Updating chat template of LFM2 to render tools in correct format | enabling minja for that model

[przepeck]

🛠 Summary

CVS-186942

Changing the chat template of LFM2 to render executed tools. This chat templates also works with Minja.

BFCL results:
Multi_turn_base:
Multiturn score with old chat template: 0.266
Multiturn score with new chat template: 0.32

Simple_python:
Llama.cpp: 37.75
OVMS with new chat template: 39.25

🧪 Checklist

• Unit tests added.
• The documentation updated.
• Change follows security best practices.
  ``

[Copilot]

Pull request overview

Updates the LFM2 chat template used by OVMS LLM tests/examples so that prior assistant tool calls are serialized in the LFM2 tool-call token format (and can be processed by Minja), aligning prompt rendering with LFM2 tool-call expectations.

Changes:

• Added Jinja macros to render tool_calls as <|tool_call_start|>[func(arg=...)]<|tool_call_end|> in the LFM2 chat template.
• Updated end-to-end Minja/Jinja tests to expect LFM2 tool-call rendering and tool-call capability detection.
• Added an extras/ example LFM2 chat template mirroring the test template.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
src/test/llm/chat_templates/chat_template_lfm2.jinja	Adds tool-call rendering macros and emits tool calls in LFM2 format during prompt construction.
src/test/llm/chat_template_end_to_end_minja_test.cpp	Updates Minja end-to-end expectations for LFM2 tool-call rendering/capabilities.
src/test/llm/chat_template_end_to_end_jinja_test.cpp	Updates Jinja end-to-end expectations for LFM2 tool-call rendering/capabilities.
extras/chat_template_examples/chat_template_lfm2.jinja	Introduces an example LFM2 template matching the updated tool-call rendering behavior.

Comments suppressed due to low confidence (2)

src/test/llm/chat_templates/chat_template_lfm2.jinja:86

• The template renders tool-call markers whenever tool_calls is defined, even if it is an empty list. That produces <|tool_call_start|>[]<|tool_call_end|> in the prompt, which is likely unintended and can confuse downstream parsing.

    {%- if message["tool_calls"] is defined -%}
        {{- render_tool_calls(message["tool_calls"]) -}}
    {%- endif -%}

extras/chat_template_examples/chat_template_lfm2.jinja:86

• The template renders tool-call markers whenever tool_calls is defined, even if it is an empty list. That produces <|tool_call_start|>[]<|tool_call_end|> in the prompt, which is likely unintended and can confuse downstream parsing.

    {%- if message["tool_calls"] is defined -%}
        {{- render_tool_calls(message["tool_calls"]) -}}
    {%- endif -%}