pengjunjie/ragflow
1
0
Fork 1
mirror of https://github.com/infiniflow/ragflow.git synced 2026-01-29 22:56:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

Refactor: Enhance delta streaming in chat functions for improved reasoning and content handling (#12453)

Browse Source 
### What problem does this PR solve?

change:
Enhance delta streaming in chat functions for improved reasoning and
content handling

### Type of change


- [x] Refactoring
This commit is contained in:
buua436
2026-01-08 13:34:16 +08:00
committed by GitHub
parent f4e2783eb4
commit 1996aa0dac
5 changed files with 325 additions and 123 deletions
Download Patch File Download Diff File Expand all files Collapse all files

73
api/apps/sdk/session.py
View File

@ -304,9 +304,12 @@ async def chat_completion_openai_like(tenant_id, chat_id):
# The choices field on the last chunk will always be an empty array [].
async def streamed_response_generator(chat_id, dia, msg):
token_used = 0
answer_cache = ""
reasoning_cache = ""
last_ans = {}
full_content = ""
full_reasoning = ""
final_answer = None
final_reference = None
in_think = False
response = {
"id": f"chatcmpl-{chat_id}",
"choices": [
@ -336,47 +339,30 @@ async def chat_completion_openai_like(tenant_id, chat_id):
chat_kwargs["doc_ids"] = doc_ids_str
async for ans in async_chat(dia, msg, True, **chat_kwargs):
last_ans = ans
answer = ans["answer"]
reasoning_match = re.search(r"<think>(.*?)</think>", answer, flags=re.DOTALL)
if reasoning_match:
reasoning_part = reasoning_match.group(1)
content_part = answer[reasoning_match.end() :]
else:
reasoning_part = ""
content_part = answer
reasoning_incremental = ""
if reasoning_part:
if reasoning_part.startswith(reasoning_cache):
reasoning_incremental = reasoning_part.replace(reasoning_cache, "", 1)
else:
reasoning_incremental = reasoning_part
reasoning_cache = reasoning_part
content_incremental = ""
if content_part:
if content_part.startswith(answer_cache):
content_incremental = content_part.replace(answer_cache, "", 1)
else:
content_incremental = content_part
answer_cache = content_part
token_used += len(reasoning_incremental) + len(content_incremental)
if not any([reasoning_incremental, content_incremental]):
if ans.get("final"):
if ans.get("answer"):
full_content = ans["answer"]
final_answer = ans.get("answer") or full_content
final_reference = ans.get("reference", {})
continue
if reasoning_incremental:
response["choices"][0]["delta"]["reasoning_content"] = reasoning_incremental
else:
response["choices"][0]["delta"]["reasoning_content"] = None
if content_incremental:
response["choices"][0]["delta"]["content"] = content_incremental
else:
if ans.get("start_to_think"):
in_think = True
continue
if ans.get("end_to_think"):
in_think = False
continue
delta = ans.get("answer") or ""
if not delta:
continue
token_used += len(delta)
if in_think:
full_reasoning += delta
response["choices"][0]["delta"]["reasoning_content"] = delta
response["choices"][0]["delta"]["content"] = None
else:
full_content += delta
response["choices"][0]["delta"]["content"] = delta
response["choices"][0]["delta"]["reasoning_content"] = None
yield f"data:{json.dumps(response, ensure_ascii=False)}\n\n"
except Exception as e:
response["choices"][0]["delta"]["content"] = "**ERROR**: " + str(e)
@ -388,8 +374,9 @@ async def chat_completion_openai_like(tenant_id, chat_id):
response["choices"][0]["finish_reason"] = "stop"
response["usage"] = {"prompt_tokens": len(prompt), "completion_tokens": token_used, "total_tokens": len(prompt) + token_used}
if need_reference:
response["choices"][0]["delta"]["reference"] = chunks_format(last_ans.get("reference", []))
response["choices"][0]["delta"]["final_content"] = last_ans.get("answer", "")
reference_payload = final_reference if final_reference is not None else last_ans.get("reference", [])
response["choices"][0]["delta"]["reference"] = chunks_format(reference_payload)
response["choices"][0]["delta"]["final_content"] = final_answer if final_answer is not None else full_content
yield f"data:{json.dumps(response, ensure_ascii=False)}\n\n"
yield "data:[DONE]\n\n"