CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-08 10:07:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

memory : remove KV cache size padding

Browse Source
This commit is contained in:
Georgi Gerganov
2025-10-28 09:26:16 +02:00
parent 3479efd112
commit e28cec364e
4 changed files with 5 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/llama-kv-cache.cpp
View File

@@ -2010,8 +2010,3 @@ void llama_kv_cache_context::set_input_kq_mask(ggml_tensor * dst, const llama_ub
void llama_kv_cache_context::set_input_pos_bucket(ggml_tensor * dst, const llama_ubatch * ubatch) const {
kv->set_input_pos_bucket(dst, ubatch);
}
uint32_t llama_kv_cache::get_padding(const llama_cparams & cparams) {
// the FA kernels require padding to avoid extra runtime boundary checks
return cparams.flash_attn ? 256u : 32u;
}