tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-28 12:25:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

server: fix tool-call of DeepSeek R1 Qwen, return reasoning_content (Command 7RB & DeepSeek R1) unless --reasoning-format none (#11607)

Browse Source 
* extract & return thoughts in reasoning_content field (unless --reasoning-format) for DeepSeek R1 & Command R7B

* tool-calls: add deepseek r1 template (models/templates/llama-cpp-deepseek-r1.jinja) + hackommodate broken official template

* tool-calls: accommodate variety of wrong tool call opening tags both R1 Qwen 32B and 7B distills like to spit out

* server/oai: ensure content is null when there are tool calls, and reasoning_content appears before content for readability

* tool-calls: add DeepSeek R1 Qwen distills to server/README.md & server tests

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Olivier Chafik
2025-02-13 10:05:16 +00:00
committed by GitHub
parent 27e8a23300
commit c7f460ab88
17 changed files with 1023 additions and 316 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
examples/server/server.cpp
View File

@ -173,6 +173,7 @@ struct slot_params {
{"grammar_trigger_words", grammar_trigger_words},
{"grammar_trigger_tokens", sampling.grammar_trigger_tokens},
{"preserved_tokens", sampling.preserved_tokens},
{"chat_format", common_chat_format_name(oaicompat_chat_format)},
{"samplers", samplers},
{"speculative.n_max", speculative.n_max},
{"speculative.n_min", speculative.n_min},
@ -724,9 +725,19 @@ struct server_task_result_cmpl_final : server_task_result {
msg.content = content;
}
json tool_calls;
json message {
{"role", "assistant"},
};
if (!msg.reasoning_content.empty()) {
message["reasoning_content"] = msg.reasoning_content;
}
if (msg.content.empty() && !msg.tool_calls.empty()) {
message["content"] = json();
} else {
message["content"] = msg.content;
}
if (!msg.tool_calls.empty()) {
tool_calls = json::array();
auto tool_calls = json::array();
for (const auto & tc : msg.tool_calls) {
tool_calls.push_back({
{"type", "function"},
@ -737,15 +748,7 @@ struct server_task_result_cmpl_final : server_task_result {
{"id", tc.id},
});
}
}
json message {
{"content", msg.content},
{"tool_calls", tool_calls},
{"role", "assistant"},
};
if (!msg.tool_plan.empty()) {
message["tool_plan"] = msg.tool_plan;
message["tool_calls"] = tool_calls;
}
json choice {
@ -4060,7 +4063,7 @@ int main(int argc, char ** argv) {
}
auto body = json::parse(req.body);
json data = oaicompat_completion_params_parse(body, params.use_jinja, ctx_server.chat_templates);
json data = oaicompat_completion_params_parse(body, params.use_jinja, params.reasoning_format, ctx_server.chat_templates);
return handle_completions_impl(
SERVER_TASK_TYPE_COMPLETION,
@ -4073,7 +4076,7 @@ int main(int argc, char ** argv) {
// same with handle_chat_completions, but without inference part
const auto handle_apply_template = [&ctx_server, &params, &res_ok](const httplib::Request & req, httplib::Response & res) {
auto body = json::parse(req.body);
json data = oaicompat_completion_params_parse(body, params.use_jinja, ctx_server.chat_templates);
json data = oaicompat_completion_params_parse(body, params.use_jinja, params.reasoning_format, ctx_server.chat_templates);
res_ok(res, {{ "prompt", std::move(data.at("prompt")) }});
};