mirror of
https://git.mirrors.martin98.com/https://github.com/infiniflow/ragflow.git
synced 2025-08-14 16:05:55 +08:00
Fix multiple generate (#1722)
### What problem does this PR solve? #1625 ### Type of change - [x] New Feature (non-breaking change which adds functionality)
This commit is contained in:
H
GitHub
parent
61096596bc
commit
013856b604
@ -59,8 +59,10 @@ class Answer(ComponentBase, ABC):
|
||||
stream = self.get_stream_input()
|
||||
if isinstance(stream, pd.DataFrame):
|
||||
res = stream
|
||||
answer = ""
|
||||
for ii, row in stream.iterrows():
|
||||
yield row.to_dict()
|
||||
answer += row.to_dict()["content"]
|
||||
yield {"content": answer}
|
||||
else:
|
||||
for st in stream():
|
||||
res = st
|
||||
|
||||
@ -67,85 +67,11 @@ class Generate(ComponentBase):
|
||||
cpnts = [para["component_id"] for para in self._param.parameters]
|
||||
return cpnts
|
||||
|
||||
def _run(self, history, **kwargs):
|
||||
chat_mdl = LLMBundle(self._canvas.get_tenant_id(), LLMType.CHAT, self._param.llm_id)
|
||||
prompt = self._param.prompt
|
||||
|
||||
retrieval_res = self.get_input()
|
||||
input = (" - " + "\n - ".join(retrieval_res["content"])) if "content" in retrieval_res else ""
|
||||
for para in self._param.parameters:
|
||||
cpn = self._canvas.get_component(para["component_id"])["obj"]
|
||||
_, out = cpn.output(allow_partial=False)
|
||||
if "content" not in out.columns:
|
||||
kwargs[para["key"]] = "Nothing"
|
||||
else:
|
||||
kwargs[para["key"]] = " - " + "\n - ".join(out["content"])
|
||||
|
||||
kwargs["input"] = input
|
||||
for n, v in kwargs.items():
|
||||
# prompt = re.sub(r"\{%s\}"%n, re.escape(str(v)), prompt)
|
||||
prompt = re.sub(r"\{%s\}" % n, str(v), prompt)
|
||||
|
||||
downstreams = self._canvas.get_component(self._id)["downstream"]
|
||||
if kwargs.get("stream") \
|
||||
and len(downstreams) == 1 \
|
||||
and self._canvas.get_component(downstreams[0])["obj"].component_name.lower() == "answer":
|
||||
return partial(self.stream_output, chat_mdl, prompt, retrieval_res)
|
||||
|
||||
if "empty_response" in retrieval_res.columns:
|
||||
return Generate.be_output(input)
|
||||
|
||||
ans = chat_mdl.chat(prompt, self._canvas.get_history(self._param.message_history_window_size),
|
||||
self._param.gen_conf())
|
||||
|
||||
if self._param.cite and "content_ltks" in retrieval_res.columns and "vector" in retrieval_res.columns:
|
||||
ans, idx = retrievaler.insert_citations(ans,
|
||||
[ck["content_ltks"]
|
||||
for _, ck in retrieval_res.iterrows()],
|
||||
[ck["vector"]
|
||||
for _, ck in retrieval_res.iterrows()],
|
||||
def set_cite(self, retrieval_res, answer):
|
||||
answer, idx = retrievaler.insert_citations(answer, [ck["content_ltks"] for _, ck in retrieval_res.iterrows()],
|
||||
[ck["vector"] for _, ck in retrieval_res.iterrows()],
|
||||
LLMBundle(self._canvas.get_tenant_id(), LLMType.EMBEDDING,
|
||||
self._canvas.get_embedding_model()),
|
||||
tkweight=0.7,
|
||||
vtweight=0.3)
|
||||
del retrieval_res["vector"]
|
||||
retrieval_res = retrieval_res.to_dict("records")
|
||||
df = []
|
||||
for i in idx:
|
||||
df.append(retrieval_res[int(i)])
|
||||
r = re.search(r"^((.|[\r\n])*? ##%s\$\$)" % str(i), ans)
|
||||
assert r, f"{i} => {ans}"
|
||||
df[-1]["content"] = r.group(1)
|
||||
ans = re.sub(r"^((.|[\r\n])*? ##%s\$\$)" % str(i), "", ans)
|
||||
if ans: df.append({"content": ans})
|
||||
return pd.DataFrame(df)
|
||||
|
||||
return Generate.be_output(ans)
|
||||
|
||||
def stream_output(self, chat_mdl, prompt, retrieval_res):
|
||||
res = None
|
||||
if "empty_response" in retrieval_res.columns and "\n- ".join(retrieval_res["content"]):
|
||||
res = {"content": "\n- ".join(retrieval_res["content"]), "reference": []}
|
||||
yield res
|
||||
self.set_output(res)
|
||||
return
|
||||
|
||||
answer = ""
|
||||
for ans in chat_mdl.chat_streamly(prompt, self._canvas.get_history(self._param.message_history_window_size),
|
||||
self._param.gen_conf()):
|
||||
res = {"content": ans, "reference": []}
|
||||
answer = ans
|
||||
yield res
|
||||
|
||||
if self._param.cite and "content_ltks" in retrieval_res.columns and "vector" in retrieval_res.columns:
|
||||
answer, idx = retrievaler.insert_citations(answer,
|
||||
[ck["content_ltks"]
|
||||
for _, ck in retrieval_res.iterrows()],
|
||||
[ck["vector"]
|
||||
for _, ck in retrieval_res.iterrows()],
|
||||
LLMBundle(self._canvas.get_tenant_id(), LLMType.EMBEDDING,
|
||||
self._canvas.get_embedding_model()),
|
||||
tkweight=0.7,
|
||||
self._canvas.get_embedding_model()), tkweight=0.7,
|
||||
vtweight=0.3)
|
||||
doc_ids = set([])
|
||||
recall_docs = []
|
||||
@ -166,6 +92,61 @@ class Generate(ComponentBase):
|
||||
if answer.lower().find("invalid key") >= 0 or answer.lower().find("invalid api") >= 0:
|
||||
answer += " Please set LLM API-Key in 'User Setting -> Model Providers -> API-Key'"
|
||||
res = {"content": answer, "reference": reference}
|
||||
|
||||
return res
|
||||
|
||||
def _run(self, history, **kwargs):
|
||||
chat_mdl = LLMBundle(self._canvas.get_tenant_id(), LLMType.CHAT, self._param.llm_id)
|
||||
prompt = self._param.prompt
|
||||
|
||||
retrieval_res = self.get_input()
|
||||
input = (" - " + "\n - ".join(retrieval_res["content"])) if "content" in retrieval_res else ""
|
||||
for para in self._param.parameters:
|
||||
cpn = self._canvas.get_component(para["component_id"])["obj"]
|
||||
_, out = cpn.output(allow_partial=False)
|
||||
if "content" not in out.columns:
|
||||
kwargs[para["key"]] = "Nothing"
|
||||
else:
|
||||
kwargs[para["key"]] = " - " + "\n - ".join(out["content"])
|
||||
|
||||
kwargs["input"] = input
|
||||
for n, v in kwargs.items():
|
||||
# prompt = re.sub(r"\{%s\}"%n, re.escape(str(v)), prompt)
|
||||
prompt = re.sub(r"\{%s\}" % n, str(v), prompt)
|
||||
|
||||
downstreams = self._canvas.get_component(self._id)["downstream"]
|
||||
if kwargs.get("stream") and len(downstreams) == 1 and self._canvas.get_component(downstreams[0])[
|
||||
"obj"].component_name.lower() == "answer":
|
||||
return partial(self.stream_output, chat_mdl, prompt, retrieval_res)
|
||||
|
||||
if "empty_response" in retrieval_res.columns:
|
||||
return Generate.be_output(input)
|
||||
|
||||
ans = chat_mdl.chat(prompt, self._canvas.get_history(self._param.message_history_window_size),
|
||||
self._param.gen_conf())
|
||||
if self._param.cite and "content_ltks" in retrieval_res.columns and "vector" in retrieval_res.columns:
|
||||
df = self.set_cite(retrieval_res, ans)
|
||||
return pd.DataFrame(df)
|
||||
|
||||
return Generate.be_output(ans)
|
||||
|
||||
def stream_output(self, chat_mdl, prompt, retrieval_res):
|
||||
res = None
|
||||
if "empty_response" in retrieval_res.columns and "\n- ".join(retrieval_res["content"]):
|
||||
res = {"content": "\n- ".join(retrieval_res["content"]), "reference": []}
|
||||
yield res
|
||||
self.set_output(res)
|
||||
return
|
||||
|
||||
answer = ""
|
||||
for ans in chat_mdl.chat_streamly(prompt, self._canvas.get_history(self._param.message_history_window_size),
|
||||
self._param.gen_conf()):
|
||||
res = {"content": ans, "reference": []}
|
||||
answer = ans
|
||||
yield res
|
||||
|
||||
if self._param.cite and "content_ltks" in retrieval_res.columns and "vector" in retrieval_res.columns:
|
||||
res = self.set_cite(retrieval_res, answer)
|
||||
yield res
|
||||
|
||||
self.set_output(res)
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user