From c818263f2a5ddab028dea5f169ea2b2266421125 Mon Sep 17 00:00:00 2001 From: Aldehir Rojas Date: Sun, 28 Jun 2026 09:53:32 -0500 Subject: [PATCH] chat : implement minicpm5 parser (#24889) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit * Add minicpm5 tool call parser * Refactor MiniCPM5 PEG parser per review feedback * Fix jinja min/max API to match Jinja2 * modify by review * MiniCPM5: use autoparser for XML tool calls and fix grammar preserved-token triggers * MiniCPM5: fix streaming tool-arg placeholder and remove alt XML markers * skip min/max attribute tests in -py mode * test-jinja: use real expected output for min/max attribute tests * MiniCPM5: revert shared mapper and history fallbacks per review Drop streaming tool-arg placeholder workarounds from the generic PEG mapper and restore strict tool-call argument JSON parsing so MiniCPM5 support stays limited to autoparser/diff-analyzer changes. * chat : refactor minicpm5 back to dedicated parser * cont : simplify grammar * cont : refactor * cont : fixes * cont : rename template to openbmb-MiniCPM5-1B.jinja * cont : add message delimiters * cont : fix tests --------- Co-authored-by: zhangtao Co-authored-by: 张涛 <> --- common/chat.cpp | 151 +++++++++++++++++ common/jinja/value.cpp | 44 +++++ models/templates/openbmb-MiniCPM5-1B.jinja | 179 +++++++++++++++++++++ tests/test-chat.cpp | 78 +++++++++ tests/test-jinja.cpp | 30 ++++ 5 files changed, 482 insertions(+) create mode 100644 models/templates/openbmb-MiniCPM5-1B.jinja diff --git a/common/chat.cpp b/common/chat.cpp index 0cee80434e..a38c62f5fe 100644 --- a/common/chat.cpp +++ b/common/chat.cpp @@ -2376,6 +2376,149 @@ static void func_args_not_string(json & messages) { } +// MiniCPM5 format: +// - Reasoning: {reasoning} (optional) +// - Tool calls: value +static common_chat_params common_chat_params_init_minicpm5(const common_chat_template & tmpl, + const autoparser::generation_params & inputs) { + common_chat_params data; + + data.prompt = common_chat_template_direct_apply_impl(tmpl, inputs); + data.generation_prompt = common_chat_template_generation_prompt_impl(tmpl, inputs); + data.format = COMMON_CHAT_FORMAT_PEG_NATIVE; + data.supports_thinking = true; + data.preserved_tokens = { + "", + "", + "", + "", + }; + + data.thinking_start_tag = ""; + data.thinking_end_tag = ""; + + data.message_delimiters = { + { COMMON_CHAT_ROLE_ASSISTANT, "<|im_start|>assistant" }, + { COMMON_CHAT_ROLE_TOOL, "<|im_start|>user\n" }, + { COMMON_CHAT_ROLE_USER, "<|im_start|>user" }, + { COMMON_CHAT_ROLE_SYSTEM, "<|im_start|>system" }, + }; + + auto has_tools = inputs.tools.is_array() && !inputs.tools.empty(); + auto has_response_format = inputs.json_schema.is_object() && !inputs.json_schema.empty(); + auto extract_reasoning = inputs.reasoning_format != COMMON_REASONING_FORMAT_NONE; + auto include_grammar = has_response_format || (has_tools && inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_NONE); + + if (inputs.has_continuation()) { + const auto & msg = inputs.continue_msg; + + data.generation_prompt = "<|im_start|>assistant\n\n" + msg.reasoning_content; + if (inputs.continue_final_message == COMMON_CHAT_CONTINUATION_CONTENT) { + data.generation_prompt += "\n\n\n" + msg.render_content(); + } + + data.prompt += data.generation_prompt; + } + + auto parser = build_chat_peg_parser([&](common_chat_peg_builder & p) { + auto generation_prompt = p.literal("<|im_start|>assistant\n"); + + auto reasoning = p.eps(); + if (extract_reasoning) { + reasoning = ("" << p.reasoning(p.until("")) << "") + p.space(); + } + + // Response format parser + if (has_response_format) { + return generation_prompt + reasoning + p.content(p.schema(p.json(), "response-format", inputs.json_schema)); + } + + if (has_tools && inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_NONE) { + // CDATA lets a value carry characters that would otherwise close the tag (e.g. + // ); capture the inner text only, excluding the CDATA markers. + auto string_value = p.choice({ + p.literal("")) + p.literal("]]>"), "]]>") + p.tool_arg_close(p.literal("")), + p.negate(p.literal("")) + p.tool_arg_close(p.literal("")), "") + }); + + auto tool_choice = p.choice(); + foreach_function(inputs.tools, [&](const json & tool) { + const auto & function = tool.at("function"); + const std::string name = function.at("name"); + auto params = function.contains("parameters") ? function.at("parameters") : json::object(); + + auto args = p.eps(); + if (params.contains("properties") && params.at("properties").is_object() && !params.at("properties").empty()) { + auto schema_info = common_schema_info(); + schema_info.resolve_refs(params); + + auto arg_choice = p.choice(); + for (const auto & [prop_name, prop_schema] : params.at("properties").items()) { + auto value_parser = p.eps(); + if (schema_info.resolves_to_string(prop_schema)) { + value_parser = string_value; + } else { + value_parser = p.tool_arg_json_value( + p.schema(p.json(), "tool-" + name + "-arg-" + prop_name + "-schema", prop_schema, false) + ) + p.tool_arg_close(p.literal("")); + } + + auto arg_rule = p.tool_arg( + p.tool_arg_open(p.literal("")) + + value_parser + ); + + arg_choice |= arg_rule; + } + args = p.zero_or_more(arg_choice + p.space()); + } + + auto tool_parser = p.tool( + p.tool_open(p.literal("")) + << p.tool_args(args) + << p.tool_close(p.literal(""))); + + tool_choice |= p.rule("tool-" + name, tool_parser); + }); + + auto max_calls = inputs.parallel_tool_calls ? -1 : 1; + auto tool_calls = p.trigger_rule("tool-call", p.repeat(tool_choice + p.space(), 1, max_calls)); + + auto content = p.content(p.until(" common_chat_try_specialized_template( return common_chat_params_init_gemma4(tmpl, params); } + // MiniCPM5 - XML tool calls with ... + if (src.find("Tool usage guidelines:") != std::string::npos && + src.find("(val) ? mk_val(std::move(arr)) : mk_val(std::move(arr)); }}, + {"min", [](const func_args & args) -> value { + args.ensure_count(1, 4); + args.ensure_vals(); + value val_case = args.get_kwarg_or_pos("case_sensitive", 1); + value attribute = args.get_kwarg_or_pos("attribute", 2); + if (!attribute->is_undefined()) { + throw not_implemented_exception("min: attribute not implemented"); + } + // FIXME: min is currently always case sensitive + (void) val_case; + const auto & arr = args.get_pos(0)->as_array(); + if (arr.empty()) { + return mk_val(); + } + value result = arr[0]; + for (size_t i = 1; i < arr.size(); ++i) { + if (value_compare(arr[i], result, value_compare_op::lt)) { + result = arr[i]; + } + } + return result; + }}, + {"max", [](const func_args & args) -> value { + args.ensure_count(1, 4); + args.ensure_vals(); + value val_case = args.get_kwarg_or_pos("case_sensitive", 1); + value attribute = args.get_kwarg_or_pos("attribute", 2); + if (!attribute->is_undefined()) { + throw not_implemented_exception("max: attribute not implemented"); + } + // FIXME: max is currently always case sensitive + (void) val_case; + const auto & arr = args.get_pos(0)->as_array(); + if (arr.empty()) { + return mk_val(); + } + value result = arr[0]; + for (size_t i = 1; i < arr.size(); ++i) { + if (value_compare(arr[i], result, value_compare_op::gt)) { + result = arr[i]; + } + } + return result; + }}, {"unique", array_unique_not_implemented}, }; return builtins; diff --git a/models/templates/openbmb-MiniCPM5-1B.jinja b/models/templates/openbmb-MiniCPM5-1B.jinja new file mode 100644 index 0000000000..cb2934c459 --- /dev/null +++ b/models/templates/openbmb-MiniCPM5-1B.jinja @@ -0,0 +1,179 @@ +{{- bos_token }}{%- if tools %} + {%- set tool_definitions %} + {{- "# Tools\n\nYou are provided with function signatures within XML tags:\n" }} + {%- for tool in tools %} + {{- "\n" }} + {{- tool | tojson(ensure_ascii=False) }} + {%- endfor %} + {{- '\n\n\nTool usage guidelines:\n- You may call zero or more functions. If no function calls are needed, just answer normally and do not include any .\n- When calling a function, return an XML object within using:\nparam-value\n- param-value may be multi-line. If it contains <, & or newline characters, wrap it in a CDATA block: ' }} + {%- endset %} + + {{- '<|im_start|>system\n' }} + {%- if messages[0].role == 'system' %} + {%- if '' in messages[0].content %} + {{- messages[0].content.replace('', tool_definitions) }} + {%- else %} + {{- messages[0].content + '\n\n' + tool_definitions }} + {%- endif %} + {%- else %} + {{- tool_definitions.lstrip() }} + {%- endif %} + {{- '<|im_end|>\n' }} +{%- else %} + {%- if messages[0].role == 'system' %} + {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }} + {%- endif %} +{%- endif %} +{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %} +{%- for message in messages[::-1] %} + {%- set index = (messages|length - 1) - loop.index0 %} + {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('') and message.content.endswith('')) %} + {%- set ns.multi_step_tool = false %} + {%- set ns.last_query_index = index %} + {%- endif %} +{%- endfor %} +{%- for message in messages %} + {%- if message.content is string %} + {%- set content = message.content %} + {%- else %} + {%- set content = '' %} + {%- endif %} + {%- if (message.role == "user") or (message.role == "system" and not loop.first) %} + {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }} + {%- elif message.role == "assistant" %} + {%- set reasoning_content = '' %} + {%- if message.reasoning_content is string %} + {%- set reasoning_content = message.reasoning_content %} + {%- else %} + {%- if '' in content %} + {%- set reasoning_content = content.split('')[0].rstrip('\n').split('')[-1].lstrip('\n') %} + {%- set content = content.split('')[-1].lstrip('\n') %} + {%- endif %} + {%- endif %} + + {%- if message.tool_calls %} + {%- set content_parts = content.split('') %} + {%- set processed_content = content_parts[0] %} + {%- set tool_calls_count = message.tool_calls|length %} + {%- set tool_sep_count = content_parts|length - 1 %} + {%- set min_count = [tool_calls_count, tool_sep_count]|min %} + + {%- for i in range(1, content_parts|length) %} + {%- set tool_index = i - 1 %} + {%- if tool_index < tool_calls_count %} + {%- set tool_call = message.tool_calls[tool_index] %} + {%- if tool_call.function %} + {%- set tool_call = tool_call.function %} + {%- endif %} + {%- set single_tool_xml %} + {{- '' }} + {%- if tool_call.arguments %} + {%- set args_dict = tool_call.arguments %} + {%- for param_name, param_value in args_dict.items() %} + {{- '' }} + {%- if param_value is string and ('<' in param_value or '&' in param_value or '\n' in param_value) %} + {{- '' }} + {%- else %} + {{- param_value }} + {%- endif %} + {{- '' }} + {%- endfor %} + {%- endif %} + {{- '' }} + {%- endset %} + {%- set processed_content = processed_content + single_tool_xml + content_parts[i] %} + {%- else %} + {%- set processed_content = processed_content + content_parts[i] %} + {%- endif %} + {%- endfor %} + + {%- if tool_calls_count > tool_sep_count %} + {%- for remaining_index in range(tool_sep_count, tool_calls_count) %} + {%- set tool_call = message.tool_calls[remaining_index] %} + {%- if tool_call.function %} + {%- set tool_call = tool_call.function %} + {%- endif %} + {%- set remaining_tool_xml %} + {{- '' }} + {%- if tool_call.arguments %} + {%- set args_dict = tool_call.arguments %} + {%- for param_name, param_value in args_dict.items() %} + {{- '' }} + {%- if param_value is string and ('<' in param_value or '&' in param_value or '\n' in param_value) %} + {{- '' }} + {%- else %} + {{- param_value }} + {%- endif %} + {{- '' }} + {%- endfor %} + {%- endif %} + {{- '' }} + {%- endset %} + {%- set processed_content = processed_content + remaining_tool_xml %} + {%- endfor %} + {%- endif %} + + {%- set content = processed_content %} + {%- endif %} + + {%- if loop.index0 > ns.last_query_index %} + {%- if reasoning_content %} + {{- '<|im_start|>' + message.role + '\n\n' + reasoning_content.strip('\n') + '\n\n\n' + content.lstrip('\n') }} + {%- else %} + {{- '<|im_start|>' + message.role + '\n' + content }} + {%- endif %} + {%- else %} + {{- '<|im_start|>' + message.role + '\n' + content }} + {%- endif %} + + {%- if message.tool_calls and not has_tool_sep %} + {%- for tool_call in message.tool_calls %} + {%- if (loop.first and content) or (not loop.first) %} + {{- '\n' }} + {%- endif %} + {%- if tool_call.function %} + {%- set tool_call = tool_call.function %} + {%- endif %} + {{- '' }} + {%- if tool_call.arguments %} + {%- set args_dict = tool_call.arguments %} + {%- for param_name, param_value in args_dict.items() %} + {{- '' }} + {%- if param_value is string and ('<' in param_value or '&' in param_value or '\n' in param_value) %} + {{- '' }} + {%- else %} + {{- param_value }} + {%- endif %} + {{- '' }} + {%- endfor %} + {%- endif %} + {{- '' }} + {%- endfor %} + {%- endif %} + {{- '<|im_end|>\n' }} + {%- elif message.role == "tool" %} + {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %} + {{- '<|im_start|>user' }} + {%- endif %} + {{- '\n\n' }} + {%- if message.content is string %} + {{- content }} + {%- else %} + {{- message.content | tojson(ensure_ascii=False) }} + {%- endif %} + {{- '\n' }} + {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %} + {{- '<|im_end|>\n' }} + {%- endif %} + {%- endif %} +{%- endfor %} +{%- if add_generation_prompt %} + {{- '<|im_start|>assistant\n' }} + {%- if enable_thinking is defined %} + {%- if enable_thinking is false %} + {{- '\n\n\n\n' }} + {%- elif enable_thinking is true %} + {{- '\n' }} + {%- endif %} + {%- endif %} +{%- endif %} diff --git a/tests/test-chat.cpp b/tests/test-chat.cpp index c38aed8cfe..5f71e5da6e 100644 --- a/tests/test-chat.cpp +++ b/tests/test-chat.cpp @@ -5593,6 +5593,77 @@ static void test_template_output_peg_parsers(bool detailed_debug) { .expect_content("Hello, world!\nWhat's up?") .run(); } + + // MiniCPM5 - XML tool calls with ... + { + auto tst = peg_tester("models/templates/openbmb-MiniCPM5-1B.jinja", detailed_debug); + + tst.test("Hello, world!\nWhat's up?") + .enable_thinking(false) + .reasoning_format(COMMON_REASONING_FORMAT_AUTO) + .expect(message_assist) + .run(); + + tst.test(R"(print('Hello, World!'))") + .enable_thinking(false) + .reasoning_format(COMMON_REASONING_FORMAT_AUTO) + .tools({ python_tool }) + .expect_tool_calls({ { "python", R"#({"code": "print('Hello, World!')"})#", {} } }) + .run(); + + tst.test(R"()") + .enable_thinking(false) + .reasoning_format(COMMON_REASONING_FORMAT_AUTO) + .tools({ empty_args_tool }) + .expect(simple_assist_msg("", "", "empty_args", "{}")) + .run(); + + tst.test(R"(print('x'))") + .enable_thinking(false) + .reasoning_format(COMMON_REASONING_FORMAT_AUTO) + .parallel_tool_calls(true) + .tools({ python_tool }) + .expect_tool_calls({ { "python", R"#({"code": "print('x')"})#", {} } }) + .run(); + + // CDATA lets a string value carry characters that would otherwise close the tag. + tst.test(R"(hi ]]>)") + .enable_thinking(false) + .reasoning_format(COMMON_REASONING_FORMAT_AUTO) + .tools({ html_tool }) + .expect_tool_calls({ { "html", R"#({"markup": "hi "})#", {} } }) + .run(); + + tst.test(R"(I'm thinkingprint('hey'))") + .enable_thinking(true) + .reasoning_format(COMMON_REASONING_FORMAT_AUTO) + .tools({ python_tool }) + .expect_reasoning("I'm thinking") + .expect_tool_calls({ { "python", R"#({"code": "print('hey')"})#", {} } }) + .run(); + + tst.test(R"(print('x') +print('y'))") + .enable_thinking(false) + .reasoning_format(COMMON_REASONING_FORMAT_AUTO) + .parallel_tool_calls(true) + .tools({ python_tool }) + .expect_tool_calls({ + { "python", R"#({"code": "print('x')"})#", {} }, + { "python", R"#({"code": "print('y')"})#", {} }, + }) + .run(); + + tst.test(" thinkingHello, world!\nWhat's up?") + .enable_thinking(true) + .reasoning_format(COMMON_REASONING_FORMAT_AUTO) + .messages({ message_user, message_assist_prefill_reasoning }) + .add_generation_prompt(false) + .continue_final_message(COMMON_CHAT_CONTINUATION_REASONING) + .expect_reasoning("I'm thinking") + .expect_content("Hello, world!\nWhat's up?") + .run(); + } } static void test_template_generation_prompt() { @@ -5740,6 +5811,13 @@ static void test_template_generation_prompt() { check(tmpls, continuation_content(), "<｜Assistant｜>I'm thinkingHello, "); check(tmpls, continuation_reasoning(), "<｜Assistant｜>I'm"); } + + { + auto tmpls = read_templates("models/templates/openbmb-MiniCPM5-1B.jinja"); + check(tmpls, basic(), "<|im_start|>assistant\n\n"); + check(tmpls, continuation_content(), "<|im_start|>assistant\n\nI'm thinking\n\n\nHello, "); + check(tmpls, continuation_reasoning(), "<|im_start|>assistant\n\nI'm"); + } } // Test the developer role to system workaround with a simple mock template diff --git a/tests/test-jinja.cpp b/tests/test-jinja.cpp index 81bbcd55a4..d8d1892a91 100644 --- a/tests/test-jinja.cpp +++ b/tests/test-jinja.cpp @@ -1584,6 +1584,36 @@ static void test_array_methods(testing & t) { "6" ); + test_template(t, "array|min", + "{{ [tool_calls_count, tool_sep_count]|min }}", + {{"tool_calls_count", 2}, {"tool_sep_count", 1}}, + "1" + ); + + test_template(t, "array|max", + "{{ [tool_calls_count, tool_sep_count]|max }}", + {{"tool_calls_count", 2}, {"tool_sep_count", 1}}, + "2" + ); + + test_template(t, "array|min attribute", + "{{ items|min(attribute='x') }}", + {{"items", json::array({ + json({{"x", 2}}), + json({{"x", 1}}), + })}}, + "{'x': 1}" + ); + + test_template(t, "array|max attribute", + "{{ items|max(attribute='x') }}", + {{"items", json::array({ + json({{"x", 2}}), + json({{"x", 1}}), + })}}, + "{'x': 2}" + ); + // not used by any chat templates // test_template(t, "array.insert()", // "{% set _ = arr.insert(1, 'x') %}{{ arr|join(',') }}",