fix: accept vLLM thinking in reasoning or reasoning_content

Newer vLLM parsers put CoT in reasoning; older builds and DeepSeek still use reasoning_content. Canonicalize to the UI field and echo only the original wire field so other ChatOpenAI models stay unchanged. Point qwen3.6 at VllmChatModel.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
董宏凯andCursor committed 2026-09-09 22:46:08 +08:00
1 parent d765afe2ac
commit f4c7d41a60
10 files changed
+371 -123

No files matched your search

+5 -3
View File
@@ -1014,9 +1014,11 @@ scheduler runs once per deployment. Per run:
### vLLM Provider (`packages/harness/deerflow/models/vllm_provider.py`)
- `VllmChatModel` subclasses `langchain_openai:ChatOpenAI` for vLLM 0.19.0 OpenAI-compatible endpoints
- Preserves vLLM's non-standard assistant `reasoning` field on full responses, streaming deltas, and follow-up tool-call turns
- Designed for configs that enable thinking through `extra_body.chat_template_kwargs.enable_thinking` on vLLM 0.19.0 Qwen reasoning models, while accepting the older `thinking` alias
- `VllmChatModel` subclasses `langchain_openai:ChatOpenAI` for vLLM OpenAI-compatible endpoints
- Thinking text is accepted from either `reasoning` (newer vLLM parsers) or `reasoning_content` (older vLLM / DeepSeek-style). The UI still reads canonical `additional_kwargs.reasoning_content`. Follow-up turns echo only the wire field that actually arrived, so other ChatOpenAI models are unchanged
- Shared helpers live in `deerflow/models/reasoning_fields.py` and are also used by `PatchedChatOpenAI` (the default `langchain_openai:ChatOpenAI` patch)
- Designed for configs that enable thinking through `extra_body.chat_template_kwargs.enable_thinking` on vLLM Qwen reasoning models, while accepting the older `thinking` alias
- Per-model switch: set only that entry's `use: deerflow.models.vllm_provider:VllmChatModel`. Do not change SiliconFlow / DeepSeek entries that already use `langchain_openai:ChatOpenAI`
### IM Channels System (`app/channels/`)