furyhawk/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-07-21 02:05:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

cli: add --output option

Browse Source
This commit is contained in:
Xuan Son Nguyen
2026-07-09 13:44:21 +02:00
parent 259f2e2a53
commit 7ab9323d1d
3 changed files with 60 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/arg.cpp
+1 -1
View File
@@ -2849,7 +2849,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
params.out_file = value;
}
).set_examples({LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_CVECTOR_GENERATOR, LLAMA_EXAMPLE_EXPORT_LORA, LLAMA_EXAMPLE_TTS, LLAMA_EXAMPLE_FINETUNE,
LLAMA_EXAMPLE_RESULTS, LLAMA_EXAMPLE_EXPORT_GRAPH_OPS}));
LLAMA_EXAMPLE_RESULTS, LLAMA_EXAMPLE_EXPORT_GRAPH_OPS, LLAMA_EXAMPLE_CLI}));
add_opt(common_arg(
{"-ofreq", "--output-frequency"}, "N",
string_format("output the imatrix every N iterations (default: %d)", params.n_out_freq),
tools/cli/cli-context.cpp
+48 -6
View File
@@ -162,6 +162,14 @@ bool cli_context::init() {
fetch_server_props();
if (!params.out_file.empty()) {
output_file.emplace(params.out_file);
if (!output_file->is_open()) {
ui::show_error(string_format("failed to open output file '%s'", params.out_file.c_str()));
return false;
}
}
return true;
}
@@ -323,7 +331,14 @@ bool cli_context::stage_media_file(const std::string & fname, const std::string
return true;
}
bool cli_context::generate_completion(std::string & assistant_content, cli_timings & timings) {
void cli_context::write_output_file(const std::string & content) {
if (output_file) {
(*output_file) << content;
output_file->flush();
}
}
bool cli_context::generate_completion(generated_content & content_out, cli_timings & timings) {
json body = {
{"messages", impl->messages},
{"stream", true},
@@ -364,13 +379,14 @@ bool cli_context::generate_completion(std::string & assistant_content, cli_timin
if (delta.contains("reasoning_content") && delta.at("reasoning_content").is_string()) {
const std::string text = delta.at("reasoning_content").get<std::string>();
if (!text.empty()) {
content_out.reasoning += text;
a.push(ui::ASSISTANT_DISPLAY_MODE_REASONING, text);
}
}
if (delta.contains("content") && delta.at("content").is_string()) {
const std::string text = delta.at("content").get<std::string>();
if (!text.empty()) {
assistant_content += text;
content_out.content += text;
a.push(ui::ASSISTANT_DISPLAY_MODE_CONTENT, text);
}
}
@@ -520,10 +536,12 @@ int cli_context::run() {
continue;
}
ui::show_message(string_format("Loaded media from '%s'", fname.c_str()));
write_output_file(string_format("User: Added media: %s\n", fname.c_str()));
continue;
} else if (string_starts_with(buffer, "/read ")) {
std::string fname = string_strip(buffer.substr(6));
add_text_file(fname);
write_output_file(string_format("User: Added text file: %s\n", fname.c_str()));
continue;
} else if (string_starts_with(buffer, "/glob ")) {
std::error_code ec;
@@ -568,9 +586,11 @@ int cli_context::run() {
continue;
}
if (!add_text_file((rel_path / rel).string())) {
const std::string full_path = (curdir / rel).string();
if (!add_text_file(full_path)) {
continue;
}
write_output_file(string_format("User: Added text file: %s\n", full_path.c_str()));
if (++count >= FILE_GLOB_MAX_RESULTS) {
ui::show_error(string_format("Maximum number of globbed files allowed (%zu) reached.", FILE_GLOB_MAX_RESULTS));
@@ -586,16 +606,34 @@ int cli_context::run() {
// generate response
if (add_user_msg) {
push_user_message(cur_msg);
write_output_file(string_format("User:\n%s\n\n", cur_msg.c_str()));
cur_msg.clear();
}
cli_timings timings;
std::string assistant_content;
generate_completion(assistant_content, timings);
generated_content content;
generate_completion(content, timings);
impl->messages.push_back({
{"role", "assistant"},
{"content", assistant_content}
{"content", content.content}
});
if (output_file) {
std::string out_content = "Assistant:\n";
if (!content.reasoning.empty()) {
out_content += "[Start thinking]\n\n";
out_content += content.reasoning;
out_content += "[End thinking]\n\n";
}
out_content += content.content;
if (!out_content.empty() && out_content.back() != '\n') {
out_content += "\n";
}
out_content += "\n";
write_output_file(out_content);
}
if (params.show_timings) {
ui::show_info(string_format(
"\n[ Prompt: %.1f t/s | Generation: %.1f t/s ]",
@@ -619,4 +657,8 @@ void cli_context::shutdown() {
server->stop();
server.reset();
}
if (output_file) {
output_file->close();
output_file.reset();
}
}
tools/cli/cli-context.h
+11 -1
View File
@@ -9,6 +9,7 @@
#include <memory>
#include <optional>
#include <string>
#include <fstream>
struct cli_timings {
double prompt_per_second = 0.0;
@@ -32,6 +33,8 @@ struct cli_context {
bool has_audio = false;
bool has_video = false;
std::optional<std::ofstream> output_file;
cli_context(const common_params & params);
~cli_context();
@@ -49,7 +52,11 @@ struct cli_context {
static std::atomic<bool> & interrupted();
private:
bool generate_completion(std::string & assistant_content, cli_timings & timings);
struct generated_content {
std::string reasoning;
std::string content;
};
bool generate_completion(generated_content & content_out, cli_timings & timings);
void fetch_server_props();
void add_system_prompt();
void push_user_message(const std::string & text);
@@ -62,5 +69,8 @@ private:
// "image", "audio", "video"; returns false if the file cannot be read
bool stage_media_file(const std::string & fname, const std::string & type);
// no-op if output file is not set
void write_output_file(const std::string & content);
std::unique_ptr<cli_context_impl> impl;
};