nvidia/Kimi-K2.6-Eagle3

Add lm_head (= moonshotai/Kimi-K2.6 lm_head) ; 16 bf16 tensors

Convert weights to bf16 (match config torch_dtype)

README: mention avg acceptance length (MT-Bench 2.62, SPEED-Bench 2.67, draft length 3) in Description

README: fix unclosed bold marker, add SPEED-Bench eval dataset entry

Initial release: Kimi-K2.6-Eagle3 (Eagle head for Kimi-K2.6, 1.8B params)

Upload folder using huggingface_hub

initial commit