v0.0.2

vllm-project/vllmv0.0.2Oct 29, 2024by ShenhanQian

AI Summary

Switches the rendering backend from OpenGL to CUDA to support higher resolutions and simplify installation requirements.

Key Highlights

Switch from OpenGL rasterizer to CUDA rasterizer
Support for non-multiple-of-8 resolutions
Support for resolutions over 2048x2048

New Features

CUDA rasterizer support
Support for resolutions > 2048x2048
Simplified installation via nvdiffrast

Full Release Notes

Since nvdiffrast now supports non-multiple-of-8 resolutions (https://github.com/NVlabs/nvdiffrast/releases/tag/v0.3.2) and resolutions over 2048x2048 (https://github.com/NVlabs/nvdiffrast/releases/tag/v0.3.3) in CUDA rasterizer, we switch from OpenGL rasterizer to CUDA rasterizer to simplify installation. You need to upgrade nvdiffrast via
  >```shell
  >  pip install nvdiffrast@git+https://github.com/ShenhanQian/nvdiffrast@backface-culling -U
  >  rm -r ~/.cache/torch_extensions/*/nvdiffrast*
>  ```