⬢github Python · 999 ★ +127 since we first saw it · pushed 2 d ago · NOASSERTION
eagle-hq/EAGLE-4
Official Inference Code of EAGLE-4
EAGLE-4 is the official inference code for EAGLE-4, a speculative decoding method that speeds up LLM generation (up to 3.4× on Qwen3.8-27B) while keeping outputs identical to the target model. Its eagle-harness package adapts draft speculation to batch size, hardware, and workload, integrating with SGLang, vLLM, and TokenSpeed.
Why now: It's newly released with an accompanying paper and draft models, and is appearing among the most-starred new GitHub repos.
Who it is for: Developers and teams serving large language models who want faster GPU inference without changing model outputs.
eaglegpullm-inferencesglangspeculative-decodingtokenspeedvllm
Stars over our 157 snapshots: 872 to 999, since 3 d ago.
Where people talked about it
- ⬢github new repos, most starred 8 min ago
- ⬢github new repos, most starred 10 h ago
- ⬢github new repos, most starred 1 d ago
- ⬢github new repos, most starred 2 d ago
API: https://socialmediatrends-api.osmike.com/v1/repos/eagle-hq/EAGLE-4