wmantly/llama.cpp-xrip
1
0 Fork 0
Code Issues Pull Requests Actions 32 Packages Projects Releases Wiki Activity

llama : enable tensor split for LFM3 (#31431)

Browse Source 
Assisted-by: Claude Opus 5.5
This commit is contained in:
Tarek Dakhran authored and GitHub committed 2026-10-31 23:40:50 +01:00
1 parent f6312ad75e
commit bc428e407a
2 files changed
+1 -2
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/llama-arch.cpp
-1
View File
@@ -1375,7 +1375,6 @@ bool llm_arch_supports_sm_tensor(const llm_arch & arch) {
case LLM_ARCH_NEMOTRON_H:
case LLM_ARCH_NEMOTRON_H_MOE:
case LLM_ARCH_GRANITE_HYBRID:
case LLM_ARCH_LFM3:
case LLM_ARCH_MINIMAX_01:
case LLM_ARCH_MINIMAX_M2:
case LLM_ARCH_MINIMAX_M3:
src/llama-model.cpp
+1 -1
View File
@@ -591,7 +591,7 @@ struct ggml_backend_meta_split_state llama_meta_device_get_split_state(const str
return get_tensor_config_impl(GGML_BACKEND_SPLIT_AXIS_1, "ssm_out.weight");
}
if (std::regex_match(tensor_name, pattern_r_cache) || std::regex_match(tensor_name, pattern_s_cache)) {
if (ud->model->arch == LLM_ARCH_LFM2 || ud->model->arch == LLM_ARCH_LFM2MOE) {
if (ud->model->arch == LLM_ARCH_LFM2 || ud->model->arch == LLM_ARCH_LFM2MOE || ud->model->arch == LLM_ARCH_LFM3) {
// the LFM2 shortconv block runs fully mirrored, so its conv state must be mirrored too
return get_tensor_config_impl(GGML_BACKEND_SPLIT_AXIS_MIRRORED, "");
}