EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

llama : remove KV cache defragmentation logic (#15473)

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov
2025-08-22 12:22:13 +03:00
committed by GitHub
parent ad5c975c2d
commit 9ebebef62f
16 changed files with 32 additions and 440 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
src/llama-kv-cache.h
View File

@@ -24,17 +24,6 @@ public:
// this callback is used to filter out layers that should not be included in the cache
using layer_filter_cb = std::function<bool(int32_t il)>;
struct defrag_info {
bool empty() const {
return ids.empty();
}
// contains information about which cell moves where:
// - cell i moves to ids[i]
// - if ids[i] == i || ids[i] == ids.size(), then cell i is not moved
std::vector<uint32_t> ids;
};
struct stream_copy_info {
bool empty() const {
assert(ssrc.size() == sdst.size());
@@ -173,7 +162,7 @@ public:
// return empty vector on failure
slot_info_vec_t prepare(const std::vector<llama_ubatch> & ubatches);
bool update(llama_context * lctx, bool do_shift, const defrag_info & dinfo, const stream_copy_info & sc_info);
bool update(llama_context * lctx, bool do_shift, const stream_copy_info & sc_info);
// find a slot of kv cells that can hold the ubatch
// if cont == true, then the slot must be continuous
@@ -254,9 +243,6 @@ private:
// model layer id -> KV cache layer id
std::unordered_map<int32_t, int32_t> map_layer_ids;
// return non-empty vector if cells have been moved
defrag_info defrag_prepare(int32_t n_max_nodes) const;
size_t total_size() const;
size_t size_k_bytes() const;
@@ -277,11 +263,6 @@ private:
llm_graph_result * res,
llama_context * lctx) const;
ggml_cgraph * build_graph_defrag(
llm_graph_result * res,
llama_context * lctx,
const defrag_info & dinfo) const;
struct cell_ranges_t {
uint32_t strm;
@@ -299,7 +280,6 @@ class llama_kv_cache_context : public llama_memory_context_i {
public:
// some shorthands
using slot_info_vec_t = llama_kv_cache::slot_info_vec_t;
using defrag_info = llama_kv_cache::defrag_info;
using stream_copy_info = llama_kv_cache::stream_copy_info;
// used for errors
@@ -314,7 +294,6 @@ public:
llama_kv_cache * kv,
llama_context * lctx,
bool do_shift,
defrag_info dinfo,
stream_copy_info sc_info);
// used to create a batch procesing context from a batch
@@ -374,8 +353,6 @@ private:
bool do_shift = false;
defrag_info dinfo;
stream_copy_info sc_info;
//