mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 28 Packages Projects Releases Wiki Activity

common : delegate assistant continuation to underlying template handlers (#23089)

Browse Source 
* common : delegate assistant continuation to template handler

* server : implement echo parameter to exclude assistant prefill in the response

* server : fix tests for prefill

* server : use existing llama template

* cont : clean up
This commit is contained in:
Aldehir Rojas
2026-05-17 07:36:05 -04:00
committed by GitHub
parent a6d6183dbc
commit 39cf5d6191
10 changed files with 1112 additions and 191 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/chat-auto-parser.h
+10 -5
View File
@@ -60,16 +60,21 @@ struct generation_params {
common_reasoning_format reasoning_format = COMMON_REASONING_FORMAT_AUTO;
bool stream = true;
std::string grammar;
bool add_generation_prompt = false;
bool enable_thinking = true;
std::chrono::system_clock::time_point now = std::chrono::system_clock::now();
std::string generation_prompt;
bool add_generation_prompt = false;
common_chat_continuation continue_final_message = COMMON_CHAT_CONTINUATION_NONE;
common_chat_msg continue_msg;
bool enable_thinking = true;
std::chrono::system_clock::time_point now = std::chrono::system_clock::now();
json extra_context;
bool add_bos = false;
bool add_eos = false;
bool is_inference = true;
bool add_inference = false;
bool mark_input = true; // whether to mark input strings in the jinja context
bool has_continuation() const {
return continue_final_message != COMMON_CHAT_CONTINUATION_NONE && !continue_msg.empty();
}
};
// ============================================================================
@@ -386,7 +391,7 @@ struct autoparser {
void analyze_template(const common_chat_template & tmpl);
// Build the PEG parser for this template
common_peg_arena build_parser(const generation_params & inputs) const;
common_peg_arena build_parser(const generation_params & inputs, const std::string & generation_prompt) const;
private:
// Collect tokens from entire analysis to preserve