ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-23 22:54:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

This seems to work

Browse Source
This commit is contained in:
Saood Karim
2025-08-30 04:45:31 -05:00
parent f529c3a808
commit 640f9b6485
2 changed files with 28 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
examples/server/server.cpp
View File

@@ -2572,7 +2572,7 @@ struct server_context {
GGML_ASSERT(slot.ga_n == 1);
// reuse any previously computed tokens that are common with the new prompt
slot.n_past = common_part(slot.cache_tokens, prompt_tokens);
slot.n_past = common_part(ctx, model, slot.cache_tokens, slot.prompt);
// push the prompt into the sampling context (do not apply grammar)
for (int i = 0; i < slot.n_past; ++i) {

27
examples/server/utils.hpp
View File

@@ -314,6 +314,33 @@ static size_t common_part(const std::string & a, const std::string & b) {
return i;
}
static size_t common_part(const llama_context * ctx, const llama_model * model, const std::vector<llama_token> & a, const std::string & b) {
size_t pos = 0;
size_t token_idx = 0;
for (const auto & token : a) {
std::string piece = llama_token_to_piece(ctx, token);
if (pos + piece.size() <= b.size() && b.compare(pos, piece.size(), piece) == 0) {
pos += piece.size();
token_idx++;
continue;
}
//Below is to handle the auto insert BOS case
if (token_idx == 0 && token == llama_token_bos(model)) {
token_idx++;
continue;
}
return token_idx;
}
return token_idx;
}
static bool ends_with(const std::string & str, const std::string & suffix) {
return str.size() >= suffix.size() && 0 == str.compare(str.size() - suffix.size(), suffix.size(), suffix);
}