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

llama : allocate indexer cache only in "full" indexer layers (#26474)

Browse Source 
Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>
This commit is contained in:
fairydreaming
2026-08-03 14:56:30 +02:00
committed by GitHub
co-authored by Stanisław Szymczyk
parent 96278e39fc
commit 563dec81c1
3 changed files with 11 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-model.cpp
+5 -3
View File
@@ -2101,10 +2101,11 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
} else {
// Main context: DSA cache for the trunk layers only - the nextn
// layer(s) are never attended by the trunk graph.
llama_kv_cache::layer_filter_cb filter = nullptr;
llama_kv_cache::layer_filter_cb filter_mla = nullptr;
if (hparams.n_layer_nextn > 0) {
filter = [&](uint32_t il) { return il < hparams.n_layer(); };
filter_mla = [&](uint32_t il) { return il < hparams.n_layer(); };
}
llama_kv_cache::layer_filter_cb filter_lid = [&](uint32_t il) { return il < hparams.n_layer() && (arch != LLM_ARCH_GLM_DSA || hparams.is_indexer_full(il)); };
res = new llama_kv_cache_dsa(
*this,
@@ -2118,7 +2119,8 @@ llama_memory_i * llama_model::create_memory(const llama_memory_params & params,
1,
hparams.n_swa,
hparams.swa_type,
filter,
filter_mla,
filter_lid,
nullptr);
}
} break;