tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-16 13:12:51 -04:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into xsn/private_batch_api

Browse Source
This commit is contained in:
Xuan Son Nguyen
2025-03-13 15:55:18 +01:00
parent 46596caf6d be7c303410
commit 17f954c8e2
173 changed files with 26425 additions and 16117 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
common/speculative.cpp
View File

@@ -5,6 +5,7 @@
#include "sampling.h"
#include <cstring>
#include <algorithm>
#define SPEC_VOCAB_MAX_SIZE_DIFFERENCE 128
#define SPEC_VOCAB_CHECK_START_TOKEN_ID 5
@@ -172,7 +173,7 @@ llama_tokens common_speculative_gen_draft(
result.reserve(params.n_draft);
if (reuse_n == 0) {
llama_kv_cache_clear(ctx);
llama_kv_self_clear(ctx);
prompt.clear();
} else {
@@ -191,14 +192,14 @@ llama_tokens common_speculative_gen_draft(
}
if (reuse_i > 0) {
llama_kv_cache_seq_rm (ctx, 0, 0, reuse_i);
llama_kv_cache_seq_add(ctx, 0, reuse_i, -1, -reuse_i);
llama_kv_self_seq_rm (ctx, 0, 0, reuse_i);
llama_kv_self_seq_add(ctx, 0, reuse_i, -1, -reuse_i);
prompt.erase(prompt.begin(), prompt.begin() + reuse_i);
}
if (reuse_n < (int) prompt.size()) {
llama_kv_cache_seq_rm (ctx, 0, reuse_n, -1);
llama_kv_self_seq_rm (ctx, 0, reuse_n, -1);
prompt.erase(prompt.begin() + reuse_n, prompt.end());
}