eelgaev/Legion
Rust LLM inference engine for MoE models on NVIDIA unified memory (IBM AC922, DGX Spark). Runs GGUF and Hugging Face safetensors (NVFP4, FP8) as shipped, with speculative decoding, continuous batching and prefix caching.
GitHub repository with 19 stars and 0 forks.
Language: Rust