Public/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-18 16:55:05 +02:00
Code Issues Packages Projects Releases Wiki Activity

memory : avoid allocating V cache for indexer (it's not used) (#28330)

Browse Source 
Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>
This commit is contained in:
fairydreaming
2026-09-10 10:55:46 +02:00
committed by GitHub
parent 72797e8919
commit 311d4211bf
1 changed files with 4 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-memory-hybrid-idx.cpp
+4
View File
@@ -55,6 +55,10 @@ llama_memory_hybrid_idx::llama_memory_hybrid_idx(
// K-shift must not rotate them while the stream copies in the same update still apply
hparams_idx.rope_type = LLAMA_ROPE_TYPE_NONE;
// fool llama_kv_cache into thinking this is a MLA cache, so it won't cache V tensors
hparams_idx.n_embd_head_k_mla_impl = model.hparams.indexer_head_size;
hparams_idx.n_embd_head_v_mla_impl = model.hparams.indexer_head_size;
LLAMA_LOG_INFO("%s: creating indexer KV cache, size = %u cells\n", __func__, kv_size);
return new llama_kv_cache(