[Roadmap] AMD Instinct Development (2025Q4) #12890
Description
Checklist
- 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
- 2. Please use English, otherwise it will be closed.
Motivation
Focus
- Distributed Inference Performance and Scaling
- Models performance with kernel optimizations
- Extended Models coverage: Thinking, Diffusion
Distributed Inference
- Large-scale EP with MoRI
- PD/D performance
- OME Integration
- Gateway support
Performance Optimizations
- Communication Kernels: fusion, quantization
- Attention Kernels: performance, fp8
- FP8 KV cache with AITER kernels
- DeepSeek 3.2 performance
- gpt-oss performance
Model coverage
Reinforcement learning training framework
- slime integration improvements
Quantization
- More MxFP4/FP8 enablement
- Quark Integration as utility
Related resources
No response