Add a response template
#1
by qgallouedec HF Staff - opened
This PR adds a response_template to tokenizer_config.json, so tokenizer.parse_response(...) turns a generation back into a message with reasoning_content, content and tool_calls (transformers >= 5.13). It's the Qwen3 one, since the chat template follows the Qwen3 format.
Checked: parsing round-trips plain content, reasoning + content, a single tool call, and parallel tool calls, and doesn't crash on truncated output.
qgallouedec changed pull request status to open
aakanksha19 changed pull request status to merged