- Prime the opening <think> in the generation prompt and remove enable_thinking (978e150108737c297b1b3f6adf47f7dfd23e24f9)
- Raise on enable_thinking=False and document the vLLM reasoning parser (e347c8246bb4a2a2053dad165c918d0caa69f7d7)
- Add --reasoning-parser qwen3 to the vLLM serve example (2219e8f6684a7c2283e03458a2c16aa1cab7e510)
4 days ago
Add preserve_thinking option to the chat template (#1)
- Add preserve_thinking option to the chat template (e99d8525a551364c22929d4a40b066ffc62299f7)
- Strip leading newline in the no-reasoning render path (6ea700af22f100101e2f5409e549f7e1026c334a)
