furyhawk/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-07-21 10:15:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

cuda-graph: GLM4_MOE - Don't load layer.nextn.embed_tokens

Browse Source
This commit is contained in:
Nexesenex
2026-06-28 01:55:55 +02:00
parent 3df798a914
commit fa7fe4eaaf
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-load-tensors.cpp
+1 -1
View File
@@ -2954,7 +2954,7 @@ bool create_tensors_helper::create_glm4_moe_tensors(const LLM_TN & tn) {
layer.nextn.embed_tokens = create_tensor(nextn_host_ctx,
tn(LLM_TENSOR_NEXTN_EMBED_TOKENS, "weight", final_layer),
{ n_embd, n_vocab },
flags | llama_model_loader::TENSOR_NOT_REQUIRED);
flags | llama_model_loader::TENSOR_NOT_REQUIRED | llama_model_loader::TENSOR_SKIP);
layer.nextn.enorm = create_tensor(nextn_ctx,
tn(LLM_TENSOR_NEXTN_ENORM, "weight", final_layer),
{ n_embd },