cedriceuh/freetoken-rdna3

AMD Radeon RX 7900 계열 GPU에서 Mixture-of-Experts 모델을 로컬 추론하는 FreeToken 튜닝 빌드가 모든 프로파일과 체크포인트의 측정값을 다시 뽑아 README 속도표를 갱신했다.

GitHub 설명: Fast local LLM inference on AMD Radeon RX 7900 XTX / XT (RDNA3, ROCm): big Mixture-of-Experts models like Qwen3.8-Flash-Next on one or two consumer GPUs. Up to 4 parallel agents with 262k context each, or one 1M-token context, OpenAI/Anthropic API, NVFP4 and EXL3 checkpoints. A tuned build of FreeToken.