diff --git a/README.md b/README.md index 86c812fa..8fe306e9 100644 --- a/README.md +++ b/README.md @@ -69,11 +69,12 @@ data. Try our demo at [https://demo.ragflow.io](https://demo.ragflow.io).
- +
## 🔥 Latest Updates +- 2024-11-01 Adds keyword extraction and related question generation to the parsed chunk to improve the accuracy of retrieval. - 2024-09-29 Optimizes multi-round conversations. - 2024-09-13 Adds search mode for knowledge base Q&A. - 2024-09-09 Adds a medical consultant agent template. diff --git a/README_ja.md b/README_ja.md index a5c9e150..bdf887b3 100644 --- a/README_ja.md +++ b/README_ja.md @@ -47,12 +47,13 @@ デモをお試しください：[https://demo.ragflow.io](https://demo.ragflow.io)。
- +
## 🔥 最新情報 +- 2024-11-01 再現の精度を向上させるために、解析されたチャンクにキーワード抽出と関連質問の生成を追加しました。 - 2024-09-29 マルチラウンドダイアログを最適化。 - 2024-09-13 ナレッジベース Q&A の検索モードを追加しました。 - 2024-09-09 エージェントに医療相談テンプレートを追加しました。 diff --git a/README_ko.md b/README_ko.md index 7d526691..5d6a8b67 100644 --- a/README_ko.md +++ b/README_ko.md @@ -49,12 +49,14 @@ 데모를 [https://demo.ragflow.io](https://demo.ragflow.io)에서 실행해 보세요.
- +
## 🔥 업데이트 +- 2024-11-01 파싱된 청크에 키워드 추출 및 관련 질문 생성을 추가하여 재현율을 향상시킵니다. + - 2024-09-29 다단계 대화를 최적화합니다. - 2024-09-13 지식베이스 Q&A 검색 모드를 추가합니다. diff --git a/README_zh.md b/README_zh.md index 2806d0b4..4cdb1164 100644 --- a/README_zh.md +++ b/README_zh.md @@ -47,12 +47,13 @@ 请登录网址 [https://demo.ragflow.io](https://demo.ragflow.io) 试用 demo。
- +
## 🔥 近期更新 +- 2024-11-01 对解析后的chunk加入关键词抽取和相关问题生成以提高召回的准确度。 - 2024-09-29 优化多轮对话. - 2024-09-13 增加知识库问答搜索模式。 - 2024-09-09 在 Agent 中加入医疗问诊模板。 diff --git a/agent/component/base.py b/agent/component/base.py index 0ae965a8..1f1fd33d 100644 --- a/agent/component/base.py +++ b/agent/component/base.py @@ -36,6 +36,7 @@ class ComponentParamBase(ABC): def __init__(self): self.output_var_name = "output" self.message_history_window_size = 22 + self.query = [] def set_name(self, name: str): self._name = name @@ -436,6 +437,16 @@ class ComponentBase(ABC): setattr(self._param, self._param.output_var_name, v) def get_input(self): + if self._param.query: + outs = [] + for q in self._param.query: + if q["value"]: outs.append(pd.DataFrame([{"content": q["value"]}])) + if q["component_id"]: outs.append(self._canvas.get_component(q["component_id"])["obj"].output(allow_partial=False)[1]) + if outs: + df = pd.concat(outs, ignore_index=True) + if "content" in df: df = df.drop_duplicates(subset=['content']).reset_index(drop=True) + return df + upstream_outs = [] reversed_cpnts = [] if len(self._canvas.path) > 1: diff --git a/agent/component/generate.py b/agent/component/generate.py index bdc431ea..19fd9159 100644 --- a/agent/component/generate.py +++ b/agent/component/generate.py @@ -130,6 +130,7 @@ class Generate(ComponentBase): msg = self._canvas.get_history(self._param.message_history_window_size) _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(chat_mdl.max_length * 0.97)) + if len(msg) < 2: msg.append({"role": "user", "content": ""}) ans = chat_mdl.chat(msg[0]["content"], msg[1:], self._param.gen_conf()) if self._param.cite and "content_ltks" in retrieval_res.columns and "vector" in retrieval_res.columns: @@ -149,6 +150,7 @@ class Generate(ComponentBase): msg = self._canvas.get_history(self._param.message_history_window_size) _, msg = message_fit_in([{"role": "system", "content": prompt}, *msg], int(chat_mdl.max_length * 0.97)) + if len(msg) < 2: msg.append({"role": "user", "content": ""}) answer = "" for ans in chat_mdl.chat_streamly(msg[0]["content"], msg[1:], self._param.gen_conf()): res = {"content": ans, "reference": []} diff --git a/agent/component/invoke.py b/agent/component/invoke.py index d497be93..9b9053f5 100644 --- a/agent/component/invoke.py +++ b/agent/component/invoke.py @@ -51,6 +51,9 @@ class Invoke(ComponentBase, ABC): for para in self._param.variables: if para.get("component_id"): cpn = self._canvas.get_component(para["component_id"])["obj"] + if cpn.component_name.lower() == "answer": + args[para["key"]] = self._canvas.get_history(1)[0]["content"] + continue _, out = cpn.output(allow_partial=False) args[para["key"]] = "\n".join(out["content"]) else: diff --git a/agent/templates/HR_callout_zh.json b/agent/templates/HR_callout_zh.json index 44643de4..b8cdbb43 100644 --- a/agent/templates/HR_callout_zh.json +++ b/agent/templates/HR_callout_zh.json @@ -152,7 +152,8 @@ "Generate:ToughLawsCheat", "Generate:KindCarrotsSit", "Generate:DirtyToolsTrain", - "Generate:FluffyPillowsGrow" + "Generate:FluffyPillowsGrow", + "Generate:ProudEarsWorry" ] }, "Retrieval:ShaggyRadiosRetire": { @@ -212,7 +213,9 @@ "top_p": 0.3 } }, - "downstream": [], + "downstream": [ + "Answer:TwentyMugsDeny" + ], "upstream": [ "categorize:0" ] @@ -331,9 +334,9 @@ "message_history_window_size": 12, "parameters": [ { + "component_id": "Retrieval:ColdEelsArrive", "id": "5166a107-e859-4c71-99a2-3a216c775347", - "key": "jd", - "component_id": "Retrieval:ColdEelsArrive" + "key": "jd" } ], "presence_penalty": 0.4, @@ -1266,9 +1269,9 @@ "parameter": "Precise", "parameters": [ { + "component_id": "Retrieval:ColdEelsArrive", "id": "5166a107-e859-4c71-99a2-3a216c775347", - "key": "jd", - "component_id": "Retrieval:ColdEelsArrive" + "key": "jd" } ], "presencePenaltyEnabled": true, @@ -1541,6 +1544,19 @@ "target": "Answer:TwentyMugsDeny", "targetHandle": "c", "type": "buttonEdge" + }, + { + "type": "buttonEdge", + "markerEnd": "logo", + "style": { + "strokeWidth": 2, + "stroke": "rgb(202 197 245)" + }, + "source": "Generate:ProudEarsWorry", + "sourceHandle": "b", + "target": "Answer:TwentyMugsDeny", + "targetHandle": "c", + "id": "reactflow__edge-Generate:ProudEarsWorryb-Answer:TwentyMugsDenyc" } ] },