Update app.py

Resolve typing_extensions version issue

.

Change port to 7899

.

Change port number to 7861

Fix issue where Qwen model_name was passed as model_path for loading

Add support for all models

Add separate num_frames slider

Add CUDA cache freeing before model re-loading

Add BF16/INT8/INT4 quantization support for LLaVA-Video to fit within 23GB VRAM HF Spaces limit

Remove s3fs version since it causes fsspec version conflict w/ other pkgs

Drop nccl-cu12 (2.27.5 -> 2.27.3) for torch==2.8

1. Add flash-attn 2.8.3 release link with ABI=False since it works on HF Spaces

Add new requirements.txt for transformers v5

Add transformers v5 integration to models

Add correct models/ repo

Add kwargs to LLaVA-Video model.chat()

Fix generation config

Add UI features

Finally worked after 10 hoursgit add .!

flash-attn ABI true -> false for HF spaces

Just the flash-attn error. Adding it to requirements.txt

fix LLaVA fork editable link issue in requirements.txt

Delete llava-next local copy, fork and update requirements.txt

Change pre-req.txt

Change pre-req.txt

Change pre-req.txt

Fix pre-requirements.txt

Add LLaVA-NExt locally

delete binary files in llava_local

Pre-install flag added in README.md

Install LLaVA-NeXT without deps to avoid version conflicts

Remove flash-attn for faster build, change LLaVA link

Drop python_version to 3.10, fix flash-attn .whl file for Python=3.10, torch=2.6

huggingface-hub>=0.21.0,<1.0

Bump torch==2.5.1 -> 2.6.0

Drop sympy==1.14.0 -> 1.13.1

Fix triton version

Fix flash-attn python version to 3.13 for demo compatibility

Fix flash-attn

Fix requirements.txt

Fix requirements.txt

Fix requirements.txt

Modify requirements.txt

Modify requirments.txt

Add videos to Git LFS

Initial commit without videos

initial commit