feat: add RAG setting panel

libs/ktem/ktem/assets/css/main.css

@@ -213,3 +213,8 @@ pdfjs-viewer-element {
   flex: 1;
   overflow: auto;
 }
+
+#chat-input {
+  resize: vertical;
+  max-height: 200px;
+}

libs/ktem/ktem/index/file/knet/knet_index.py

@@ -1,3 +1,4 @@
+import os
 from typing import Any
 
 from ktem.index.file import FileIndex
@@ -5,6 +6,8 @@ from ktem.index.file import FileIndex
 from ..base import BaseFileIndexIndexing, BaseFileIndexRetriever
 from .pipelines import KnetIndexingPipeline, KnetRetrievalPipeline
 
+DEFAULT_RAG_PIPELINE = os.environ.get("KNET_RAG_PIPELINE", "Direct_Query")
+
 
 class KnowledgeNetworkFileIndex(FileIndex):
     @classmethod
@@ -43,5 +46,11 @@ class KnowledgeNetworkFileIndex(FileIndex):
             # also set the collection_name for API call
             obj.VS = None
             obj.collection_name = f"kh_index_{self.id}"
+            obj.retrieval_expansion = settings.get("rag_settings", {}).get(
+                "retrieval_expansion", False
+            )
+            obj.pipeline_name = settings.get("rag_settings", {}).get(
+                "pipeline", DEFAULT_RAG_PIPELINE
+            )
 
         return retrievers

libs/ktem/ktem/index/file/knet/pipelines.py

@@ -47,6 +47,8 @@ class KnetRetrievalPipeline(BaseFileIndexRetriever):
 
     collection_name: str = "default"
     rerankers: Sequence[BaseReranking] = [LLMReranking.withx()]
+    retrieval_expansion: bool = False
+    pipeline_name: str | None = None
 
     def encode_image_base64(self, image_path: str | Path) -> bytes | str:
         """Convert image to base64"""
@@ -81,7 +83,10 @@ class KnetRetrievalPipeline(BaseFileIndexRetriever):
             "collection": self.collection_name,
             "meta_filters": {"doc_name": doc_ids},
             "conv_id": conv_id,
+            "assert_query_type": self.pipeline_name,
+            "retrieval_expansion": self.retrieval_expansion,
         }
+        print(params)
         params["meta_filters"] = json.dumps(params["meta_filters"])
         response = requests.get(self.DEFAULT_KNET_ENDPOINT, params=params)
         metadata_translation = {

libs/ktem/ktem/pages/chat/__init__.py

@@ -24,6 +24,7 @@ from .chat_panel import ChatPanel
 from .chat_suggestion import ChatSuggestion
 from .common import STATE
 from .control import ConversationControl
+from .rag_setting import RAGSetting
 from .report import ReportIssue
 
 DEFAULT_SETTING = "(default)"
@@ -95,8 +96,11 @@ class ChatPage(BasePage):
                                 self._indices_input.append(gr_index)
                         setattr(self, f"_index_{index.id}", index_ui)
 
+                with gr.Accordion(label="Retrieval Settings") as _:
+                    self.retrieval_settings = RAGSetting(self._app)
+
                 if len(self._app.index_manager.indices) > 0:
-                    with gr.Accordion(label="Quick Upload") as _:
+                    with gr.Accordion(label="Quick Upload", open=False) as _:
                         self.quick_file_upload = File(
                             file_types=list(KH_DEFAULT_FILE_EXTRACTORS.keys()),
                             file_count="multiple",
@@ -180,6 +184,7 @@ class ChatPage(BasePage):
                 self._reasoning_type,
                 self._llm_type,
                 self.chat_state,
+                self.retrieval_settings.setting_state,
                 self._app.user_id,
             ]
             + self._indices_input,
@@ -251,6 +256,7 @@ class ChatPage(BasePage):
                 self._reasoning_type,
                 self._llm_type,
                 self.chat_state,
+                self.retrieval_settings.setting_state,
                 self._app.user_id,
             ]
             + self._indices_input,
@@ -648,6 +654,7 @@ class ChatPage(BasePage):
         session_reasoning_type: str,
         session_llm: str,
         state: dict,
+        rag_state: dict,
         user_id: int,
         *selecteds,
     ):
@@ -665,6 +672,8 @@ class ChatPage(BasePage):
         # override reasoning_mode by temporary chat page state
         print("Session reasoning type", session_reasoning_type)
         print("Session LLM", session_llm)
+        print("RAG settings", rag_state)
+
         reasoning_mode = (
             settings["reasoning.use"]
             if session_reasoning_type in (DEFAULT_SETTING, None)
@@ -675,6 +684,8 @@ class ChatPage(BasePage):
         reasoning_id = reasoning_cls.get_info()["id"]
 
         settings = deepcopy(settings)
+        settings["rag_settings"] = rag_state
+
         llm_setting_key = f"reasoning.options.{reasoning_id}.llm"
         if llm_setting_key in settings and session_llm not in (DEFAULT_SETTING, None):
             settings[llm_setting_key] = session_llm
@@ -711,6 +722,7 @@ class ChatPage(BasePage):
         reasoning_type,
         llm_type,
         state,
+        rag_state,
         user_id,
         *selecteds,
     ):
@@ -722,7 +734,7 @@ class ChatPage(BasePage):
 
         # construct the pipeline
         pipeline, reasoning_state = self.create_pipeline(
-            settings, reasoning_type, llm_type, state, user_id, *selecteds
+            settings, reasoning_type, llm_type, state, rag_state, user_id, *selecteds
         )
         print("Reasoning state", reasoning_state)
         pipeline.set_output_queue(queue)
@@ -785,6 +797,7 @@ class ChatPage(BasePage):
         reasoning_type,
         llm_type,
         state,
+        rag_state,
         user_id,
         *selecteds,
     ):
@@ -802,6 +815,7 @@ class ChatPage(BasePage):
             reasoning_type,
             llm_type,
             state,
+            rag_state,
             user_id,
             *selecteds,
         ):

libs/ktem/ktem/pages/chat/chat_panel.py

@@ -26,6 +26,7 @@ class ChatPanel(BasePage):
                 scale=15,
                 container=False,
                 max_lines=10,
+                elem_id="chat-input",
             )
             self.submit_btn = gr.Button(
                 value="Send",

libs/ktem/ktem/pages/chat/rag_setting.py

@@ -0,0 +1,70 @@
+import logging
+import os
+
+import gradio as gr
+import requests
+from ktem.app import BasePage
+
+logger = logging.getLogger(__name__)
+
+
+DEFAULT_KNET_ENDPOINT = "http://127.0.0.1:8081"
+KNET_ENDPOINT = os.environ.get("KN_ENDPOINT", DEFAULT_KNET_ENDPOINT)
+
+
+class RAGSetting(BasePage):
+    """Manage RAG settings from KNet"""
+
+    def __init__(self, app):
+        self._app = app
+        self.setting_state = gr.State(value={})
+        self.on_building_ui()
+        self.on_register_events()
+
+    def get_pipelines(self):
+        """Retrieve pipeline list from KNet endpoint"""
+        try:
+            response = requests.get(KNET_ENDPOINT + "/query_type", timeout=5)
+            if response.status_code == 200:
+                output = response.json()
+                return [item["name"] for item in output["pipelines"]]
+            else:
+                raise IOError(f"{response.status_code}: {response.text}")
+        except Exception as e:
+            logger.error(f"Failed to retrieve KNet pipelines: {e}")
+            return []
+
+    def on_building_ui(self):
+        pipline_options = self.get_pipelines()
+        self.pipeline_select = gr.Dropdown(
+            label="Pipeline",
+            choices=pipline_options,
+            value=pipline_options[0] if pipline_options else None,
+            container=False,
+            interactive=True,
+        )
+        self.retrieval_expansion = gr.Checkbox(
+            label="Enable retrieval expansion",
+            value=False,
+            container=False,
+        )
+
+    def store_setting_state(self, pipeline, retrieval_expansion):
+        return {
+            "pipeline": pipeline,
+            "retrieval_expansion": retrieval_expansion,
+        }
+
+    def on_register_events(self):
+        gr.on(
+            triggers=[
+                self.pipeline_select.change,
+                self.retrieval_expansion.change,
+            ],
+            fn=self.store_setting_state,
+            inputs=[
+                self.pipeline_select,
+                self.retrieval_expansion,
+            ],
+            outputs=self.setting_state,
+        )