ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-07 12:30:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Be able to set reduce op data type for split mode "graph" (#1087)

Browse Source 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-12-24 14:01:29 +01:00
committed by GitHub
parent 2421a7e12b
commit 1ace5b7526
7 changed files with 23 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
src/llama.cpp
View File

@@ -4055,6 +4055,7 @@ struct llama_context_params llama_context_default_params() {
/*.only_active_experts =*/ false,
/*.k_cache_hadamard =*/ false,
/*.split_mode_graph_scheduling =*/ false,
/*.split_mode_f16 =*/ true,
/*.abort_callback =*/ nullptr,
/*.abort_callback_data =*/ nullptr,
/*.offload_policy =*/ nullptr,
@@ -4344,6 +4345,7 @@ struct llama_context * llama_new_context_with_model(
cparams.graph_reuse = params.graph_reuse;
cparams.k_cache_hadamard = params.k_cache_hadamard;
cparams.split_mode_graph_scheduling = params.split_mode_graph_scheduling;
cparams.split_mode_f16 = params.split_mode_f16;
cparams.min_experts = params.min_experts;
cparams.thresh_experts = params.thresh_experts;
cparams.cuda_params = params.cuda_params;
@@ -4433,6 +4435,7 @@ struct llama_context * llama_new_context_with_model(
LLAMA_LOG_INFO("%s: graph_reuse = %d\n", __func__, cparams.graph_reuse);
LLAMA_LOG_INFO("%s: k_cache_hadam = %d\n", __func__, cparams.k_cache_hadamard);
LLAMA_LOG_INFO("%s: split_mode_graph_scheduling = %d\n", __func__, cparams.split_mode_graph_scheduling);
LLAMA_LOG_INFO("%s: split_mode_f16= %d\n", __func__, cparams.split_mode_f16);
LLAMA_LOG_INFO("%s: ser = %d, %g\n", __func__, cparams.min_experts, cparams.thresh_experts);
LLAMA_LOG_INFO("%s: freq_base = %.1f\n", __func__, cparams.rope_freq_base);
LLAMA_LOG_INFO("%s: freq_scale = %g\n", __func__, cparams.rope_freq_scale);