CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-03 09:22:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama : more robust cell_max heuristic + wip shift

Browse Source
This commit is contained in:
Georgi Gerganov
2023-09-18 17:15:25 +03:00
parent 4d76d762ef
commit f015b26689
3 changed files with 39 additions and 52 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
llama.h
View File

@@ -321,7 +321,7 @@ extern "C" {
LLAMA_API DEPRECATED(int llama_get_kv_cache_token_count(const struct llama_context * ctx),
"avoid using this, it will be removed in the future, instead - count the tokens in user code");
// Remove all tokens between cells [c0, c1)
// Remove all tokens data of cells in [c0, c1)
LLAMA_API void llama_kv_cache_rm_tokens(struct llama_context * ctx, int32_t c0, int32_t c1);
// Removes all tokens that belong to the specified sequence
@@ -330,6 +330,10 @@ extern "C" {
// Removes all tokens that do not belong to the specified sequence
LLAMA_API void llama_kv_cache_keep_seq(struct llama_context * ctx, llama_seq_id seq_id);
// Adds relative position "delta" to all tokens that belong to the specified sequence and have positions in [p0, p1)
// If the KV cache is RoPEd, the KV data is updated accordingly
LLAMA_API void llama_kv_cache_shift(struct llama_context * ctx, llama_seq_id seq_id, llama_pos p0, llama_pos p1, llama_pos delta);
//
// State / sessions
//