New
efficientmoe/moe-infinity — Qwen3.5-MoE paged KV cache serving
efficientmoe/moe-infinity added a commit that connects the gated full attention layers of Qwen3.5-MoE to a paged KV cache, fixing the broken multi-token decoding.
efficientmoe/moe-infinity
efficientmoe/moe-infinity added a commit that connects the gated full attention layers of Qwen3.5-MoE to a paged KV cache, fixing the broken multi-token decoding.
View on GitHub ↗- stars
- ★ 373
- Forks
- 46
- Language
- Python
- License
- Apache-2.0
- Latest release
- v0.2.2
Posts
1No matching stories.