Pinned post
๐ค We just released Kimi-K3-W4AFP8 (@Kimi_Moonshot) on @huggingface !
We requantized Moonshot AIโs 2.8T MoE model into INT4-weight / FP8-activation (W4AFP8) optimized for SGLang. Get faster inference with virtually zero accuracy loss!
We requantized Moonshot AIโs 2.8T MoE model into INT4-weight / FP8-activation (W4AFP8) optimized for SGLang. Get faster inference with virtually zero accuracy loss!