ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-23 06:34:13 +00:00
Code Issues Packages Projects Releases Wiki Activity

Turn off merge_up_gate_exps if split mode graph

Browse Source 
It is not yet implemented
This commit is contained in:
Kawrakow
2026-01-12 13:00:50 +02:00
parent 7ad7d8339b
commit ec105a80bc
1 changed files with 8 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
src/llama-load-tensors.cpp
View File

@@ -2933,11 +2933,18 @@ bool create_tensors_helper::create_tensors() {
bool use_mmap_buffer = true;
if (ml.merge_qkv && (model.split_mode == LLAMA_SPLIT_MODE_GRAPH || model.split_mode == LLAMA_SPLIT_MODE_ATTN)) {
LLAMA_LOG_WARN("\n========================================================\n");
LLAMA_LOG_WARN("merge_qkv is not compatible with split model 'graph'\n");
LLAMA_LOG_WARN("merge_qkv is not compatible with split mode 'graph'\n");
LLAMA_LOG_WARN(" => turning off merge_qkv\n");
LLAMA_LOG_WARN("========================================================\n\n");
ml.merge_qkv = false;
}
if (ml.merge_up_gate_exps && (model.split_mode == LLAMA_SPLIT_MODE_GRAPH || model.split_mode == LLAMA_SPLIT_MODE_ATTN)) {
LLAMA_LOG_WARN("\n========================================================\n");
LLAMA_LOG_WARN("merge_up_gate_exps is not compatible with split mode 'graph'\n");
LLAMA_LOG_WARN(" => turning off merge_up_gate_exps\n");
LLAMA_LOG_WARN("========================================================\n\n");
ml.merge_up_gate_exps = false;
}
switch (model.arch) {
case LLM_ARCH_LLAMA:
case LLM_ARCH_REFACT: