Hongji Zhu comments

Results 64 comments of


                                            Hongji Zhu

Implement chatbot functionality using Streamlit

Thanks for your contribution，we will review this PR soon.

方便提供openai_api接口吗，这样我就可以做成服务调用了，VLLM就更好了

已支持[vLLM](https://github.com/OpenBMB/MiniCPM-V?tab=readme-ov-file#vllm) 推理，欢迎试用～

Request for a 4bit quantization model for omnilmm 12B

We released the more powerful [MiniCPM-Llama3-V 2.5](https://github.com/OpenBMB/MiniCPM-V/blob/main/README_en.md#minicpm-llama3-v-25) today，which has 8.5B parameters and much better performance， the int4 version is [here](https://github.com/OpenBMB/MiniCPM-V/blob/main/README_en.md#model-zoo). Welcome to use.

疑问

目前模型在文字全文识别时能力有限，针对各个字段单独提问效果好很多。近期我们也将更新提升全文识别能力，敬请关注～

MiniCPM-v2进行文字识别与提取的demo有建议吗？

我们发布了更强大的[MiniCPM-Llama3-V 2.5](https://github.com/OpenBMB/MiniCPM-V/tree/main?tab=readme-ov-file#minicpm-llama3-v-25), 文字识别能力显著提升，欢迎关注～

web_demo.py 运行需要10G显存。为啥 vllm 启动api模式运行，需要23G显存啊？

你好，vLLM有自己的显存管理机制，api模式初始化LLM时默认gpu_memory_utilization=0.9，所以用了22.5G，代码见：https://github.com/vllm-project/vllm/blob/main/vllm/entrypoints/llm.py#L93

MiniCPM-v2的对话能力

感谢关注！近期我们将会有新的版本发布，敬请期待～

MiniCPM-v2的对话能力

> > 请问VLLM 0.4.2报Model architectures ['MiniCPMV'] are not supported for now 应该怎么解决，VLLM版本降级吗？目前MiniCPM-V 2.0暂未合入官方vllm，请从https://github.com/OpenBMB/vllm 编译使用

MiniCPM-v2的对话能力

@yayafengzi @zRzRzRzRzRzRzR 更大参数版本MiniCPM-Llama3-V 2.5现已发布，更大更强，欢迎关注

What's the learning rate scheduler when training the perceiver resampler from scratch?

We will release the technical report in weeks, please stay tuned.