happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Refactor to use a common function to do file output, which both outputs to file and selects different outputs for special token and plain text cases

This commit is contained in:
David Baker 2026-03-11 13:53:31 +00:00
parent a0319dd06c
commit c2baff9161
No known key found for this signature in database
GPG Key ID: 89298D31E5B7B548
1 changed files with 33 additions and 43 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

76
tools/cli/cli.cpp
View File

@ -58,6 +58,8 @@ struct cli_context {
task_params defaults; task_params defaults;
bool verbose_prompt; bool verbose_prompt;
common_reasoning_format reasoning_format; common_reasoning_format reasoning_format;
bool file_streaming = false;
std::ofstream * file_out = nullptr;
// thread for showing "loading" animation // thread for showing "loading" animation
std::atomic<bool> loading_show; std::atomic<bool> loading_show;
@ -78,7 +80,7 @@ struct cli_context {
reasoning_format = params.reasoning_format; reasoning_format = params.reasoning_format;
} }
std::string generate_completion(result_timings & out_timings, std::ofstream * file_out = nullptr) { std::string generate_completion(result_timings & out_timings, std::ofstream * file_to_use = nullptr) {
server_response_reader rd = ctx_server.get_response_reader(); server_response_reader rd = ctx_server.get_response_reader();
auto chat_params = format_chat(); auto chat_params = format_chat();
{ {
@ -108,16 +110,12 @@ struct cli_context {
} }
// check if we are doing file output // check if we are doing file output
bool file_streaming = (file_out != nullptr && file_out->is_open()); file_out = file_to_use;
if (file_streaming) { file_streaming = (file_out != nullptr && file_out->is_open());
if (defaults.special_characters) { append_file_out(
*file_out << chat_params.prompt; "[Prompt]: " + messages.back()["content"].get<std::string>() + "\n\n",
} chat_params.prompt
else { );
*file_out << "[Prompt]: " << messages.back()["content"].get<std::string>() << "\n\n";
}
file_out->flush();
}
// wait for first result // wait for first result
console::spinner::start(); console::spinner::start();
@ -149,60 +147,40 @@ struct cli_context {
if (is_thinking) { if (is_thinking) {
console::log("\n[End thinking]\n\n"); console::log("\n[End thinking]\n\n");
console::set_display(DISPLAY_TYPE_RESET); console::set_display(DISPLAY_TYPE_RESET);
if (file_streaming && is_thinking) { append_file_out("\n\n", "</think>");
if (defaults.special_characters) {
*file_out << "<\\think>";
}
else {
*file_out << "\n\n";
}
}
is_thinking = false; is_thinking = false;
} }
curr_content += diff.content_delta; curr_content += diff.content_delta;
console::log("%s", diff.content_delta.c_str()); console::log("%s", diff.content_delta.c_str());
console::flush(); console::flush();
if (file_streaming) { if (!content_started) {
if (!content_started && !defaults.special_characters) { append_file_out("[Assistant]: ", "");
*file_out << "[Assistant]: ";
}
content_started = true; content_started = true;
*file_out << diff.content_delta;
file_out->flush();
} }
append_file_out(diff.content_delta);
} }
if (!diff.reasoning_content_delta.empty()) { if (!diff.reasoning_content_delta.empty()) {
console::set_display(DISPLAY_TYPE_REASONING); console::set_display(DISPLAY_TYPE_REASONING);
std::string reasoning_delta = diff.reasoning_content_delta;
if (!is_thinking) { if (!is_thinking) {
console::log("[Start thinking]\n"); console::log("[Start thinking]\n");
if (file_streaming) { append_file_out("[Thinking]: ", "<think>");
if (defaults.special_characters) { if (reasoning_delta == "<think>") {
*file_out << "<think>"; reasoning_delta = "";
}
else {
*file_out << "[Thinking]: ";
}
} }
} }
is_thinking = true; is_thinking = true;
console::log("%s", diff.reasoning_content_delta.c_str()); console::log("%s", reasoning_delta.c_str());
console::flush(); console::flush();
if (file_streaming) { append_file_out(reasoning_delta);
*file_out << diff.reasoning_content_delta;
file_out->flush();
}
} }
} }
} }
auto res_final = dynamic_cast<server_task_result_cmpl_final *>(result.get()); auto res_final = dynamic_cast<server_task_result_cmpl_final *>(result.get());
if (res_final) { if (res_final) {
out_timings = std::move(res_final->timings); out_timings = std::move(res_final->timings);
if (file_streaming) { append_file_out("\n\n","");
if (!defaults.special_characters) {
*file_out << "\n\n";
}
file_out->flush();
}
break; break;
} }
result = rd.next(should_stop); result = rd.next(should_stop);
@ -229,6 +207,18 @@ struct cli_context {
} }
} }
void append_file_out(const std::string & content, const std::optional<std::string> & special_characters_content = std::nullopt) {
if (!file_streaming) {
return;
}
if (defaults.special_characters && special_characters_content.has_value()) {
*file_out << special_characters_content.value();
} else {
*file_out << content;
}
file_out->flush();
}
common_chat_params format_chat() { common_chat_params format_chat() {
auto meta = ctx_server.get_meta(); auto meta = ctx_server.get_meta();
auto & chat_params = meta.chat_params; auto & chat_params = meta.chat_params;