First, thank you for your excellent work on this quantization library and sharing your code.
Is there a quantization scheme available for the kimi-k2 model? From what I understand, its architecture is consistent with the deepseek-v3 model, differing only in the number of experts and dense blocks. In this case, is moe-quant still applicable?
First, thank you for your excellent work on this quantization library and sharing your code.
Is there a quantization scheme available for the kimi-k2 model? From what I understand, its architecture is consistent with the deepseek-v3 model, differing only in the number of experts and dense blocks. In this case, is moe-quant still applicable?