feat(turbo): add direct LLM chat mode with runtime toggle
Add 极速 mode that streams via /api/direct-chat without the lead agent stack, plus VITE_TURBO_MODE_VISIBLE in runtime-config.js (default off). Also disable global memory, tighten Qwen/vLLM thinking control for flash mode, and force thinking off on the direct-chat endpoint. Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
1 parent
4f7f07504a
commit
d765afe2ac
23 files changed
+1295
-394
No files matched your search
+27
-17
@@ -29,11 +29,11 @@ from typing import Any
|
||||
|
||||
from langchain_core.language_models import LanguageModelInput
|
||||
from langchain_core.messages import AIMessage, AIMessageChunk
|
||||
from langchain_core.outputs import ChatGeneration, ChatGenerationChunk, ChatResult
|
||||
from langchain_openai import ChatOpenAI
|
||||
|
||||
from deerflow.models.message_ordering import normalize_chat_payload_messages
|
||||
|
||||
from langchain_core.outputs import ChatGeneration, ChatGenerationChunk, ChatResult
|
||||
from langchain_openai import ChatOpenAI
|
||||
|
||||
from deerflow.models.message_ordering import normalize_chat_payload_messages
|
||||
|
||||
|
||||
class PatchedChatOpenAI(ChatOpenAI):
|
||||
"""ChatOpenAI with ``thought_signature`` preservation for Gemini thinking via OpenAI gateway.
|
||||
@@ -89,16 +89,16 @@ class PatchedChatOpenAI(ChatOpenAI):
|
||||
if payload_msg.get("role") == "assistant" and isinstance(orig_msg, AIMessage):
|
||||
_restore_tool_call_signatures(payload_msg, orig_msg)
|
||||
_restore_reasoning_content(payload_msg, orig_msg)
|
||||
else:
|
||||
# Fallback: match assistant-role entries positionally against AIMessages.
|
||||
ai_messages = [m for m in original_messages if isinstance(m, AIMessage)]
|
||||
assistant_payloads = [(i, m) for i, m in enumerate(payload_messages) if m.get("role") == "assistant"]
|
||||
for (_, payload_msg), ai_msg in zip(assistant_payloads, ai_messages):
|
||||
_restore_tool_call_signatures(payload_msg, ai_msg)
|
||||
_restore_reasoning_content(payload_msg, ai_msg)
|
||||
|
||||
normalize_chat_payload_messages(payload)
|
||||
return payload
|
||||
else:
|
||||
# Fallback: match assistant-role entries positionally against AIMessages.
|
||||
ai_messages = [m for m in original_messages if isinstance(m, AIMessage)]
|
||||
assistant_payloads = [(i, m) for i, m in enumerate(payload_messages) if m.get("role") == "assistant"]
|
||||
for (_, payload_msg), ai_msg in zip(assistant_payloads, ai_messages):
|
||||
_restore_tool_call_signatures(payload_msg, ai_msg)
|
||||
_restore_reasoning_content(payload_msg, ai_msg)
|
||||
|
||||
normalize_chat_payload_messages(payload)
|
||||
return payload
|
||||
|
||||
def _convert_chunk_to_generation_chunk(
|
||||
self,
|
||||
@@ -208,11 +208,21 @@ def _restore_reasoning_content(payload_msg: dict, orig_msg: AIMessage) -> None:
|
||||
|
||||
|
||||
def _extract_reasoning_content(payload: dict | None) -> str | None:
|
||||
"""Extract reasoning text from provider payload delta/message objects."""
|
||||
"""Extract reasoning text from provider payload delta/message objects.
|
||||
|
||||
vLLM Qwen reasoning parsers use ``reasoning_content`` on streaming deltas
|
||||
and ``reasoning`` on some non-stream / older builds. Read both so the UI
|
||||
can show the first thinking token instead of waiting for answer content.
|
||||
"""
|
||||
if not isinstance(payload, dict):
|
||||
return None
|
||||
reasoning_content = payload.get("reasoning_content")
|
||||
return reasoning_content if isinstance(reasoning_content, str) and reasoning_content else None
|
||||
if isinstance(reasoning_content, str) and reasoning_content:
|
||||
return reasoning_content
|
||||
reasoning = payload.get("reasoning")
|
||||
if isinstance(reasoning, str) and reasoning:
|
||||
return reasoning
|
||||
return None
|
||||
|
||||
|
||||
def _append_reasoning_content(message: AIMessage | AIMessageChunk, reasoning_content: str) -> AIMessage | AIMessageChunk:
|
||||
|
||||
Reference in new issue
Block a user