Christopher Chou

Add callback on save for LoRA

7

## Why are these changes needed? When we save the model checkpoint, it saves the entire model in `pytorch_model.bin` which is extremely large, but we often only want the adapter...

## Why are these changes needed? Update the documentation to include the training script to fine-tune using LoRA/QLoRA. If necessary, I could also add the reference `deepspeed_config.json`, but I didn't...

Adding support for the Yi-VL Model: https://huggingface.co/01-ai/Yi-VL-6B

## Why are these changes needed? Side-by-side multimodal chatbot arena. Please comment any advice in terms of the UI or logic! This branch is a fork of the direct chat...

## Why are these changes needed? Converts the current imagebox + textbox setup to just use the multimodal textbox from gradio. ## Related issue number (if applicable) ## Checks -...

## Why are these changes needed? 1. Add CSAM and NSFW moderation filter. Check the README for how to run. Notably, the NSFW endpoint should be the full endpoint now,...

[WIP] Add Qwen-VL

1

Add Qwen-VL model to SGLang. Rough roadmap: - [ ] Write out classes and test model loading - [ ] Test out CLIPVisionModel forward and Cross attention forward - [...

Change log to separate data that has been flagged.

flag preset images more easily. note: at some point we probably want to move these image features to the image class and create a todict() class for that

[Bug]: Pixtral fails when limit_mm_per_prompt not set

11

### Your current environment The output of `python collect_env.py` ```text Your output of `python collect_env.py` here ``` ### 🐛 Describe the bug The below command does not work ``` CUDA_VISIBLE_DEVICES=3...