Public/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-20 01:31:31 +02:00
Code Issues Packages Projects Releases Wiki Activity

quant : allow quantization of ffn_gate_inp tensors

Browse Source
This commit is contained in:
Georgi Gerganov
2026-08-04 18:11:42 +03:00
parent 6b5224cfcc
commit 58137f0f47
1 changed files with 1 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-quant.cpp
+1 -5
View File
@@ -141,7 +141,7 @@ static tensor_category tensor_get_category(const std::string & tensor_name) {
if (tensor_name.find("ffn_up") != std::string::npos) {
return tensor_category::FFN_UP;
}
if (tensor_name.find("ffn_gate") != std::string::npos) {
if (tensor_name.find("ffn_gate") != std::string::npos && tensor_name.find("ffn_gate_inp") == std::string::npos) {
return tensor_category::FFN_GATE;
}
if (tensor_name.find("ffn_down") != std::string::npos) {
@@ -303,10 +303,6 @@ static bool tensor_allows_quantization(const llama_model_quantize_params * param
quantize &= params->quantize_output_tensor || name != "output.weight";
// do not quantize expert gating tensors
// NOTE: can't use LLM_TN here because the layer number is not known
quantize &= name.find("ffn_gate_inp.weight") == std::string::npos;
// do not quantize the i32 token-id -> expert-id routing table (DeepSeek-V4)
quantize &= name.find("ffn_gate_tid2eid.weight") == std::string::npos;