These are prerequisite for making mlc-serve an independent package.
- Mixtral support @vinx13
- vLLM v2 kernel @vinx13
- Misc changes in core.py for mlc-serve-specific artifact dump @sunggg
- Batched model support for split + rotary fusion (
mlc_llm/transform/fuse_split_rotary_embedding.py). This one depends on a hack to TVM