上显卡
@@ -654,6 +654,12 @@ mineru -p <input_path> -o <output_path> -b vlm-sglang-engine
mineru-sglang-server --port 30000
```
> [!TIP]
> sglang加速需设备有Ampere及以后架构24G显存及以上显卡如您有两张12G或16G显卡可以通过张量并行TP模式使用。
> `mineru-sglang-server --port 30000 --tp 2`
>
> 如使用两张卡仍出现显存不足错误或需要使用多卡并行增加吞吐量或推理速度,请参考 [sglang官方文档](https://docs.sglang.ai/backend/server_arguments.html#common-launch-commands)
2. 在另一个终端中使用 Client 调用:
```bash