yudian0504
yudian0504
> > 6B,10B,130B是参数量为60亿、100亿、1300亿，一般来说参数量越多脑容量越大。 是否+chat我理解是代表有没有经过中文QA和对话数据集的训练。 > > 是的，但是他们的模型结构也有调整，必须弄清楚调整了哪里，才能在FasterTransformer进行对应的调整。 准备跟你做一样的事情；请问有一些进展了吗？
> Hi! I'm also interested in this model and have a couple of questions: > > 1. Do we need to register the model config in both **python/sglang/srt/hf_transformers_utils.py** and **python/sglang/srt/configs/init.py**?...
> @yudian0504 I tested the pr with your command, the result seems incorrect.  Please refer to the cited section above from https://huggingface.co/PaddlePaddle/PaddleOCR-VL; In SGLang, only the VL model part...
> > . In fact, the full functionality of PaddleOCR depends on Baidu's official CLI or Python API, whereas the service launched by SGLang only replaces the vl_rec_server_url part of...
> Could you rebase on main? This will fix the errors in the CI, and then it should be ready to merge. done
/tag-and-rerun-ci