Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

TP: enable tensor split for LFM2/LFM2MOE (#26993)

Browse Source 
Assisted-by: deepseek-v4-flash
This commit is contained in:
Chris Danis
2026-08-21 08:13:58 +03:00
committed by GitHub
parent a298422da7
commit b2e5e9b28b
2 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-arch.cpp
-2
View File
@@ -1062,8 +1062,6 @@ bool llm_arch_supports_sm_tensor(const llm_arch & arch) {
case LLM_ARCH_NEMOTRON_H: case LLM_ARCH_NEMOTRON_H:
case LLM_ARCH_NEMOTRON_H_MOE: case LLM_ARCH_NEMOTRON_H_MOE:
case LLM_ARCH_GRANITE_HYBRID: case LLM_ARCH_GRANITE_HYBRID:
case LLM_ARCH_LFM2:
case LLM_ARCH_LFM2MOE:
case LLM_ARCH_MINIMAX_01: case LLM_ARCH_MINIMAX_01:
case LLM_ARCH_MINIMAX_M2: case LLM_ARCH_MINIMAX_M2:
case LLM_ARCH_MINIMAX_M3: case LLM_ARCH_MINIMAX_M3:
src/llama-model.cpp
+4
View File
@@ -487,6 +487,10 @@ 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"); 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 (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) {
// 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, "");
}
return get_tensor_config_impl(GGML_BACKEND_SPLIT_AXIS_0, "ssm_out.weight"); return get_tensor_config_impl(GGML_BACKEND_SPLIT_AXIS_0, "ssm_out.weight");
} }
if (std::regex_match(tensor_name, pattern_ssm_conv1d)) { if (std::regex_match(tensor_name, pattern_ssm_conv1d)) {