vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge 01db429161 into d7b31a9d84

Browse source
This commit is contained in:
Olivier Chafik 2025-02-10 17:34:14 +08:00 • committed by GitHub
commit 41de60d728
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
17 changed files with 979 additions and 272 deletions
Download patch file Download diff file Expand all files Collapse all files

11
common/arg.cpp
View file

@ -1975,6 +1975,17 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
params.use_jinja = true;
}
).set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_MAIN}).set_env("LLAMA_ARG_JINJA"));
add_opt(common_arg(
{"--reasoning-format"}, "FORMAT",
"reasoning format (default: deepseek; allowed values: deepseek, none)\n"
"controls whether thought tags are extracted from the response, and in which format they're returned. 'none' leaves thoughts unparsed in `message.content`, 'deepseek' puts them in `message.reasoning_content` (for DeepSeek R1 & Command R7B only).\n"
"only supported for non-streamed responses",
[](common_params & params, const std::string & value) {
/**/ if (value == "deepseek") { params.reasoning_format = COMMON_REASONING_FORMAT_DEEPSEEK; }
else if (value == "none") { params.reasoning_format = COMMON_REASONING_FORMAT_NONE; }
else { std::invalid_argument("invalid value"); }
}
).set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_MAIN}).set_env("LLAMA_ARG_THINK"));
add_opt(common_arg(
{"--chat-template"}, "JINJA_TEMPLATE",
string_format(

231
common/chat.cpp
View file

@ -12,11 +12,13 @@ std::string common_chat_format_name(common_chat_format format) {
case COMMON_CHAT_FORMAT_LLAMA_3_X: return "Llama 3.x";
case COMMON_CHAT_FORMAT_LLAMA_3_X_WITH_BUILTIN_TOOLS: return "Llama 3.x with builtin tools";
case COMMON_CHAT_FORMAT_DEEPSEEK_R1: return "DeepSeek R1";
case COMMON_CHAT_FORMAT_DEEPSEEK_R1_EXTRACT_REASONING: return "DeepSeek R1 (extract reasoning)";
case COMMON_CHAT_FORMAT_FIREFUNCTION_V2: return "FireFunction v2";
case COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2: return "Functionary v3.2";
case COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1: return "Functionary v3.1 Llama 3.1";
case COMMON_CHAT_FORMAT_HERMES_2_PRO: return "Hermes 2 Pro";
case COMMON_CHAT_FORMAT_COMMAND_R7B: return "Command R7B";
case COMMON_CHAT_FORMAT_COMMAND_R7B_EXTRACT_REASONING: return "Command R7B (extract reasoning)";
default:
throw std::runtime_error("Unknown chat format");
}
@ -105,7 +107,6 @@ static common_chat_msg parse_json_tool_calls(
std::sregex_iterator rend;
std::sregex_iterator rit(it, end, function_regex);
if (rit == rend) {
fprintf(stderr, "No more tool calls found\n");
result.content += std::string(it, end);
break;
}
@ -115,14 +116,21 @@ static common_chat_msg parse_json_tool_calls(
json arguments;
if (!parse_json(it, end, arguments)) {
throw std::runtime_error("Failed to parse json tool call arguments");
throw std::runtime_error("Failed to parse json tool call arguments: " + input);
}
if (!std::regex_search(it, end, match, close_regex)) {
throw std::runtime_error("Malformed input, missing closing pattern");
throw std::runtime_error("Malformed input, missing closing pattern: " + input);
}
it = match.suffix().first;
result.tool_calls.push_back({name, arguments.is_string() ? arguments.get<std::string>() : arguments.dump(), /* id= */ ""});
}
if (!result.tool_calls.empty()) {
if (!string_strip(result.content).empty()) {
LOG_WRN("Content found with tool calls: %s", result.content.c_str());
}
result.content = "";
}
return result;
}
@ -382,22 +390,46 @@ static common_chat_params common_chat_params_init_command_r7b(const common_chat_
"<|END_THINKING|>",
"<|END_ACTION|>",
};
data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.format = COMMON_CHAT_FORMAT_COMMAND_R7B;
auto adjusted_messages = json::array();
for (const auto & msg : inputs.messages) {
auto has_reasoning_content = msg.contains("reasoning_content") && msg["reasoning_content"].is_string();
auto has_tool_calls = msg.contains("tool_calls") && msg["tool_calls"].is_array();
if (has_reasoning_content && has_tool_calls) {
auto adjusted_message = msg;
adjusted_message["tool_plan"] = msg["reasoning_content"];
adjusted_message.erase("reasoning_content");
adjusted_messages.push_back(adjusted_message);
} else {
adjusted_messages.push_back(msg);
}
}
data.prompt = apply(tmpl, adjusted_messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt, {});
data.format = inputs.extract_reasoning ? COMMON_CHAT_FORMAT_COMMAND_R7B_EXTRACT_REASONING : COMMON_CHAT_FORMAT_COMMAND_R7B;
return data;
}
static common_chat_msg common_chat_parse_command_r7b(const std::string & input) {
static std::regex response_regex("<\\|START_RESPONSE\\|>([\\s\\S\\n\\r]*?)<\\|END_RESPONSE\\|>");
static std::regex thought_action_regex("<\\|START_THINKING\\|>([\\s\\S\\n\\r]*?)<\\|END_THINKING\\|><\\|START_ACTION\\|>([\\s\\S\\n\\r]*?)<\\|END_ACTION\\|>");
static common_chat_msg common_chat_parse_command_r7b(const std::string & input, bool extract_reasoning) {
static std::regex thought_regex("(<\\|START_THINKING\\|>([\\s\\S\\n\\r]*?)<\\|END_THINKING\\|>)([\\s\\S\\n\\r]*)");
static std::regex action_regex("<\\|START_ACTION\\|>([\\s\\S\\n\\r]*?)<\\|END_ACTION\\|>");
static std::regex response_regex("(?:<\\|START_RESPONSE\\|>)?([\\s\\S\\n\\r]*?)<\\|END_RESPONSE\\|>");
std::smatch match;
common_chat_msg result;
result.role = "assistant";
if (std::regex_match(input, match, response_regex)) {
result.content = match[1].str();
} else if (std::regex_match(input, match, thought_action_regex)) {
result.tool_plan = match[1].str();
auto actions_str = match[2].str();
std::string rest = input;
if (std::regex_match(rest, match, thought_regex)) {
if (extract_reasoning) {
result.reasoning_content = match[2].str();
} else if (!match[2].str().empty()) {
// Let the unparsed thinking tags through in content only if their insides aren't empty.
result.content = match[1].str();
}
rest = match[3].str();
}
if (std::regex_match(rest, match, action_regex)) {
auto actions_str = match[1].str();
auto actions = json::parse(actions_str);
for (const auto & action : actions) {
result.tool_calls.push_back({
@ -406,9 +438,11 @@ static common_chat_msg common_chat_parse_command_r7b(const std::string & input)
/* .id = */ action["tool_call_id"],
});
}
} else if (std::regex_match(rest, match, response_regex)) {
auto response = match[1].str();
result.content += response;
} else {
LOG_ERR("Failed to parse command_r output");
result.content = input;
result.content += rest;
}
return result;
}
@ -546,34 +580,90 @@ static common_chat_msg common_chat_parse_llama_3_1(const std::string & input, bo
static common_chat_params common_chat_params_init_deepseek_r1(const common_chat_template & tmpl, const struct common_chat_inputs & inputs) {
common_chat_params data;
data.grammar_lazy = inputs.tool_choice != "required";
data.grammar = build_grammar([&](const common_grammar_builder & builder) {
std::vector<std::string> tool_rules;
foreach_function(inputs.tools, [&](const json & tool) {
const auto & function = tool["function"];
std::string name = function["name"];
auto parameters = function["parameters"];
auto args_rule = builder.add_schema(name + "-args", parameters);
tool_rules.push_back(builder.add_rule(name + "-call",
"\"<｜tool▁call▁begin｜>function<｜tool▁sep｜>" + name + "\\n```json\\n\" " + args_rule + " \"```<｜tool▁call▁end｜>\""));
});
data.grammar_triggers.push_back({"<｜tool▁calls▁begin｜>", /* .at_start = */ false});
data.preserved_tokens = {
"<｜tool▁sep｜>",
"<｜tool▁call▁end｜>",
};
builder.add_rule("root", "\"<｜tool▁calls▁begin｜>\" (" + string_join(tool_rules, " | ") + ")" + (inputs.parallel_tool_calls ? "*" : "") + " space");
}, grammar_options);
if (inputs.tools.is_array() && !inputs.tools.empty()) {
data.grammar_lazy = inputs.tool_choice != "required" && inputs.json_schema.is_null();
data.grammar = build_grammar([&](const common_grammar_builder & builder) {
std::vector<std::string> tool_rules;
foreach_function(inputs.tools, [&](const json & tool) {
const auto & function = tool["function"];
std::string name = function["name"];
auto parameters = function["parameters"];
auto args_rule = builder.add_schema(name + "-args", parameters);
tool_rules.push_back(builder.add_rule(name + "-call",
"\"<｜tool▁call▁begin｜>function<｜tool▁sep｜>" + name + "\\n"
"```json\\n\" " + args_rule + " \"```<｜tool▁call▁end｜>\""));
});
// Distill Qwen 7B & 32B models seem confused re/ syntax of their tool call opening tag,
// so we accept common variants (then it's all constrained)
builder.add_rule("root",
"( \"<｜tool▁calls▁begin｜>\" | \"<｜tool_calls_begin｜>\" | \"<｜tool calls begin｜>\" | \"<｜tool\\\\_calls\\\\_begin｜>\" ) "
"(" + string_join(tool_rules, " | ") + ")" + (inputs.parallel_tool_calls ? "*" : "") + " "
"\"<｜tool▁calls▁end｜>\""
" space");
data.grammar_triggers.push_back({"<｜tool▁calls▁begin｜>", /* .at_start = */ false});
data.grammar_triggers.push_back({"<｜tool_calls_begin｜>", /* .at_start = */ false});
data.grammar_triggers.push_back({"<｜tool calls begin｜>", /* .at_start = */ false});
data.grammar_triggers.push_back({"<｜tool\\_calls\\_begin｜>", /* .at_start = */ false});
data.preserved_tokens = {
"<think>",
"</think>",
"<｜tool▁sep｜>",
"<｜tool▁calls▁end｜",
"<｜tool▁call▁end｜>",
};
}, grammar_options);
}
auto prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
// Hacks to fix the official (broken) prompt.
// It is advisable to use --chat-template-file models/templates/llama-cpp-deepseek-r1.jinja instead,
// until the official template is fixed.
if (tmpl.source().find("{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}") != std::string::npos) {
// Don't leave the chat dangling after tool results
if (string_ends_with(prompt, "<｜tool▁outputs▁end｜>")) {
prompt += "<｜end▁of▁sentence｜>";
if (inputs.add_generation_prompt) {
prompt += "<｜Assistant｜>";
}
}
// Fix up tool call delta example added by Minja
prompt = std::regex_replace(
prompt,
std::regex("(<｜tool▁call▁end｜>)[\\s\\r\\n]*(<｜tool▁outputs▁begin｜>|<｜User｜>)"),
"$1<｜tool▁calls▁end｜><｜end▁of▁sentence｜>$2");
}
data.prompt = prompt;
data.format = COMMON_CHAT_FORMAT_DEEPSEEK_R1;
data.format = inputs.extract_reasoning ? COMMON_CHAT_FORMAT_DEEPSEEK_R1_EXTRACT_REASONING : COMMON_CHAT_FORMAT_DEEPSEEK_R1;
return data;
}
static common_chat_msg common_chat_parse_deepseek_r1(const std::string & input) {
static std::regex trigger_regex("<｜tool▁calls▁begin｜>");
static common_chat_msg common_chat_parse_deepseek_r1(const std::string & input, bool extract_reasoning) {
static std::regex function_regex("<｜tool▁call▁begin｜>function<｜tool▁sep｜>([^\n]+)\n```json\n");
static std::regex close_regex("```<｜tool▁call▁end｜>");
return parse_json_tool_calls(input, trigger_regex, function_regex, close_regex);
static std::regex close_regex("```[\\s\\r\\n]*<｜tool▁call▁end｜>");
static std::regex reasoning_content_regex("((?:<think>)?([\\s\\S\\r\\n]*?)</think>)?([\\s\\S\\r\\n]*)");
static std::regex tool_calls_regex("[\\s\\r\\n]*(?:<｜tool▁calls▁begin｜>|<｜tool_calls_begin｜>|<｜tool calls begin｜>|<｜tool\\\\_calls\\\\_begin｜>)([\\s\\S\\r\\n]*?)<｜tool▁calls▁end｜>");
common_chat_msg msg;
msg.role = "assistant";
std::smatch match;
if (std::regex_match(input, match, reasoning_content_regex)) {
std::string rest;
if (extract_reasoning) {
msg.reasoning_content = string_strip(match[2].str());
} else {
msg.content = match[1].str();
}
rest = match[3].str();
if (std::regex_search(rest, match, tool_calls_regex)) {
auto tool_calls = match[1].str();
auto msg2 = parse_json_tool_calls(tool_calls, std::nullopt, function_regex, close_regex);
msg.tool_calls = std::move(msg2.tool_calls);
} else {
msg.content += std::string(rest.begin() + rest.find_first_not_of(" \r\n"), rest.end());
}
} else {
msg.content = input;
}
return msg;
}
static common_chat_params common_chat_params_init_firefunction_v2(const common_chat_template & tmpl, const struct common_chat_inputs & inputs) {
@ -583,7 +673,7 @@ static common_chat_params common_chat_params_init_firefunction_v2(const common_c
{"datetime", "Jan 29 2025 13:00:00 GMT"},
{"functions", json(inputs.tools.empty() ? "" : inputs.tools.dump(2))},
});
if (!inputs.tools.is_null() && !inputs.tools.empty()) {
if (inputs.tools.is_array() && !inputs.tools.empty()) {
data.grammar_lazy = inputs.tool_choice != "required";
data.grammar = build_grammar([&](const common_grammar_builder & builder) {
auto schemas = json::array();
@ -628,7 +718,7 @@ static common_chat_params common_chat_params_init_functionary_v3_2(const common_
common_chat_params data;
data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
data.format = COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2;
if (!inputs.tools.is_null() && !inputs.tools.empty()) {
if (inputs.tools.is_array() && !inputs.tools.empty()) {
data.grammar_lazy = inputs.tool_choice != "required";
data.grammar = build_grammar([&](const common_grammar_builder & builder) {
std::vector<std::string> first_tool_rules;
@ -884,47 +974,72 @@ static common_chat_params common_chat_params_init_without_tools(const common_cha
}
common_chat_params common_chat_params_init(const common_chat_template & tmpl, const struct common_chat_inputs & inputs) {
auto has_tools = !inputs.tools.is_null() && inputs.tool_choice != "none";
LOG_DBG("[%s] has_tools=%s\n", __func__, has_tools ? "true" : "false");
const auto & src = tmpl.source();
const auto & caps = tmpl.original_caps();
if (has_tools && !inputs.grammar.empty()) {
throw std::runtime_error("Cannot specify grammar with tools");
if (inputs.tools.is_array()) {
if (inputs.tool_choice != "none" && !inputs.grammar.empty()) {
throw std::runtime_error("Cannot specify grammar with tools");
}
if (caps.supports_tool_calls && !caps.supports_tools) {
LOG_WRN("Template supports tool calls but does not natively describe tools. The fallback behaviour used may produce bad results, inspect prompt w/ --verbose & consider overriding the template.");
}
}
const auto & src = tmpl.source();
// DeepSeek R1: use handler in all cases except json schema (thinking / tools).
if (src.find("<｜tool▁calls▁begin｜>") != std::string::npos && inputs.json_schema.is_null()) {
return common_chat_params_init_deepseek_r1(tmpl, inputs);
}
// Command R7B: : use handler in all cases except json schema (thinking / tools).
if (src.find("<|END_THINKING|><|START_ACTION|>") != std::string::npos && inputs.json_schema.is_null()) {
return common_chat_params_init_command_r7b(tmpl, inputs);
}
// Use generic handler when mixing tools + JSON schema.
// TODO: support that mix in handlers below.
if ((!inputs.tools.is_array() && inputs.json_schema.is_object())) {
return common_chat_params_init_generic(tmpl, inputs);
}
// Functionary prepends "all\n" to plain content outputs, so we use its handler in all cases.
if (src.find(">>>all") != std::string::npos) {
// Functionary prepends "all\n" to plain content outputs, so we use the parser no matter when
return common_chat_params_init_functionary_v3_2(tmpl, inputs);
}
// Firefunction v2 requires datetime and functions in the context even w/o tools, so we also use its handler in all cases.
if (src.find(" functools[") != std::string::npos) {
// Firefunction v2 requires datetime and functions in the context, even w/o tools.
return common_chat_params_init_firefunction_v2(tmpl, inputs);
}
if (!has_tools) {
// Plain handler (no tools)
if (inputs.tools.is_null() || inputs.tool_choice == "none") {
return common_chat_params_init_without_tools(tmpl, inputs);
}
// Hermes 2/3 Pro, Qwen 2.5 Instruct (w/ tools)
if (src.find("<tool_call>") != std::string::npos) {
return common_chat_params_init_hermes_2_pro(tmpl, inputs);
}
// Functionary v3.1 (w/ tools)
if (src.find("<|start_header_id|>") != std::string::npos
&& src.find("<function=") != std::string::npos) {
return common_chat_params_init_functionary_v3_1_llama_3_1(tmpl, inputs);
}
// Llama 3.1, 3.2, 3.3 (w/ tools)
if (src.find("<|start_header_id|>ipython<|end_header_id|>") != std::string::npos) {
auto allow_python_tag_builtin_tools = src.find("<|python_tag|>") != std::string::npos;
return common_chat_params_init_llama_3_1_tool_calls(tmpl, inputs, allow_python_tag_builtin_tools);
}
if (src.find("<｜tool▁calls▁begin｜>") != std::string::npos) {
return common_chat_params_init_deepseek_r1(tmpl, inputs);
}
// Mistral Nemo (w/ tools)
if (src.find("[TOOL_CALLS]") != std::string::npos) {
return common_chat_params_init_mistral_nemo(tmpl, inputs);
}
if (src.find("<|END_THINKING|><|START_ACTION|>") != std::string::npos) {
return common_chat_params_init_command_r7b(tmpl, inputs);
}
// Generic fallback
return common_chat_params_init_generic(tmpl, inputs);
}
@ -949,7 +1064,9 @@ common_chat_msg common_chat_parse(const std::string & input, common_chat_format
case COMMON_CHAT_FORMAT_LLAMA_3_X_WITH_BUILTIN_TOOLS:
return common_chat_parse_llama_3_1(input, /* with_builtin_tools= */ true);
case COMMON_CHAT_FORMAT_DEEPSEEK_R1:
return common_chat_parse_deepseek_r1(input);
return common_chat_parse_deepseek_r1(input, /* extract_reasoning= */ false);
case COMMON_CHAT_FORMAT_DEEPSEEK_R1_EXTRACT_REASONING:
return common_chat_parse_deepseek_r1(input, /* extract_reasoning= */ true);
case COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2:
return common_chat_parse_functionary_v3_2(input);
case COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1:
@ -959,7 +1076,9 @@ common_chat_msg common_chat_parse(const std::string & input, common_chat_format
case COMMON_CHAT_FORMAT_FIREFUNCTION_V2:
return common_chat_parse_firefunction_v2(input);
case COMMON_CHAT_FORMAT_COMMAND_R7B:
return common_chat_parse_command_r7b(input);
return common_chat_parse_command_r7b(input, /* extract_reasoning= */ false);
case COMMON_CHAT_FORMAT_COMMAND_R7B_EXTRACT_REASONING:
return common_chat_parse_command_r7b(input, /* extract_reasoning= */ true);
default:
throw std::runtime_error("Unsupported format: " + common_chat_format_name(format));
}

3
common/chat.hpp
View file

@ -19,6 +19,7 @@ struct common_chat_inputs {
bool stream;
std::string grammar;
bool add_generation_prompt = true;
bool extract_reasoning = true;
};
enum common_chat_format {
@ -28,11 +29,13 @@ enum common_chat_format {
COMMON_CHAT_FORMAT_LLAMA_3_X,
COMMON_CHAT_FORMAT_LLAMA_3_X_WITH_BUILTIN_TOOLS,
COMMON_CHAT_FORMAT_DEEPSEEK_R1,
COMMON_CHAT_FORMAT_DEEPSEEK_R1_EXTRACT_REASONING,
COMMON_CHAT_FORMAT_FIREFUNCTION_V2,
COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2,
COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1,
COMMON_CHAT_FORMAT_HERMES_2_PRO,
COMMON_CHAT_FORMAT_COMMAND_R7B,
COMMON_CHAT_FORMAT_COMMAND_R7B_EXTRACT_REASONING,
COMMON_CHAT_FORMAT_COUNT, // Not a format, just the # formats
};

8
common/common.h
View file

@ -202,6 +202,11 @@ struct common_params_vocoder {
bool use_guide_tokens = false; // enable guide tokens to improve TTS accuracy // NOLINT
};
enum common_reasoning_format {
COMMON_REASONING_FORMAT_NONE,
COMMON_REASONING_FORMAT_DEEPSEEK, // Extract thinking tag contents and return as `message.reasoning_content`
};
struct common_params {
int32_t n_predict = -1; // new tokens to predict
int32_t n_ctx = 4096; // context size
@ -346,6 +351,7 @@ struct common_params {
std::string chat_template = ""; // NOLINT
bool use_jinja = false; // NOLINT
bool enable_chat_template = true;
common_reasoning_format reasoning_format = COMMON_REASONING_FORMAT_DEEPSEEK;
std::vector<std::string> api_keys;
@ -623,7 +629,7 @@ struct common_chat_msg {
std::string role;
std::string content;
std::vector<common_tool_call> tool_calls;
std::string tool_plan = "";
std::string reasoning_content = "";
};
// Check if the template supplied via "--chat-template" is supported or not. Returns true if it's valid

12
common/sampling.cpp
View file

@ -151,12 +151,6 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
lparams.no_perf = params.no_perf;
std::vector<const char *> trigger_words;
trigger_words.reserve(params.grammar_trigger_words.size());
for (const auto & str : params.grammar_trigger_words) {
trigger_words.push_back(str.word.c_str());
}
struct llama_sampler * grmr;
if (params.grammar.compare(0, 11, "%llguidance") == 0) {
#ifdef LLAMA_USE_LLGUIDANCE
@ -165,6 +159,12 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
GGML_ABORT("llguidance (cmake -DLLAMA_LLGUIDANCE=ON) is not enabled");
#endif // LLAMA_USE_LLGUIDANCE
} else {
std::vector<const char *> trigger_words;
trigger_words.reserve(params.grammar_trigger_words.size());
for (const auto & str : params.grammar_trigger_words) {
trigger_words.push_back(str.word.c_str());
}
grmr = params.grammar_lazy
? llama_sampler_init_grammar_lazy(vocab, params.grammar.c_str(), "root",
trigger_words.data(), trigger_words.size(),

301
examples/server/README.md
View file

@ -127,6 +127,7 @@ The project is under active development, and we are [looking for feedback and co
| `--grammar-file FNAME` | file to read grammar from |
| `-j, --json-schema SCHEMA` | JSON schema to constrain generations (https://json-schema.org/), e.g. `{}` for any JSON object<br/>For schemas w/ external $refs, use --grammar + example/json_schema_to_grammar.py instead |
| `--jinja` | Enable experimental Jinja templating engine (required for tool use) |
| `--reasoning-format FORMAT` | Controls extraction of model thinking traces and the format / field in which they are returned (default: `deepseek`; allowed values: `deepseek`, `none`; requires `--jinja`). `none` will leave thinking traces inline in `message.content` in a model-specific format, while `deepseek` will return them separately under `message.reasoning_content` |
**Example-specific params**
@ -1136,61 +1137,252 @@ curl http://localhost:8080/v1/chat/completions \
| Template | Format |
|----------|--------|
| CohereForAI-c4ai-command-r-plus-default.jinja | generic tool calls |
| CohereForAI-c4ai-command-r-plus-rag.jinja | generic tool calls |
| CohereForAI-c4ai-command-r-plus-tool_use.jinja | generic tool calls |
| MiniMaxAI-MiniMax-Text-01.jinja | generic tool calls |
| NexaAIDev-Octopus-v2.jinja | generic tool calls |
| NousResearch-Hermes-2-Pro-Llama-3-8B-default.jinja | generic tool calls |
| NousResearch-Hermes-2-Pro-Llama-3-8B-tool_use.jinja | hermes 2 pro tool calls |
| NousResearch-Hermes-2-Pro-Mistral-7B-default.jinja | generic tool calls |
| NousResearch-Hermes-2-Pro-Mistral-7B-tool_use.jinja | hermes 2 pro tool calls |
| NousResearch-Hermes-3-Llama-3.1-70B-default.jinja | generic tool calls |
| NousResearch-Hermes-3-Llama-3.1-70B-tool_use.jinja | hermes 2 pro tool calls |
| OrionStarAI-Orion-14B-Chat.jinja | generic tool calls |
| Qwen-QwQ-32B-Preview.jinja | hermes 2 pro tool calls |
| Qwen-Qwen2-7B-Instruct.jinja | generic tool calls |
| Qwen-Qwen2-VL-7B-Instruct.jinja | generic tool calls |
| Qwen-Qwen2.5-7B-Instruct.jinja | hermes 2 pro tool calls |
| Qwen-Qwen2.5-Math-7B-Instruct.jinja | hermes 2 pro tool calls |
| TheBloke-FusionNet_34Bx2_MoE-AWQ.jinja | generic tool calls |
| abacusai-Fewshot-Metamath-OrcaVicuna-Mistral.jinja | generic tool calls |
| bofenghuang-vigogne-2-70b-chat.jinja | generic tool calls |
| databricks-dbrx-instruct.jinja | generic tool calls |
| deepseek-ai-DeepSeek-Coder-V2-Instruct.jinja | generic tool calls |
| deepseek-ai-DeepSeek-R1-Distill-Llama-8B.jinja | deepseek r1 tool calls |
| deepseek-ai-DeepSeek-R1-Distill-Qwen-32B.jinja | deepseek r1 tool calls |
| deepseek-ai-DeepSeek-R1-Distill-Qwen-7B.jinja | deepseek r1 tool calls |
| deepseek-ai-DeepSeek-V2.5.jinja | deepseek r1 tool calls |
| deepseek-ai-deepseek-coder-33b-instruct.jinja | generic tool calls |
| google-gemma-2-2b-it.jinja | generic tool calls |
| google-gemma-7b-it.jinja | generic tool calls |
| indischepartij-MiniCPM-3B-OpenHermes-2.5-v2.jinja | generic tool calls |
| mattshumer-Reflection-Llama-3.1-70B.jinja | generic tool calls |
| meetkai-functionary-medium-v3.2.jinja | functionary v3.2 tool calls |
| meta-llama-Llama-3.1-8B-Instruct.jinja | llama 3.x tool calls (w/ builtin tools) |
| meta-llama-Llama-3.2-3B-Instruct.jinja | llama 3.x tool calls |
| meta-llama-Llama-3.3-70B-Instruct.jinja | llama 3.x tool calls (w/ builtin tools) |
| meta-llama-Meta-Llama-3.1-8B-Instruct.jinja | llama 3.x tool calls (w/ builtin tools) |
| microsoft-Phi-3-medium-4k-instruct.jinja | generic tool calls |
| microsoft-Phi-3-mini-4k-instruct.jinja | generic tool calls |
| microsoft-Phi-3-small-8k-instruct.jinja | generic tool calls |
| microsoft-Phi-3.5-mini-instruct.jinja | generic tool calls |
| microsoft-Phi-3.5-vision-instruct.jinja | generic tool calls |
| mistralai-Mistral-7B-Instruct-v0.2.jinja | generic tool calls |
| mistralai-Mistral-Large-Instruct-2407.jinja | mistral nemo tool calls |
| mistralai-Mistral-Large-Instruct-2411.jinja | generic tool calls |
| mistralai-Mistral-Nemo-Instruct-2407.jinja | mistral nemo tool calls |
| mistralai-Mixtral-8x7B-Instruct-v0.1.jinja | generic tool calls |
| mlabonne-AlphaMonarch-7B.jinja | generic tool calls |
| nvidia-Llama-3.1-Nemotron-70B-Instruct-HF.jinja | llama 3.x tool calls (w/ builtin tools) |
| openchat-openchat-3.5-0106.jinja | generic tool calls |
| teknium-OpenHermes-2.5-Mistral-7B.jinja | generic tool calls |
| Almawave-Velvet-14B.jinja | Hermes 2 Pro |
| AtlaAI-Selene-1-Mini-Llama-3.1-8B.jinja | Llama 3.x |
| CohereForAI-aya-expanse-8b.jinja | Generic |
| CohereForAI-c4ai-command-r-plus-default.jinja | Generic |
| CohereForAI-c4ai-command-r-plus-rag.jinja | Generic |
| CohereForAI-c4ai-command-r-plus-tool_use.jinja | Generic |
| CohereForAI-c4ai-command-r7b-12-2024-default.jinja | Command R7B (extract reasoning) |
| CohereForAI-c4ai-command-r7b-12-2024-rag.jinja | Command R7B (extract reasoning) |
| CohereForAI-c4ai-command-r7b-12-2024-tool_use.jinja | Command R7B (extract reasoning) |
| CohereForAI-c4ai-command-r7b-12-2024.jinja | Generic |
| DavieLion-Llama-3.2-1B-SPIN-iter3.jinja | Generic |
| Delta-Vector-Rei-12B.jinja | Mistral Nemo |
| EpistemeAI-Mistral-Nemo-Instruct-12B-Philosophy-Math.jinja | Mistral Nemo |
| FlofloB-83k_continued_pretraining_Qwen2.5-0.5B-Instruct_Unsloth_merged_16bit.jinja | Hermes 2 Pro |
| FlofloB-test_continued_pretraining_Phi-3-mini-4k-instruct_Unsloth_merged_16bit.jinja | Generic |
| HelpingAI-HAI-SER.jinja | Generic |
| HuggingFaceTB-SmolLM2-1.7B-Instruct.jinja | Generic |
| HuggingFaceTB-SmolLM2-135M-Instruct.jinja | Generic |
| HuggingFaceTB-SmolLM2-360M-Instruct.jinja | Generic |
| INSAIT-Institute-BgGPT-Gemma-2-27B-IT-v1.0.jinja | Generic |
| Ihor-Text2Graph-R1-Qwen2.5-0.5b.jinja | Hermes 2 Pro |
| Infinigence-Megrez-3B-Instruct.jinja | Generic |
| Josephgflowers-TinyLlama_v1.1_math_code-world-test-1.jinja | Generic |
| LGAI-EXAONE-EXAONE-3.5-2.4B-Instruct.jinja | Generic |
| LGAI-EXAONE-EXAONE-3.5-7.8B-Instruct.jinja | Generic |
| LatitudeGames-Wayfarer-12B.jinja | Generic |
| Magpie-Align-Llama-3-8B-Magpie-Align-v0.1.jinja | Generic |
| Magpie-Align-Llama-3.1-8B-Magpie-Align-v0.1.jinja | Generic |
| MaziyarPanahi-calme-3.2-instruct-78b.jinja | Generic |
| MiniMaxAI-MiniMax-Text-01.jinja | Generic |
| MiniMaxAI-MiniMax-VL-01.jinja | Generic |
| NaniDAO-deepseek-r1-qwen-2.5-32B-ablated.jinja | DeepSeek R1 (extract reasoning) |
| NexaAIDev-Octopus-v2.jinja | Generic |
| NousResearch-Hermes-2-Pro-Llama-3-8B-default.jinja | Generic |
| NousResearch-Hermes-2-Pro-Llama-3-8B-tool_use.jinja | Hermes 2 Pro |
| NousResearch-Hermes-2-Pro-Mistral-7B-default.jinja | Generic |
| NousResearch-Hermes-2-Pro-Mistral-7B-tool_use.jinja | Hermes 2 Pro |
| NousResearch-Hermes-3-Llama-3.1-70B-default.jinja | Generic |
| NousResearch-Hermes-3-Llama-3.1-70B-tool_use.jinja | Hermes 2 Pro |
| NovaSky-AI-Sky-T1-32B-Flash.jinja | Hermes 2 Pro |
| NovaSky-AI-Sky-T1-32B-Preview.jinja | Hermes 2 Pro |
| OnlyCheeini-greesychat-turbo.jinja | Generic |
| Orenguteng-Llama-3.1-8B-Lexi-Uncensored-V2.jinja | Llama 3.x |
| OrionStarAI-Orion-14B-Chat.jinja | Generic |
| PowerInfer-SmallThinker-3B-Preview.jinja | Generic |
| PrimeIntellect-INTELLECT-1-Instruct.jinja | Generic |
| Qwen-QVQ-72B-Preview.jinja | Generic |
| Qwen-QwQ-32B-Preview.jinja | Hermes 2 Pro |
| Qwen-Qwen1.5-7B-Chat.jinja | Generic |
| Qwen-Qwen2-7B-Instruct.jinja | Generic |
| Qwen-Qwen2-VL-72B-Instruct.jinja | Generic |
| Qwen-Qwen2-VL-7B-Instruct.jinja | Generic |
| Qwen-Qwen2.5-0.5B.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-1.5B-Instruct.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-14B-Instruct-1M.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-14B.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-32B-Instruct.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-32B.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-3B-Instruct.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-72B-Instruct.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-7B-Instruct-1M.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-7B-Instruct.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-7B.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-Coder-32B-Instruct.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-Coder-7B-Instruct.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-Math-1.5B.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-Math-7B-Instruct.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-VL-3B-Instruct.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-VL-72B-Instruct.jinja | Hermes 2 Pro |
| Qwen-Qwen2.5-VL-7B-Instruct.jinja | Hermes 2 Pro |
| RWKV-Red-Team-ARWKV-7B-Preview-0.1.jinja | Hermes 2 Pro |
| SakanaAI-TinySwallow-1.5B-Instruct.jinja | Hermes 2 Pro |
| SakanaAI-TinySwallow-1.5B.jinja | Hermes 2 Pro |
| Sao10K-70B-L3.3-Cirrus-x1.jinja | Llama 3.x |
| SentientAGI-Dobby-Mini-Leashed-Llama-3.1-8B.jinja | Llama 3.x |
| SentientAGI-Dobby-Mini-Unhinged-Llama-3.1-8B.jinja | Llama 3.x |
| Steelskull-L3.3-Damascus-R1.jinja | Llama 3.x |
| Steelskull-L3.3-MS-Nevoria-70b.jinja | Llama 3.x |
| Steelskull-L3.3-Nevoria-R1-70b.jinja | Llama 3.x |
| THUDM-glm-4-9b-chat.jinja | Generic |
| THUDM-glm-edge-1.5b-chat.jinja | Generic |
| Tarek07-Progenitor-V1.1-LLaMa-70B.jinja | Llama 3.x |
| TheBloke-FusionNet_34Bx2_MoE-AWQ.jinja | Generic |
| TinyLlama-TinyLlama-1.1B-Chat-v1.0.jinja | Generic |
| UCLA-AGI-Mistral7B-PairRM-SPPO-Iter3.jinja | Generic |
| ValiantLabs-Llama3.1-8B-Enigma.jinja | Llama 3.x |
| abacusai-Fewshot-Metamath-OrcaVicuna-Mistral.jinja | Generic |
| ai21labs-AI21-Jamba-1.5-Large.jinja | Generic |
| allenai-Llama-3.1-Tulu-3-405B-SFT.jinja | Generic |
| allenai-Llama-3.1-Tulu-3-405B.jinja | Generic |
| allenai-Llama-3.1-Tulu-3-8B.jinja | Generic |
| arcee-ai-Virtuoso-Lite.jinja | Hermes 2 Pro |
| arcee-ai-Virtuoso-Medium-v2.jinja | Hermes 2 Pro |
| arcee-ai-Virtuoso-Small-v2.jinja | Hermes 2 Pro |
| avemio-GRAG-NEMO-12B-ORPO-HESSIAN-AI.jinja | Generic |
| bespokelabs-Bespoke-Stratos-7B.jinja | Hermes 2 Pro |
| bfuzzy1-acheron-m1a-llama.jinja | Generic |
| bofenghuang-vigogne-2-70b-chat.jinja | Generic |
| bytedance-research-UI-TARS-72B-DPO.jinja | Generic |
| bytedance-research-UI-TARS-7B-DPO.jinja | Generic |
| bytedance-research-UI-TARS-7B-SFT.jinja | Generic |
| carsenk-phi3.5_mini_exp_825_uncensored.jinja | Generic |
| cyberagent-DeepSeek-R1-Distill-Qwen-14B-Japanese.jinja | DeepSeek R1 (extract reasoning) |
| cyberagent-DeepSeek-R1-Distill-Qwen-32B-Japanese.jinja | DeepSeek R1 (extract reasoning) |
| databricks-dbrx-instruct.jinja | Generic |
| deepseek-ai-DeepSeek-Coder-V2-Instruct.jinja | Generic |
| deepseek-ai-DeepSeek-Coder-V2-Lite-Base.jinja | Generic |
| deepseek-ai-DeepSeek-Coder-V2-Lite-Instruct.jinja | Generic |
| deepseek-ai-DeepSeek-R1-Distill-Llama-70B.jinja | DeepSeek R1 (extract reasoning) |
| deepseek-ai-DeepSeek-R1-Distill-Llama-8B.jinja | DeepSeek R1 (extract reasoning) |
| deepseek-ai-DeepSeek-R1-Distill-Qwen-1.5B.jinja | DeepSeek R1 (extract reasoning) |
| deepseek-ai-DeepSeek-R1-Distill-Qwen-14B.jinja | DeepSeek R1 (extract reasoning) |
| deepseek-ai-DeepSeek-R1-Distill-Qwen-32B.jinja | DeepSeek R1 (extract reasoning) |
| deepseek-ai-DeepSeek-R1-Distill-Qwen-7B.jinja | DeepSeek R1 (extract reasoning) |
| deepseek-ai-DeepSeek-R1-Zero.jinja | DeepSeek R1 (extract reasoning) |
| deepseek-ai-DeepSeek-R1.jinja | DeepSeek R1 (extract reasoning) |
| deepseek-ai-DeepSeek-V2-Lite.jinja | Generic |
| deepseek-ai-DeepSeek-V2.5.jinja | DeepSeek R1 (extract reasoning) |
| deepseek-ai-DeepSeek-V3.jinja | DeepSeek R1 (extract reasoning) |
| deepseek-ai-deepseek-coder-33b-instruct.jinja | Generic |
| deepseek-ai-deepseek-coder-6.7b-instruct.jinja | Generic |
| deepseek-ai-deepseek-coder-7b-instruct-v1.5.jinja | Generic |
| deepseek-ai-deepseek-llm-67b-chat.jinja | Generic |
| deepseek-ai-deepseek-llm-7b-chat.jinja | Generic |
| dicta-il-dictalm2.0-instruct.jinja | Generic |
| ehristoforu-Falcon3-8B-Franken-Basestruct.jinja | Hermes 2 Pro |
| fireworks-ai-llama-3-firefunction-v2.jinja | FireFunction v2 |
| godlikehhd-alpaca_data_sampled_ifd_new_5200.jinja | Hermes 2 Pro |
| godlikehhd-alpaca_data_score_max_0.7_2600.jinja | Hermes 2 Pro |
| google-gemma-2-27b-it.jinja | Generic |
| google-gemma-2-2b-it.jinja | Generic |
| google-gemma-2-2b-jpn-it.jinja | Generic |
| google-gemma-7b-it.jinja | Generic |
| huihui-ai-DeepSeek-R1-Distill-Llama-70B-abliterated.jinja | DeepSeek R1 (extract reasoning) |
| huihui-ai-DeepSeek-R1-Distill-Llama-8B-abliterated.jinja | DeepSeek R1 (extract reasoning) |
| huihui-ai-DeepSeek-R1-Distill-Qwen-14B-abliterated-v2.jinja | DeepSeek R1 (extract reasoning) |
| huihui-ai-DeepSeek-R1-Distill-Qwen-32B-abliterated.jinja | DeepSeek R1 (extract reasoning) |
| huihui-ai-DeepSeek-R1-Distill-Qwen-7B-abliterated-v2.jinja | DeepSeek R1 (extract reasoning) |
| huihui-ai-Qwen2.5-14B-Instruct-1M-abliterated.jinja | Hermes 2 Pro |
| ibm-granite-granite-3.1-8b-instruct.jinja | Generic |
| indischepartij-MiniCPM-3B-OpenHermes-2.5-v2.jinja | Generic |
| inflatebot-MN-12B-Mag-Mell-R1.jinja | Generic |
| jinaai-ReaderLM-v2.jinja | Generic |
| kms7530-chemeng_qwen-math-7b_24_1_100_1_nonmath.jinja | Hermes 2 Pro |
| knifeayumu-Cydonia-v1.3-Magnum-v4-22B.jinja | Mistral Nemo |
| langgptai-qwen1.5-7b-chat-sa-v0.1.jinja | Generic |
| lightblue-DeepSeek-R1-Distill-Qwen-7B-Japanese.jinja | DeepSeek R1 (extract reasoning) |
| mattshumer-Reflection-Llama-3.1-70B.jinja | Generic |
| meetkai-functionary-medium-v3.1.jinja | Functionary v3.1 Llama 3.1 |
| meetkai-functionary-medium-v3.2.jinja | Functionary v3.2 |
| meta-llama-Llama-2-7b-chat-hf.jinja | Generic |
| meta-llama-Llama-3.1-8B-Instruct.jinja | Llama 3.x |
| meta-llama-Llama-3.2-11B-Vision-Instruct.jinja | Llama 3.x |
| meta-llama-Llama-3.2-1B-Instruct.jinja | Llama 3.x |
| meta-llama-Llama-3.2-3B-Instruct.jinja | Llama 3.x |
| meta-llama-Llama-3.3-70B-Instruct.jinja | Llama 3.x |
| meta-llama-Meta-Llama-3-8B-Instruct.jinja | Generic |
| meta-llama-Meta-Llama-3.1-8B-Instruct.jinja | Llama 3.x |
| microsoft-Phi-3-medium-4k-instruct.jinja | Generic |
| microsoft-Phi-3-mini-4k-instruct.jinja | Generic |
| microsoft-Phi-3-small-8k-instruct.jinja | Generic |
| microsoft-Phi-3.5-mini-instruct.jinja | Generic |
| microsoft-Phi-3.5-vision-instruct.jinja | Generic |
| microsoft-phi-4.jinja | Generic |
| migtissera-Tess-3-Mistral-Nemo-12B.jinja | Generic |
| ministral-Ministral-3b-instruct.jinja | Generic |
| mistralai-Codestral-22B-v0.1.jinja | Generic |
| mistralai-Mistral-7B-Instruct-v0.1.jinja | Generic |
| mistralai-Mistral-7B-Instruct-v0.2.jinja | Generic |
| mistralai-Mistral-7B-Instruct-v0.3.jinja | Mistral Nemo |
| mistralai-Mistral-Large-Instruct-2407.jinja | Mistral Nemo |
| mistralai-Mistral-Large-Instruct-2411.jinja | Generic |
| mistralai-Mistral-Nemo-Instruct-2407.jinja | Mistral Nemo |
| mistralai-Mistral-Small-24B-Instruct-2501.jinja | Generic |
| mistralai-Mixtral-8x7B-Instruct-v0.1.jinja | Generic |
| mkurman-Qwen2.5-14B-DeepSeek-R1-1M.jinja | Hermes 2 Pro |
| mlabonne-AlphaMonarch-7B.jinja | Generic |
| mlx-community-Josiefied-Qwen2.5-0.5B-Instruct-abliterated-v1-float32.jinja | Hermes 2 Pro |
| mlx-community-Qwen2.5-VL-7B-Instruct-8bit.jinja | Hermes 2 Pro |
| mobiuslabsgmbh-DeepSeek-R1-ReDistill-Qwen-1.5B-v1.1.jinja | DeepSeek R1 (extract reasoning) |
| netcat420-MFANNv0.20.jinja | Generic |
| netcat420-MFANNv0.24.jinja | Generic |
| netease-youdao-Confucius-o1-14B.jinja | Hermes 2 Pro |
| nvidia-AceMath-7B-RM.jinja | Hermes 2 Pro |
| nvidia-Eagle2-1B.jinja | Hermes 2 Pro |
| nvidia-Eagle2-9B.jinja | Hermes 2 Pro |
| nvidia-Llama-3.1-Nemotron-70B-Instruct-HF.jinja | Llama 3.x |
| onnx-community-DeepSeek-R1-Distill-Qwen-1.5B-ONNX.jinja | DeepSeek R1 (extract reasoning) |
| open-thoughts-OpenThinker-7B.jinja | Hermes 2 Pro |
| openchat-openchat-3.5-0106.jinja | Generic |
| pankajmathur-orca_mini_v6_8b.jinja | Generic |
| princeton-nlp-Mistral-7B-Base-SFT-RDPO.jinja | Generic |
| princeton-nlp-Mistral-7B-Instruct-DPO.jinja | Generic |
| princeton-nlp-Mistral-7B-Instruct-RDPO.jinja | Generic |
| prithivMLmods-Bellatrix-Tiny-1.5B-R1.jinja | Hermes 2 Pro |
| prithivMLmods-Bellatrix-Tiny-1B-R1.jinja | Llama 3.x |
| prithivMLmods-Bellatrix-Tiny-1B-v3.jinja | Generic |
| prithivMLmods-Bellatrix-Tiny-3B-R1.jinja | Llama 3.x |
| prithivMLmods-Blaze-14B-xElite.jinja | Generic |
| prithivMLmods-Calcium-Opus-14B-Elite2-R1.jinja | Hermes 2 Pro |
| prithivMLmods-Calme-Ties-78B.jinja | Generic |
| prithivMLmods-Calme-Ties2-78B.jinja | Generic |
| prithivMLmods-Calme-Ties3-78B.jinja | Generic |
| prithivMLmods-ChemQwen2-vL.jinja | Generic |
| prithivMLmods-GWQ2b.jinja | Generic |
| prithivMLmods-LatexMind-2B-Codec.jinja | Generic |
| prithivMLmods-Llama-3.2-6B-AlgoCode.jinja | Llama 3.x |
| prithivMLmods-Megatron-Opus-14B-Exp.jinja | Hermes 2 Pro |
| prithivMLmods-Megatron-Opus-14B-Stock.jinja | Hermes 2 Pro |
| prithivMLmods-Megatron-Opus-7B-Exp.jinja | Hermes 2 Pro |
| prithivMLmods-Omni-Reasoner-Merged.jinja | Hermes 2 Pro |
| prithivMLmods-Omni-Reasoner4-Merged.jinja | Hermes 2 Pro |
| prithivMLmods-Primal-Opus-14B-Optimus-v1.jinja | Hermes 2 Pro |
| prithivMLmods-QwQ-Math-IO-500M.jinja | Hermes 2 Pro |
| prithivMLmods-Qwen-7B-Distill-Reasoner.jinja | DeepSeek R1 (extract reasoning) |
| prithivMLmods-Qwen2.5-1.5B-DeepSeek-R1-Instruct.jinja | Hermes 2 Pro |
| prithivMLmods-Qwen2.5-14B-DeepSeek-R1-1M.jinja | Hermes 2 Pro |
| prithivMLmods-Qwen2.5-32B-DeepSeek-R1-Instruct.jinja | Hermes 2 Pro |
| prithivMLmods-Qwen2.5-7B-DeepSeek-R1-1M.jinja | Hermes 2 Pro |
| prithivMLmods-Triangulum-v2-10B.jinja | Hermes 2 Pro |
| qingy2024-Falcon3-2x10B-MoE-Instruct.jinja | Hermes 2 Pro |
| rubenroy-Zurich-14B-GCv2-5m.jinja | Hermes 2 Pro |
| rubenroy-Zurich-7B-GCv2-5m.jinja | Hermes 2 Pro |
| silma-ai-SILMA-Kashif-2B-Instruct-v1.0.jinja | Generic |
| simplescaling-s1-32B.jinja | Hermes 2 Pro |
| sometimesanotion-Lamarck-14B-v0.7.jinja | Hermes 2 Pro |
| sonthenguyen-zephyr-sft-bnb-4bit-DPO-mtbr-180steps.jinja | Generic |
| sthenno-tempesthenno-icy-0130.jinja | Generic |
| sumink-qwft.jinja | Hermes 2 Pro |
| teknium-OpenHermes-2.5-Mistral-7B.jinja | Generic |
| thirdeyeai-elevate360m.jinja | Generic |
| tiiuae-Falcon3-10B-Instruct.jinja | Hermes 2 Pro |
| unsloth-DeepSeek-R1-Distill-Llama-8B-unsloth-bnb-4bit.jinja | DeepSeek R1 (extract reasoning) |
| unsloth-DeepSeek-R1-Distill-Llama-8B.jinja | DeepSeek R1 (extract reasoning) |
| unsloth-DeepSeek-R1.jinja | DeepSeek R1 (extract reasoning) |
| unsloth-Mistral-Small-24B-Instruct-2501-unsloth-bnb-4bit.jinja | Generic |
| upstage-solar-pro-preview-instruct.jinja | Generic |
| whyhow-ai-PatientSeek.jinja | Generic |
| xwen-team-Xwen-72B-Chat.jinja | Hermes 2 Pro |
| xwen-team-Xwen-7B-Chat.jinja | Hermes 2 Pro |
This table can be generated with:
```bash
./build/bin/test-chat ../minja/build/tests/*.jinja 2>/dev/null
```
</details>
@ -1202,11 +1394,20 @@ curl http://localhost:8080/v1/chat/completions \
```shell
# Native support:
llama-server --jinja -fa -hf bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M
llama-server --jinja -fa -hf bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q6_K_L
llama-server --jinja -fa -hf bartowski/functionary-small-v3.2-GGUF:Q4_K_M
llama-server --jinja -fa -hf bartowski/Llama-3.3-70B-Instruct-GGUF:Q4_K_M
# Native support for DeepSeek R1 works best w/ our own template (official template buggy)
llama-server --jinja -fa -hf bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q6_K_L \
--chat-template-file models/templates/llama-cpp-deepseek-r1.jinja
llama-server --jinja -fa -hf bartowski/DeepSeek-R1-Distill-Qwen-32B-GGUF:Q4_K_M \
--chat-template-file models/templates/llama-cpp-deepseek-r1.jinja
# Native support requires the right template for these GGUFs:
llama-server --jinja -fa -hf bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M \

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 == "" && !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 {
@ -4056,7 +4059,7 @@ 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,
@ -4069,7 +4072,7 @@ 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")) }});
};

177
examples/server/tests/unit/test_tool_call.py
View file

@ -92,6 +92,7 @@ def do_test_completion_with_required_tool_tiny(template_name: str, tool: dict, a
tool_calls = choice["message"].get("tool_calls")
assert tool_calls and len(tool_calls) == 1, f'Expected 1 tool call in {choice["message"]}'
tool_call = tool_calls[0]
assert choice["message"].get("content") is None, f'Expected no content in {choice["message"]}'
expected_function_name = "python" if tool["type"] == "code_interpreter" else tool["function"]["name"]
assert expected_function_name == tool_call["function"]["name"]
actual_arguments = tool_call["function"]["arguments"]
@ -155,11 +156,11 @@ def test_completion_with_required_tool_tiny_slow(template_name: str, tool: dict,
(TEST_TOOL, "success", "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M", ("NousResearch/Hermes-2-Pro-Llama-3-8B", "tool_use")),
(PYTHON_TOOL, "code", "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M", ("NousResearch/Hermes-2-Pro-Llama-3-8B", "tool_use")),
(PYTHON_TOOL, "code", "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M", "chatml"),
# (PYTHON_TOOL, "code", "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M", "chatml"),
(TEST_TOOL, "success", "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M", ("NousResearch/Hermes-3-Llama-3.1-8B", "tool_use")),
(PYTHON_TOOL, "code", "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M", ("NousResearch/Hermes-3-Llama-3.1-8B", "tool_use")),
(PYTHON_TOOL, "code", "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M", "chatml"),
# (PYTHON_TOOL, "code", "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M", "chatml"),
(TEST_TOOL, "success", "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M", None),
(PYTHON_TOOL, "code", "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M", None),
@ -175,7 +176,7 @@ def test_completion_with_required_tool_tiny_slow(template_name: str, tool: dict,
(TEST_TOOL, "success", "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M", ("meta-llama/Llama-3.2-3B-Instruct", None)),
(PYTHON_TOOL, "code", "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M", ("meta-llama/Llama-3.2-3B-Instruct", None)),
(PYTHON_TOOL, "code", "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M", "chatml"),
# (PYTHON_TOOL, "code", "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M", "chatml"),
# TODO: fix these
# (TEST_TOOL, "success", "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
# (PYTHON_TOOL, "code", "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
@ -214,6 +215,7 @@ def test_completion_with_required_tool_real_model(tool: dict, argument_key: str
tool_calls = choice["message"].get("tool_calls")
assert tool_calls and len(tool_calls) == 1, f'Expected 1 tool call in {choice["message"]}'
tool_call = tool_calls[0]
assert choice["message"].get("content") is None, f'Expected no content in {choice["message"]}'
expected_function_name = "python" if tool["type"] == "code_interpreter" else tool["function"]["name"]
assert expected_function_name == tool_call["function"]["name"]
actual_arguments = tool_call["function"]["arguments"]
@ -273,7 +275,6 @@ def test_completion_without_tool_call_slow(template_name: str, n_predict: int, t
@pytest.mark.slow
@pytest.mark.parametrize("hf_repo,template_override", [
("bartowski/c4ai-command-r7b-12-2024-GGUF:Q4_K_M", ("CohereForAI/c4ai-command-r7b-12-2024", "tool_use")),
("bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", None),
("bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", "chatml"),
@ -298,13 +299,16 @@ def test_completion_without_tool_call_slow(template_name: str, n_predict: int, t
("bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M", ("meta-llama/Llama-3.2-3B-Instruct", None)),
("bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M", "chatml"),
("bartowski/c4ai-command-r7b-12-2024-GGUF:Q6_K_L", ("CohereForAI/c4ai-command-r7b-12-2024", "tool_use")),
("bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
# Note: gemma-2-2b-it knows itself as "model", not "assistant", so we don't test the ill-suited chatml on it.
("bartowski/gemma-2-2b-it-GGUF:Q4_K_M", None),
# ("bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M", ("meta-llama/Llama-3.2-3B-Instruct", None)),
# ("bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
])
def test_weather(hf_repo: str, template_override: Tuple[str, str | None] | None):
def test_weather(hf_repo: str, template_override: str | Tuple[str, str | None] | None):
global server
n_predict = 512
server.n_slots = 1
@ -323,6 +327,7 @@ def test_weather(hf_repo: str, template_override: Tuple[str, str | None] | None)
res = server.make_request("POST", "/chat/completions", data={
"max_tokens": n_predict,
"messages": [
{"role": "system", "content": "You are a chatbot that uses tools/functions. Dont overthink things."},
{"role": "user", "content": "What is the weather in Istanbul?"},
],
"tools": [WEATHER_TOOL],
@ -332,6 +337,7 @@ def test_weather(hf_repo: str, template_override: Tuple[str, str | None] | None)
tool_calls = choice["message"].get("tool_calls")
assert tool_calls and len(tool_calls) == 1, f'Expected 1 tool call in {choice["message"]}'
tool_call = tool_calls[0]
assert choice["message"].get("content") is None, f'Expected no content in {choice["message"]}'
assert tool_call["function"]["name"] == WEATHER_TOOL["function"]["name"]
actual_arguments = json.loads(tool_call["function"]["arguments"])
assert 'location' in actual_arguments, f"location not found in {json.dumps(actual_arguments)}"
@ -340,22 +346,166 @@ def test_weather(hf_repo: str, template_override: Tuple[str, str | None] | None)
assert re.match('^Istanbul(, (TR|Turkey|Türkiye))?$', location), f'Expected Istanbul for location, got {location}'
@pytest.mark.slow
@pytest.mark.parametrize("result_override,n_predict,hf_repo,template_override", [
(None, 128, "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M", "chatml"),
(None, 128, "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M", None),
(None, 128, "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M", "chatml"),
(None, 128, "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M", ("NousResearch/Hermes-2-Pro-Llama-3-8B", "tool_use")),
(None, 128, "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M", ("NousResearch/Hermes-3-Llama-3.1-8B", "tool_use")),
(None, 128, "bartowski/functionary-small-v3.2-GGUF:Q8_0", ("meetkai/functionary-medium-v3.2", None)),
(None, 128, "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", None),
(None, 128, "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M", None),
("^> 0.56$", 128, "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M", "chatml"),
(None, 128, "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M", None),
# TODO: fix these (wrong results, either didn't respect decimal instruction or got wrong value)
("^The y-coordinate [\\s\\S]*?\\*\\*0.5\\*\\*", 8192, "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
("[\\s\\S]*?\\*\\*0\\.5\\*\\*", 8192, "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", ("llama-cpp-deepseek-r1", None)),
])
def test_calc_result(result_override: str | None, n_predict: int, hf_repo: str, template_override: str | Tuple[str, str | None] | None):
global server
# n_predict = 512
server.n_slots = 1
server.jinja = True
server.n_ctx = 8192 * 2
server.n_predict = n_predict
server.model_hf_repo = hf_repo
server.model_hf_file = None
if isinstance(template_override, tuple):
(template_hf_repo, template_variant) = template_override
server.chat_template_file = f"../../../models/templates/{template_hf_repo.replace('/', '-') + ('-' + template_variant if template_variant else '')}.jinja"
assert os.path.exists(server.chat_template_file), f"Template file {server.chat_template_file} does not exist. Run `python scripts/get_chat_template.py {template_hf_repo} {template_variant} > {server.chat_template_file}` to download the template."
elif isinstance(template_override, str):
server.chat_template = template_override
server.start(timeout_seconds=TIMEOUT_SERVER_START)
res = server.make_request("POST", "/chat/completions", data={
"max_tokens": n_predict,
"messages": [
{"role": "system", "content": "You are a chatbot that uses tools/functions. Dont overthink things, and provide very concise answers. Do not explain your reasoning to the user. Provide any numerical values back to the user with at most two decimals."},
{"role": "user", "content": "What's the y coordinate of a point on the unit sphere at angle 30 degrees?"},
{
"role": "assistant",
"content": None,
"tool_calls": [
{
"id": "call_6789",
"type": "function",
"function": {
"name": "calculate",
"arguments": "{\"expression\":\"sin(30 * pi / 180)\"}"
}
}
]
},
{
"role": "tool",
"name": "calculate",
"content": 0.55644242476,
"tool_call_id": "call_6789"
}
],
"tools": [
{
"type":"function",
"function":{
"name":"calculate",
"description":"A calculator function that computes values of arithmetic expressions in the Python syntax",
"parameters":{
"type":"object",
"properties":{
"expression":{
"type":"string",
"description":"An arithmetic expression to compute the value of (Python syntad, assuming all floats)"
}
},
"required":["expression"]
}
}
}
]
}, timeout=TIMEOUT_HTTP_REQUEST)
assert res.status_code == 200, f"Expected status code 200, got {res.status_code}"
choice = res.body["choices"][0]
tool_calls = choice["message"].get("tool_calls")
assert tool_calls is None, f'Expected no tool call in {choice["message"]}'
content = choice["message"].get("content")
assert content is not None, f'Expected content in {choice["message"]}'
if result_override is not None:
assert re.match(result_override, content), f'Expected {result_override}, got {content}'
else:
assert re.match('^[\\s\\S]*?The (y[ -])?coordinate [\\s\\S]*?is (approximately )?0\\.56\\b|^0\\.56$', content), \
f'Expected something like "The y coordinate is 0.56.", got {content}'
@pytest.mark.slow
@pytest.mark.parametrize("n_predict,reasoning_format,expect_content,expect_reasoning_content,hf_repo,template_override", [
(128, 'deepseek', "^The sum of 102 and 7 is 109.*", None, "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M", None),
(128, None, "^The sum of 102 and 7 is 109.*", None, "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M", None),
(1024, 'deepseek', "To find the sum of.*", "I need to calculate the sum of 102 and 7.*", "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
(1024, 'none', "<think>\n?I need[\\s\\S]*?</think>\n?To find.*", None, "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
(1024, 'deepseek', "To find the sum of.*", "First, I [\\s\\S]*", "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", ("llama-cpp-deepseek-r1", None)),
])
def test_thoughts(n_predict: int, reasoning_format: Literal['deepseek', 'none'] | None, expect_content: str | None, expect_reasoning_content: str | None, hf_repo: str, template_override: str | Tuple[str, str | None] | None):
global server
server.n_slots = 1
server.reasoning_format = reasoning_format
server.jinja = True
server.n_ctx = 8192 * 2
server.n_predict = n_predict
server.model_hf_repo = hf_repo
server.model_hf_file = None
if isinstance(template_override, tuple):
(template_hf_repo, template_variant) = template_override
server.chat_template_file = f"../../../models/templates/{template_hf_repo.replace('/', '-') + ('-' + template_variant if template_variant else '')}.jinja"
assert os.path.exists(server.chat_template_file), f"Template file {server.chat_template_file} does not exist. Run `python scripts/get_chat_template.py {template_hf_repo} {template_variant} > {server.chat_template_file}` to download the template."
elif isinstance(template_override, str):
server.chat_template = template_override
server.start(timeout_seconds=TIMEOUT_SERVER_START)
res = server.make_request("POST", "/chat/completions", data={
"max_tokens": n_predict,
"messages": [
{"role": "user", "content": "What's the sum of 102 and 7?"},
]
}, timeout=TIMEOUT_HTTP_REQUEST)
assert res.status_code == 200, f"Expected status code 200, got {res.status_code}"
choice = res.body["choices"][0]
assert choice["message"].get("tool_calls") is None, f'Expected no tool call in {choice["message"]}'
content = choice["message"].get("content")
if expect_content is None:
assert content is None, f'Expected no content in {choice["message"]}'
else:
assert re.match(expect_content, content), f'Expected {expect_content}, got {content}'
reasoning_content = choice["message"].get("reasoning_content")
if expect_reasoning_content is None:
assert reasoning_content is None, f'Expected no reasoning content in {choice["message"]}'
else:
assert re.match(expect_reasoning_content, reasoning_content), f'Expected {expect_reasoning_content}, got {reasoning_content}'
@pytest.mark.slow
@pytest.mark.parametrize("expected_arguments_override,hf_repo,template_override", [
(None, "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
# (None, "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", "chatml"),
(None, "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M", None),
(None, "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M", "chatml"),
(None, "bartowski/functionary-small-v3.2-GGUF:Q8_0", ("meetkai-functionary-medium-v3.2", None)),
(None, "bartowski/functionary-small-v3.2-GGUF:Q8_0", "chatml"),
(None, "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", None),
('{"code":"print("}', "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", "chatml"),
('{"code":"print("}', "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", None),
(None, "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", "chatml"),
('{"code":"print("}', "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M", ("meta-llama-Llama-3.2-3B-Instruct", None)),
(None, "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M", ("meta-llama-Llama-3.2-3B-Instruct", None)),
(None, "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M", "chatml"),
('{"code":"print("}', "bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M", ("meta-llama-Llama-3.2-3B-Instruct", None)),
('{"code":"print("}', "bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M", "chatml"),
(None, "bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M", "chatml"),
(None, "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M", None),
(None, "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M", "chatml"),
@ -371,15 +521,13 @@ def test_weather(hf_repo: str, template_override: Tuple[str, str | None] | None)
# Note: gemma-2-2b-it knows itself as "model", not "assistant", so we don't test the ill-suited chatml on it.
(None, "bartowski/gemma-2-2b-it-GGUF:Q4_K_M", None),
# (None, "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
])
def test_hello_world_tool_call(expected_arguments_override: str | None, hf_repo: str, template_override: str | Tuple[str, str | None] | None):
def test_hello_world(expected_arguments_override: str | None, hf_repo: str, template_override: str | Tuple[str, str | None] | None):
global server
server.n_slots = 1
server.jinja = True
server.n_ctx = 8192
server.n_predict = 128
server.n_predict = 512 # High because of DeepSeek R1
server.model_hf_repo = hf_repo
server.model_hf_file = None
if isinstance(template_override, tuple):
@ -406,6 +554,7 @@ def test_hello_world_tool_call(expected_arguments_override: str | None, hf_repo:
tool_calls = choice["message"].get("tool_calls")
assert tool_calls and len(tool_calls) == 1, f'Expected 1 tool call in {choice["message"]}'
tool_call = tool_calls[0]
assert choice["message"].get("content") is None, f'Expected no content in {choice["message"]}'
assert tool_call["function"]["name"] == PYTHON_TOOL["function"]["name"]
actual_arguments = tool_call["function"]["arguments"]
if expected_arguments_override is not None:

3
examples/server/tests/utils.py
View file

@ -78,6 +78,7 @@ class ServerProcess:
draft_max: int | None = None
no_webui: bool | None = None
jinja: bool | None = None
reasoning_format: Literal['deepseek', 'none'] | None = None
chat_template: str | None = None
chat_template_file: str | None = None
@ -172,6 +173,8 @@ class ServerProcess:
server_args.append("--no-webui")
if self.jinja:
server_args.append("--jinja")
if self.reasoning_format is not None:
server_args.extend(("--reasoning-format", self.reasoning_format))
if self.chat_template:
server_args.extend(["--chat-template", self.chat_template])
if self.chat_template_file:

8
examples/server/utils.hpp
View file

@ -578,6 +578,7 @@ static json oaicompat_completion_params_parse(const json & body) {
static json oaicompat_completion_params_parse(
const json & body, /* openai api json semantics */
bool use_jinja,
common_reasoning_format reasoning_format,
const common_chat_templates & chat_templates)
{
json llama_params;
@ -633,9 +634,10 @@ static json oaicompat_completion_params_parse(
throw std::runtime_error("Cannot use custom grammar constraints with tools.");
}
common_chat_inputs inputs;
inputs.messages = body.at("messages");
inputs.tools = tools;
inputs.tool_choice = tool_choice;
inputs.extract_reasoning = reasoning_format != COMMON_REASONING_FORMAT_NONE;
inputs.messages = body.at("messages");
inputs.tools = tools;
inputs.tool_choice = tool_choice;
inputs.parallel_tool_calls = json_value(body, "parallel_tool_calls", false);
if (inputs.parallel_tool_calls && !tmpl.original_caps().supports_parallel_tool_calls) {
LOG_DBG("Disabling parallel_tool_calls because the template does not support it\n");

22
models/templates/README.md Normal file
View file

@ -0,0 +1,22 @@
These templates can be updated with the following commands:
```bash
./scripts/get_chat_template.py CohereForAI/c4ai-command-r-plus tool_use > models/templates/CohereForAI-c4ai-command-r-plus-tool_use.jinja
./scripts/get_chat_template.py CohereForAI/c4ai-command-r7b-12-2024 default > models/templates/CohereForAI-c4ai-command-r7b-12-2024-default.jinja
./scripts/get_chat_template.py CohereForAI/c4ai-command-r7b-12-2024 rag > models/templates/CohereForAI-c4ai-command-r7b-12-2024-rag.jinja
./scripts/get_chat_template.py CohereForAI/c4ai-command-r7b-12-2024 tool_use > models/templates/CohereForAI-c4ai-command-r7b-12-2024-tool_use.jinja
./scripts/get_chat_template.py deepseek-ai/DeepSeek-R1-Distill-Llama-8B > models/templates/deepseek-ai-DeepSeek-R1-Distill-Llama-8B.jinja
./scripts/get_chat_template.py deepseek-ai/DeepSeek-R1-Distill-Qwen-32B > models/templates/deepseek-ai-DeepSeek-R1-Distill-Qwen-32B.jinja
./scripts/get_chat_template.py fireworks-ai/llama-3-firefunction-v2 > models/templates/fireworks-ai-llama-3-firefunction-v2.jinja
./scripts/get_chat_template.py google/gemma-2-2b-it > models/templates/google-gemma-2-2b-it.jinja
./scripts/get_chat_template.py meetkai/functionary-medium-v3. > models/templates/meetkai-functionary-medium-v3.jinja
./scripts/get_chat_template.py meetkai/functionary-medium-v3.2 > models/templates/meetkai-functionary-medium-v3.2.jinja
./scripts/get_chat_template.py meta-llama/Llama-3.1-8B-Instruct > models/templates/meta-llama-Llama-3.1-8B-Instruct.jinja
./scripts/get_chat_template.py meta-llama/Llama-3.2-3B-Instruct > models/templates/meta-llama-Llama-3.2-3B-Instruct.jinja
./scripts/get_chat_template.py meta-llama/Llama-3.3-70B-Instruct > models/templates/meta-llama-Llama-3.3-70B-Instruct.jinja
./scripts/get_chat_template.py microsoft/Phi-3.5-mini-instruct > models/templates/microsoft-Phi-3.5-mini-instruct.jinja
./scripts/get_chat_template.py mistralai/Mistral-Nemo-Instruct-2407 > models/templates/mistralai-Mistral-Nemo-Instruct-2407.jinja
./scripts/get_chat_template.py NousResearch/Hermes-2-Pro-Llama-3-8B tool_use > models/templates/NousResearch-Hermes-2-Pro-Llama-3-8B-tool_use.jinja
./scripts/get_chat_template.py NousResearch/Hermes-3-Llama-3.1-8B tool_use > models/templates/NousResearch-Hermes-3-Llama-3.1-8B-tool_use.jinja
./scripts/get_chat_template.py Qwen/Qwen2.5-7B-Instruct > models/templates/Qwen-Qwen2.5-7B-Instruct.jinja
```

2
models/templates/deepseek-ai-DeepSeek-R1-Distill-Llama-8B.jinja
View file

@ -1 +1 @@
{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜>'}}{% endif %}
{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜><think>\n'}}{% endif %}

57
models/templates/deepseek-ai-DeepSeek-R1-Distill-Qwen-32B.jinja
View file

@ -1,56 +1 @@
{% if not add_generation_prompt is defined %}
{% set add_generation_prompt = false %}
{% endif %}
{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}
{%- for message in messages %}
{%- if message['role'] == 'system' %}
{% set ns.system_prompt = message['content'] %}
{%- endif %}
{%- endfor %}
{{bos_token}}
{{ns.system_prompt}}
{%- for message in messages %}
{%- if message['role'] == 'user' %}
{%- set ns.is_tool = false -%}
{{'<｜User｜>' + message['content']}}
{%- endif %}
{%- if message['role'] == 'assistant' and message['content'] is none %}
{%- set ns.is_tool = false -%}
{%- for tool in message['tool_calls']%}
{%- if not ns.is_first %}
{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}
{%- set ns.is_first = true -%}
{%- else %}
{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}
{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}
{%- endif %}
{%- endfor %}
{%- endif %}
{%- if message['role'] == 'assistant' and message['content'] is not none %}
{%- if ns.is_tool %}
{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}
{%- set ns.is_tool = false -%}
{%- else %}
{% set content = message['content'] %}
{% if '</think>' in content %}
{% set content = content.split('</think>')[-1] %}
{% endif %}
{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}
{%- endif %}
{%- endif %}
{%- if message['role'] == 'tool' %}
{%- set ns.is_tool = true -%}
{%- if ns.is_output_first %}
{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}
{%- set ns.is_output_first = false %}
{%- else %}
{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}
{%- endif %}
{%- endif %}
{%- endfor -%}
{% if ns.is_tool %}
{{'<｜tool▁outputs▁end｜>'}}
{% endif %}
{% if add_generation_prompt and not ns.is_tool %}
{{'<｜Assistant｜>'}}
{% endif %}
{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + tool['function']['arguments'] + '\n' + '```' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜><think>\n'}}{% endif %}

76
models/templates/llama-cpp-deepseek-r1.jinja Normal file
View file

@ -0,0 +1,76 @@
{%- if not add_generation_prompt is defined -%}
{%- set add_generation_prompt = false -%}
{%- endif -%}
{%- set ns = namespace(is_first=false, is_tool_outputs=false, is_output_first=true, system_prompt='') -%}
{%- for message in messages -%}
{%- if message['role'] == 'system' -%}
{%- set ns.system_prompt = message['content'] -%}
{%- endif -%}
{%- endfor -%}
{{bos_token}}
{%- if tools %}
You can call any of the following function tools to satisfy the user's requests: {{tools | map(attribute='function') | tojson(indent=2)}}
Example function tool call syntax:
<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>function<｜tool▁sep｜>example_function_name
```json
{
"arg1": "some_value"
...
}
```
<｜tool▁call▁end｜><｜tool▁calls▁end｜>
{% endif -%}
{{ns.system_prompt}}
{%- macro flush_tool_outputs() -%}
{%- if ns.is_tool_outputs -%}
{{- '<｜tool▁outputs▁end｜><｜end▁of▁sentence｜>' -}}
{%- set ns.is_tool_outputs = false -%}
{%- endif -%}
{%- endmacro -%}
{{- flush_tool_outputs() -}}
{%- for message in messages -%}
{%- if message['role'] != 'tool' -%}
{{- flush_tool_outputs() -}}
{%- endif -%}
{%- if message['role'] == 'user' -%}
{{- '<｜User｜>' + message['content'] + '<｜end▁of▁sentence｜>' -}}
{%- endif -%}
{%- if message['role'] == 'assistant' and message['content'] is none -%}
{{- '<｜Assistant｜><｜tool▁calls▁begin｜>' -}}
{%- set ns.is_first = true -%}
{%- for tc in message['tool_calls'] -%}
{%- if ns.is_first -%}
{%- set ns.is_first = false -%}
{%- else -%}
{{- '\n' -}}
{%- endif -%}
{%- set tool_name = tc['function']['name'] -%}
{%- set tool_args = tc['function']['arguments'] -%}
{{- '<｜tool▁call▁begin｜>' + tc['type'] + '<｜tool▁sep｜>' + tool_name + '\n' + '```json' + '\n' + tool_args + '\n' + '```' + '<｜tool▁call▁end｜>' -}}
{%- endfor -%}
{{- '<｜tool▁calls▁end｜><｜end▁of▁sentence｜>' -}}
{%- endif -%}
{%- if message['role'] == 'assistant' and message['content'] is not none -%}
{{- flush_tool_outputs() -}}
{%- set content = message['content'] -%}
{%- if '</think>' in content -%}
{%- set content = content.split('</think>')[-1] -%}
{%- endif -%}
{{- '<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>' -}}
{%- endif -%}
{%- if message['role'] == 'tool' -%}
{%- set ns.is_tool_outputs = true -%}
{%- if ns.is_output_first -%}
{{- '<｜tool▁outputs▁begin｜>' -}}
{%- set ns.is_output_first = false -%}
{%- endif -%}
{{- '\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>' -}}
{%- endif -%}
{%- endfor -%}
{{- flush_tool_outputs() -}}
{%- if add_generation_prompt and not ns.is_tool_outputs -%}
{{- '<｜Assistant｜><think>\n' -}}
{%- endif -%}

5
scripts/get_chat_template.py Normal file → Executable file
View file

@ -7,9 +7,8 @@
./scripts/get_chat_template.py model_id [variant]
Examples:
./scripts/get_chat_template.py NousResearch/Meta-Llama-3-8B-Instruct
./scripts/get_chat_template.py NousResearch/Hermes-3-Llama-3.1-8B tool_use
./scripts/get_chat_template.py meta-llama/Llama-3.2-3B-Instruct
./scripts/get_chat_template.py CohereForAI/c4ai-command-r-plus tool_use
./scripts/get_chat_template.py microsoft/Phi-3.5-mini-instruct
'''
import json

2
src/llama-grammar.cpp
View file

@ -1186,7 +1186,7 @@ void llama_grammar_accept_impl(struct llama_grammar & grammar, llama_token token
return;
}
}
LLAMA_LOG_DEBUG("Grammar still awaiting trigger after token %d (`%s`) (buffer: `%s`)\n", token, piece.c_str(), grammar.trigger_buffer.c_str());
LLAMA_LOG_DEBUG("Grammar still awaiting trigger after token %d (`%s`)\n", token, piece.c_str());
return;
}
}

304
tests/test-chat.cpp
View file

@ -24,7 +24,10 @@ static common_chat_msg msg_from_json(const json & message) {
ret.content = message.at("content");
}
if (message.contains("tool_plan")) {
ret.tool_plan = message.at("tool_plan");
ret.reasoning_content = message.at("tool_plan");
}
if (message.contains("reasoning_content")) {
ret.reasoning_content = message.at("reasoning_content");
}
auto has_tool_calls = message.contains("tool_calls");
if (has_tool_calls) {
@ -105,6 +108,7 @@ static std::string dump(const json & j) {
static void assert_msg_equals(const common_chat_msg & expected, const common_chat_msg & actual) {
assert_equals(expected.role, actual.role);
assert_equals(expected.content, actual.content);
assert_equals(expected.reasoning_content, actual.reasoning_content);
assert_equals(expected.tool_calls.size(), actual.tool_calls.size());
for (size_t i = 0; i < expected.tool_calls.size(); i++) {
const auto & expected_tool_call = expected.tool_calls[i];
@ -176,13 +180,15 @@ struct delta_data {
static delta_data init_delta(const common_chat_template & tmpl, const std::vector<std::string> & end_tokens,
const json & user_message, const json & delta_message, const json & tools,
const json & tool_choice) {
const json & tool_choice,
bool think = false) {
common_chat_inputs inputs;
inputs.parallel_tool_calls = true;
inputs.messages = json::array();
inputs.messages.push_back(user_message);
inputs.tools = tools;
inputs.tool_choice = tool_choice;
inputs.extract_reasoning = think;
auto params_prefix = common_chat_params_init(tmpl, inputs);
inputs.messages.push_back(delta_message);
@ -192,17 +198,24 @@ static delta_data init_delta(const common_chat_template & tmpl, const std::vecto
std::string prefix = params_prefix.prompt;
std::string full = params_full.prompt;
// Check full starts with prefix
if (full.find(prefix) != 0) {
fprintf(stderr, "Full:\n%s\n\nPrefix:\n%s\n\n", full.c_str(), prefix.c_str());
throw std::runtime_error("Full message does not start with prefix");
}
if (full == prefix) {
throw std::runtime_error("Full message is the same as the prefix");
}
auto delta = full.substr(prefix.size());
size_t common_prefix_length = 0;
for (size_t i = 0; i < prefix.size() && i < full.size(); ++i) {
if (prefix[i] != full[i]) {
break;
}
if (prefix[i] == '<') {
// DeepSeek R1's template (as of 20250209) adds a trailing <think> if add_generation_prompt,
// but it removes thinking tags for past messages.
// The prefix and full strings diverge at <think> vs. <｜tool▁calls▁begin｜>, we avoid consuming the leading <.
continue;
}
common_prefix_length = i + 1;
}
auto delta = full.substr(common_prefix_length);
// Strip end tokens
for (const auto & end_token : end_tokens) {
@ -223,7 +236,9 @@ static delta_data init_delta(const common_chat_template & tmpl, const std::vecto
*/
static void test_template(const common_chat_template & tmpl, const std::vector<std::string> & end_tokens,
const json & test_message, const json & tools = {}, const std::string & expected_delta = "",
bool expect_grammar_triggered = true) {
bool expect_grammar_triggered = true,
bool test_grammar_if_triggered = true,
bool think = false) {
common_chat_msg expected_msg = msg_from_json(test_message);
auto user_message = json{
@ -232,7 +247,7 @@ static void test_template(const common_chat_template & tmpl, const std::vector<s
};
for (const auto & tool_choice : json({ "auto", "required" })) {
auto data = init_delta(tmpl, end_tokens, user_message, test_message, tools, tool_choice);
auto data = init_delta(tmpl, end_tokens, user_message, test_message, tools, tool_choice, think);
if (!expected_delta.empty()) {
assert_equals(expected_delta, data.delta);
}
@ -274,7 +289,7 @@ static void test_template(const common_chat_template & tmpl, const std::vector<s
assert_equals(expect_grammar_triggered, grammar_triggered);
}
if (grammar_triggered && !match_string(constrained, grammar.get())) {
if (grammar_triggered && test_grammar_if_triggered && !match_string(constrained, grammar.get())) {
throw std::runtime_error("Failed to match delta against grammar:\n\n" + data.delta +
"\n\nGrammar: " + data.params.grammar);
}
@ -283,16 +298,33 @@ static void test_template(const common_chat_template & tmpl, const std::vector<s
}
static void test_template_output_parsers() {
json text_message {
json message_user {
{ "role", "user" },
{ "content", "Hey there!" },
};
json message_assist {
{ "role", "assistant" },
{ "content", "Hello, world!\nWhat's up?" },
};
json message_assist_thoughts_unparsed_think {
{ "role", "assistant" },
{ "content", "<think>I'm thinking</think>Hello, world!\nWhat's up?" },
};
json message_assist_thoughts_unparsed_r7b {
{ "role", "assistant" },
{ "content", "<|START_THINKING|>I'm thinking<|END_THINKING|>Hello, world!\nWhat's up?" },
};
json message_assist_thoughts {
{ "role", "assistant" },
{ "content", "Hello, world!\nWhat's up?" },
{ "reasoning_content", "I'm thinking" },
};
json tool_calls = json::array({{
{ "type", "function" },
{ "function", { { "name", "special_function" }, { "arguments", "{\"arg1\": 1}" } } },
}});
json tool_call_message {
json message_assist_call {
{ "role", "assistant"},
{ "content", {}},
{ "tool_calls", {
@ -305,7 +337,34 @@ static void test_template_output_parsers() {
},
}},
};
json tool_call_message_with_id {
json message_assist_call_thoughts = {
{ "role", "assistant" },
{ "content", nullptr },
{ "reasoning_content", "I'm\nthinking" },
{ "tool_calls", {
{
{ "type", "function" },
{ "function", {
{ "name", "special_function" },
{ "arguments", "{\"arg1\": 1}" },
}},
},
}},
};
json message_assist_call_thoughts_unparsed = {
{ "role", "assistant" },
{ "content", "<think>I'm\nthinking</think>" },
{ "tool_calls", {
{
{ "type", "function" },
{ "function", {
{ "name", "special_function" },
{ "arguments", "{\"arg1\": 1}" },
}},
},
}},
};
json message_assist_call_id {
{ "role", "assistant"},
{ "content", {}},
{ "tool_calls", {
@ -322,10 +381,9 @@ static void test_template_output_parsers() {
{ "content", {} },
{ "tool_calls", tool_calls }
};
json tool_call_plan_message_with_idx {
json message_assist_call_idx {
{ "role", "assistant"},
{ "content", {}},
{ "tool_plan", "I'm not so sure"},
{ "tool_calls", {
{
{ "type", "function" },
@ -341,8 +399,10 @@ static void test_template_output_parsers() {
{ "content", {} },
{ "tool_calls", tool_calls }
};
json message_assist_call_tool_plan_idx = message_assist_call_idx;
message_assist_call_tool_plan_idx["tool_plan"] = "I'm thinking";
auto python_tool_call_message = json{
auto python_message_assist_call = json{
{ "role", "assistant" },
{ "content", {} },
{ "tool_calls", json{ {
@ -357,7 +417,7 @@ static void test_template_output_parsers() {
} },
} } }
};
auto code_interpreter_tool_call_message = json{
auto code_interpreter_message_assist_call = json{
{ "role", "assistant" },
{ "content", {} },
{ "tool_calls", json{ {
@ -374,17 +434,27 @@ static void test_template_output_parsers() {
};
common_chat_inputs inputs_no_tools;
inputs_no_tools.messages = {
{ { "role", "user" }, { "content", "Hey\nThere" } }
};
inputs_no_tools.messages = json::array({message_user});
inputs_no_tools.extract_reasoning = false;
common_chat_inputs inputs_tools = inputs_no_tools;
inputs_tools.tools = json::array();
inputs_tools.tools.push_back(special_function_tool);
common_chat_inputs inputs_no_tools_think;
inputs_no_tools_think.messages = json::array({message_user});
inputs_no_tools_think.extract_reasoning = true;
common_chat_inputs inputs_tools_builtin = inputs_no_tools;
inputs_tools_builtin.tools = json::array();
inputs_tools_builtin.tools.push_back(python_tool);
common_chat_inputs inputs_tools;
inputs_tools.messages = json::array({message_user});
inputs_tools.tools = json::array({special_function_tool});
inputs_tools.extract_reasoning = false;
common_chat_inputs inputs_tools_think;
inputs_tools_think.messages = json::array({message_user});
inputs_tools_think.tools = json::array({special_function_tool});
inputs_tools_think.extract_reasoning = true;
common_chat_inputs inputs_tools_builtin;
inputs_tools_builtin.messages = json::array({message_user});
inputs_tools_builtin.tools = json::array({python_tool});
inputs_tools_builtin.extract_reasoning = false;
{
// Not supported yet
@ -395,15 +465,53 @@ static void test_template_output_parsers() {
const common_chat_template tmpl(read_file("models/templates/CohereForAI-c4ai-command-r7b-12-2024-tool_use.jinja"), "<s>", "</s>");
std::vector<std::string> end_tokens{ "<|END_OF_TURN_TOKEN|>" };
assert_equals(COMMON_CHAT_FORMAT_CONTENT_ONLY, common_chat_params_init(tmpl, inputs_no_tools).format);
assert_equals(COMMON_CHAT_FORMAT_COMMAND_R7B, common_chat_params_init(tmpl, inputs_tools).format);
assert_equals(COMMON_CHAT_FORMAT_COMMAND_R7B, common_chat_params_init(tmpl, inputs_no_tools).format);
assert_equals(COMMON_CHAT_FORMAT_COMMAND_R7B, common_chat_params_init(tmpl, inputs_tools).format);
assert_equals(COMMON_CHAT_FORMAT_COMMAND_R7B_EXTRACT_REASONING, common_chat_params_init(tmpl, inputs_tools_think).format);
test_template(tmpl, end_tokens, tool_call_plan_message_with_idx, tools,
"<|START_THINKING|>I'm not so sure<|END_THINKING|>"
assert_msg_equals(msg_from_json(message_assist),
common_chat_parse(
"Hello, world!\nWhat's up?",
COMMON_CHAT_FORMAT_COMMAND_R7B));
assert_msg_equals(msg_from_json(message_assist),
common_chat_parse(
"Hello, world!\nWhat's up?<|END_RESPONSE|>",
COMMON_CHAT_FORMAT_COMMAND_R7B));
assert_msg_equals(msg_from_json(message_assist),
common_chat_parse(
"<|START_RESPONSE|>Hello, world!\nWhat's up?<|END_RESPONSE|>",
COMMON_CHAT_FORMAT_COMMAND_R7B));
assert_msg_equals(msg_from_json(message_assist_thoughts_unparsed_r7b),
common_chat_parse(
"<|START_THINKING|>I'm thinking<|END_THINKING|>"
"<|START_RESPONSE|>Hello, world!\nWhat's up?<|END_RESPONSE|>",
COMMON_CHAT_FORMAT_COMMAND_R7B));
assert_msg_equals(msg_from_json(message_assist_thoughts_unparsed_r7b),
common_chat_parse(
"<|START_THINKING|>I'm thinking<|END_THINKING|>"
"Hello, world!\nWhat's up?<|END_RESPONSE|>",
COMMON_CHAT_FORMAT_COMMAND_R7B));
assert_msg_equals(msg_from_json(message_assist_thoughts),
common_chat_parse(
"<|START_THINKING|>I'm thinking<|END_THINKING|>"
"<|START_RESPONSE|>Hello, world!\nWhat's up?<|END_RESPONSE|>",
COMMON_CHAT_FORMAT_COMMAND_R7B_EXTRACT_REASONING));
test_template(tmpl, end_tokens, message_assist_call_idx, tools,
"<|START_THINKING|><|END_THINKING|>"
"<|START_ACTION|>[\n"
" {\"tool_call_id\": \"0\", \"tool_name\": \"special_function\", \"parameters\": {\"arg1\": 1}}\n"
"]<|END_ACTION|>");
test_template(tmpl, end_tokens, text_message, tools,
test_template(tmpl, end_tokens, message_assist_call_tool_plan_idx, tools,
"<|START_THINKING|>I'm thinking<|END_THINKING|>"
"<|START_ACTION|>[\n"
" {\"tool_call_id\": \"0\", \"tool_name\": \"special_function\", \"parameters\": {\"arg1\": 1}}\n"
"]<|END_ACTION|>",
/* expect_grammar_triggered= */ true,
/* test_grammar_if_triggered= */ true,
/* think= */ true);
test_template(tmpl, end_tokens, message_assist, tools,
"<|START_RESPONSE|>Hello, world!\n"
"What's up?<|END_RESPONSE|>",
/* expect_grammar_triggered= */ false);
@ -423,12 +531,12 @@ static void test_template_output_parsers() {
// Generic tool calls doesn't generate / parse content-only messages symmetrically.
assert_msg_equals(msg_from_json(text_message),
assert_msg_equals(msg_from_json(message_assist),
common_chat_parse("{\n"
" \"response\": \"Hello, world!\\nWhat's up?\"\n"
"}",
common_chat_params_init(tmpl, inputs_tools).format));
test_template(tmpl, end_tokens, tool_call_message_with_id, tools,
test_template(tmpl, end_tokens, message_assist_call_id, tools,
"{\n"
" \"tool_calls\": [\n"
" {\n"
@ -448,9 +556,9 @@ static void test_template_output_parsers() {
assert_equals(COMMON_CHAT_FORMAT_MISTRAL_NEMO, common_chat_params_init(tmpl, inputs_tools).format);
test_template(tmpl, end_tokens, text_message, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, message_assist, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(
tmpl, end_tokens, tool_call_message_with_id, tools,
tmpl, end_tokens, message_assist_call_id, tools,
"[TOOL_CALLS][{\"name\": \"special_function\", \"arguments\": {\"arg1\": 1}, \"id\": \"123456789\"}]");
}
{
@ -473,12 +581,12 @@ static void test_template_output_parsers() {
inputs_tools)
.format);
test_template(tmpl, end_tokens, text_message, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, tool_call_message, tools,
test_template(tmpl, end_tokens, message_assist, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, message_assist_call, tools,
"<tool_call>\n"
"{\"name\": \"special_function\", \"arguments\": {\"arg1\": 1}}\n"
"</tool_call>");
test_template(tmpl, end_tokens, python_tool_call_message, tools,
test_template(tmpl, end_tokens, python_message_assist_call, tools,
"<tool_call>\n"
"{\"name\": \"python\", \"arguments\": {\"code\": \"print('hey')\"}}\n"
"</tool_call>");
@ -498,12 +606,12 @@ static void test_template_output_parsers() {
inputs_tools_builtin)
.format);
// test_template(tmpl, end_tokens, text_message, tools, R"(?)", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, code_interpreter_tool_call_message, llama_3_1_tools,
// test_template(tmpl, end_tokens, message_assist, tools, R"(?)", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, code_interpreter_message_assist_call, llama_3_1_tools,
"<|python_tag|>code_interpreter.call(code=\"print('hey')\")");
test_template(tmpl, end_tokens, python_tool_call_message, tools,
test_template(tmpl, end_tokens, python_message_assist_call, tools,
"<|python_tag|>python.call(code=\"print('hey')\")");
test_template(tmpl, end_tokens, tool_call_message, tools,
test_template(tmpl, end_tokens, message_assist_call, tools,
"{\"name\": \"special_function\", \"parameters\": {\"arg1\": 1}}");
}
{
@ -513,8 +621,8 @@ static void test_template_output_parsers() {
assert_equals(COMMON_CHAT_FORMAT_LLAMA_3_X, common_chat_params_init(tmpl, inputs_tools).format);
test_template(tmpl, end_tokens, text_message, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, tool_call_message, tools,
test_template(tmpl, end_tokens, message_assist, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, message_assist_call, tools,
"{\"name\": \"special_function\", \"parameters\": {\"arg1\": 1}}");
}
{
@ -525,8 +633,8 @@ static void test_template_output_parsers() {
assert_equals(COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1,
common_chat_params_init(tmpl, inputs_tools).format);
test_template(tmpl, end_tokens, text_message, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, tool_call_message, tools,
test_template(tmpl, end_tokens, message_assist, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, message_assist_call, tools,
"<function=special_function>{\"arg1\": 1}</function>");
}
{
@ -537,12 +645,12 @@ static void test_template_output_parsers() {
assert_equals(COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2, common_chat_params_init(tmpl, inputs_no_tools).format);
assert_equals(COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2, common_chat_params_init(tmpl, inputs_tools).format);
test_template(tmpl, end_tokens, text_message, {},
test_template(tmpl, end_tokens, message_assist, {},
"all\n"
"Hello, world!\n"
"What's up?",
/* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, tool_call_message, tools,
test_template(tmpl, end_tokens, message_assist_call, tools,
"special_function\n"
"{\"arg1\": 1}");
}
@ -553,23 +661,79 @@ static void test_template_output_parsers() {
assert_equals(COMMON_CHAT_FORMAT_FIREFUNCTION_V2, common_chat_params_init(tmpl, inputs_tools).format);
test_template(tmpl, end_tokens, text_message, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, tool_call_message, tools,
test_template(tmpl, end_tokens, message_assist, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, message_assist_call, tools,
" functools[{\"name\": \"special_function\", \"arguments\": {\"arg1\": 1}}]");
}
{
// Original DeepSeek R1 template. Leaves <｜tool▁calls▁begin｜> and others unclosed. Our logic fixes the prompt.
const common_chat_template tmpl(read_file("models/templates/deepseek-ai-DeepSeek-R1-Distill-Llama-8B.jinja"),
"<s>", "</s>");
std::vector<std::string> end_tokens{ "<｜end▁of▁sentence｜>" };
assert_equals(COMMON_CHAT_FORMAT_DEEPSEEK_R1, common_chat_params_init(tmpl, inputs_tools).format);
assert_equals(COMMON_CHAT_FORMAT_DEEPSEEK_R1, common_chat_params_init(tmpl, inputs_tools).format);
assert_equals(COMMON_CHAT_FORMAT_DEEPSEEK_R1_EXTRACT_REASONING, common_chat_params_init(tmpl, inputs_tools_think).format);
test_template(tmpl, end_tokens, text_message, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, tool_call_message, tools,
"<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>function<｜tool▁sep｜>special_function\n"
"```json\n"
"{\"arg1\": 1}\n"
"```<｜tool▁call▁end｜>");
test_template(tmpl, end_tokens, message_assist, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, message_assist_thoughts, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
assert_msg_equals(msg_from_json(message_assist_thoughts_unparsed_think),
common_chat_parse("<think>I'm thinking</think>Hello, world!\nWhat's up?",
COMMON_CHAT_FORMAT_DEEPSEEK_R1));
assert_msg_equals(msg_from_json(message_assist_thoughts),
common_chat_parse("<think>I'm thinking</think>Hello, world!\nWhat's up?",
COMMON_CHAT_FORMAT_DEEPSEEK_R1_EXTRACT_REASONING));
assert_msg_equals(msg_from_json(message_assist_thoughts),
// Latest template update (ast of 20250209) adds a trailing <think>\n if add_generation_prompt is true.
common_chat_parse("I'm thinking</think>Hello, world!\nWhat's up?",
COMMON_CHAT_FORMAT_DEEPSEEK_R1_EXTRACT_REASONING));
// test_template(tmpl, end_tokens, message_assist_call, tools,
// "<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>function<｜tool▁sep｜>special_function\n"
// "```json\n"
// "{\"arg1\": 1}\n"
// // Look what's not here: <｜tool▁calls▁end｜> (also missing the <｜end▁of▁sentence｜>, but that is removed lazily by the test's delta logic)
// "```<｜tool▁call▁end｜>",
// /* expect_grammar_triggered= */ true,
// /* test_grammar_if_triggered= */ false);
}
{
// Replacement DeepSeek R1 template. Makes the Distill Qwen 7B/32B models happy to call tools and all.
const common_chat_template tmpl(read_file("models/templates/llama-cpp-deepseek-r1.jinja"),
"<s>", "</s>");
std::vector<std::string> end_tokens{ "<｜end▁of▁sentence｜>" };
assert_equals(COMMON_CHAT_FORMAT_DEEPSEEK_R1, common_chat_params_init(tmpl, inputs_tools).format);
assert_equals(COMMON_CHAT_FORMAT_DEEPSEEK_R1_EXTRACT_REASONING, common_chat_params_init(tmpl, inputs_tools_think).format);
test_template(tmpl, end_tokens, message_assist, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
test_template(tmpl, end_tokens, message_assist_thoughts, tools, "Hello, world!\nWhat's up?", /* expect_grammar_triggered= */ false);
assert_msg_equals(msg_from_json(message_assist_thoughts_unparsed_think),
common_chat_parse("<think>I'm thinking</think>Hello, world!\nWhat's up?",
COMMON_CHAT_FORMAT_DEEPSEEK_R1));
assert_msg_equals(msg_from_json(message_assist_thoughts),
common_chat_parse("<think>I'm thinking</think>Hello, world!\nWhat's up?",
COMMON_CHAT_FORMAT_DEEPSEEK_R1_EXTRACT_REASONING));
assert_msg_equals(msg_from_json(message_assist_call_thoughts_unparsed),
common_chat_parse(
"<think>I'm\nthinking</think>\n\n"
"<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>function<｜tool▁sep｜>special_function\n"
"```json\n"
"{\"arg1\": 1}\n"
"```<｜tool▁call▁end｜><｜tool▁calls▁end｜>",
COMMON_CHAT_FORMAT_DEEPSEEK_R1));
assert_msg_equals(msg_from_json(message_assist_call_thoughts),
common_chat_parse(
"<think>I'm\nthinking</think>\n\n"
"<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>function<｜tool▁sep｜>special_function\n"
"```json\n"
"{\"arg1\": 1}\n"
"```<｜tool▁call▁end｜><｜tool▁calls▁end｜>",
COMMON_CHAT_FORMAT_DEEPSEEK_R1_EXTRACT_REASONING));
test_template(tmpl, end_tokens, message_assist_call, tools,
"<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>function<｜tool▁sep｜>special_function\n"
"```json\n"
"{\"arg1\": 1}\n"
"```<｜tool▁call▁end｜><｜tool▁calls▁end｜>");
}
}
@ -586,16 +750,20 @@ int main(int argc, char ** argv) {
std::cout << "|----------|--------|\n";
for (int i = 1; i < argc; i++) {
std::string path = argv[i];
if (path.rfind(".jinja") != path.size() - 6) {
std::cerr << "Skipping non-jinja file: " << path << std::endl;
continue;
try {
std::string path = argv[i];
if (path.rfind(".jinja") != path.size() - 6) {
std::cerr << "Skipping non-jinja file: " << path << std::endl;
continue;
}
common_chat_template tmpl(read_file(path), "", "");
auto parts = string_split(path, "/");
auto name = parts[parts.size() - 1];
auto format = common_chat_format_name(common_chat_params_init(tmpl, inputs).format);
std::cout << "| " << name << " | " << format << " |\n";
} catch (const std::exception & e) {
std::cerr << "Failed to process " << argv[i] << ": " << e.what() << std::endl;
}
common_chat_template tmpl(read_file(path), "", "");
auto parts = string_split(path, "/");
auto name = parts[parts.size() - 1];
std::cout << "| " << name << " | " << common_chat_format_name(common_chat_params_init(tmpl, inputs).format)
<< " |\n";
}
} else
#endif