mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 30 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.h
+21 -2
View File
@@ -89,6 +89,8 @@ struct common_chat_msg {
nlohmann::ordered_json to_json_oaicompat(bool concat_typed_text = false) const;
std::string render_content(const std::string & delimiter = "\n\n") const;
bool empty() const {
return content.empty() && content_parts.empty() && tool_calls.empty() && reasoning_content.empty() &&
tool_name.empty() && tool_call_id.empty();
@@ -164,12 +166,22 @@ enum common_chat_format {
COMMON_CHAT_FORMAT_COUNT, // Not a format, just the # formats
};
// Continuation method provided via `continue_final_message`
enum common_chat_continuation {
COMMON_CHAT_CONTINUATION_NONE,
COMMON_CHAT_CONTINUATION_AUTO,
COMMON_CHAT_CONTINUATION_REASONING,
COMMON_CHAT_CONTINUATION_CONTENT,
};
struct common_chat_templates_inputs {
std::vector<common_chat_msg> messages;
std::string grammar;
std::string json_schema;
bool add_generation_prompt = true;
bool use_jinja = true;
bool add_generation_prompt = true;
common_chat_continuation continue_final_message = COMMON_CHAT_CONTINUATION_NONE;
bool use_jinja = true;
// Parameters below only supported when use_jinja is true
std::vector<common_chat_tool> tools;
common_chat_tool_choice tool_choice = COMMON_CHAT_TOOL_CHOICE_AUTO;
@@ -207,6 +219,7 @@ struct common_chat_parser_params {
bool reasoning_in_content = false;
std::string generation_prompt;
bool parse_tool_calls = true;
bool echo = false; // Include assistant prefilled msg in output
bool debug = false; // Enable debug output for PEG parser
common_peg_arena parser = {};
common_chat_parser_params() = default;
@@ -267,6 +280,8 @@ std::vector<common_chat_msg> common_chat_msgs_parse_oaicompat(const nlohmann::or
std::vector<common_chat_tool> common_chat_tools_parse_oaicompat(const nlohmann::ordered_json & tools);
common_chat_continuation common_chat_continuation_parse(const nlohmann::ordered_json & value);
// DEPRECATED: only used in tests
nlohmann::ordered_json common_chat_msgs_to_json_oaicompat(const std::vector<common_chat_msg> & msgs, bool concat_typed_text = false);
@@ -279,6 +294,10 @@ std::string common_chat_template_direct_apply(
const common_chat_template & tmpl,
const autoparser::generation_params & inputs);
std::string common_chat_template_generation_prompt(
const common_chat_template & tmpl,
const autoparser::generation_params & inputs);
std::optional<common_chat_params> common_chat_try_specialized_template(
const common_chat_template & tmpl,
const std::string & src,