From a4020d867f55bf4721ddd1c5df713f4ade14ad49 Mon Sep 17 00:00:00 2001 From: Yee Man Chan Date: Sat, 3 Jan 2026 08:27:29 +0800 Subject: [PATCH] don't quantize conv1d of Kimi Linear --- src/llama-quant.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/llama-quant.cpp b/src/llama-quant.cpp index 7b8bf6e524..bae907f92c 100644 --- a/src/llama-quant.cpp +++ b/src/llama-quant.cpp @@ -869,9 +869,9 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std:: quantize &= name != LLM_TN(model.arch)(LLM_TENSOR_POS_EMBD, "weight"); quantize &= name != LLM_TN(model.arch)(LLM_TENSOR_TOKEN_TYPES, "weight"); - // do not quantize Mamba's small yet 2D weights + // do not quantize Mamba /Kimi's small conv1d weights // NOTE: can't use LLM_TN here because the layer number is not known - quantize &= name.find("ssm_conv1d.weight") == std::string::npos; + quantize &= name.find("ssm_conv1d") == std::string::npos; quantize &= name.find("shortconv.conv.weight") == std::string::npos; // do not quantize RWKV's small yet 2D weights