How to install and configure RolmOCR on your own computer?

2025-08-26

1.6 K

RolmOCR Detailed Installation Guide

The installation process is divided into four core steps:

environmental preparation: Confirm Python ≥ 3.8 environment, recommended to use Anaconda to create a virtual environment:
```
conda create -n rolmocr python=3.10
```
Frame mounting: After installing the vLLM inference engine via pip, environment variables must be set:
```
export VLLM_USE_MODELSCOPE=1
```
This is a key step in ensuring the compatibility of domestic models
Model Acquisition: Accelerated downloads via mirrored sources are available, except for Hugging Face:
```
git clone https://www.modelscope.cn/reducto/RolmOCR.git
```
service activation: It is recommended to add the -token flow-limiting parameter to ensure stability:
```
vllm serve --model reducto/RolmOCR --max-num-batched-tokens 4096
```

Troubleshooting: If startup fails, check whether the CUDA version is ≥11.7; if memory is insufficient, add the -gpu-memory-utilization 0.8 parameter.