tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 03:55:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

server : added --no-prefill-assistant flag (#13608)

Browse Source 
* added no-prefill-assistant flag

* reworded documentation comment

* updated server README.md
This commit is contained in:
Isaac McFadyen
2025-05-17 17:59:48 -04:00
committed by GitHub
parent e3a7cf6c5b
commit 6a2bc8bfb7
5 changed files with 17 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
tools/server/utils.hpp
View File

@ -583,6 +583,7 @@ static json oaicompat_completion_params_parse(const json & body) {
static json oaicompat_completion_params_parse(
const json & body, /* openai api json semantics */
bool use_jinja,
bool prefill_assistant,
common_reasoning_format reasoning_format,
const struct common_chat_templates * tmpls,
bool allow_non_text,
@ -732,7 +733,7 @@ static json oaicompat_completion_params_parse(
// if the assistant message appears at the end of list, we do not add end-of-turn token
// for ex. this can be useful to modify the reasoning process in reasoning models
bool prefill_assistant_message = !inputs.messages.empty() && inputs.messages.back().role == "assistant";
bool prefill_assistant_message = !inputs.messages.empty() && inputs.messages.back().role == "assistant" && prefill_assistant;
common_chat_msg last_message;
if (prefill_assistant_message) {
last_message = inputs.messages.back();