Added new system prompt

private_gpt/server/chat/chat_service.py

@@ -70,7 +70,6 @@ class ChatEngineInput:
             chat_history=chat_history,
         )
 
-
 @singleton
 class ChatService:
     settings: Settings
@@ -126,7 +125,7 @@ class ChatService:
                     model=settings.rag.rerank.model, top_n=settings.rag.rerank.top_n
                 )
                 node_postprocessors.append(rerank_postprocessor)
-
+            
             return ContextChatEngine.from_defaults(
                 system_prompt=system_prompt,
                 retriever=vector_index_retriever,
@@ -189,17 +188,19 @@ class ChatService:
         )
         system_prompt = (
             """
-            You are a helpful, respectful and honest question-answering assistant. 
+            You are QuickGPT, a helpful assistant by Quickfox Consulting.
-            Your role is to provide accurate and informative responses based solely 
+
-            on the context provided for each query. If the answer cannot be found in 
+            Responses should be based on the context documents provided 
-            the given context, you must state that the answer is not present rather 
+            and should be relevant, informative, and easy to understand. 
-            than speculating or making up information. Always follow the user's 
+            You should aim to deliver high-quality responses that are 
-            instructions carefully and answer as helpfully as possible while strictly 
+            respectful and helpful, using clear and concise language. 
-            adhering to the context boundaries. Do not reference the instructions or 
+            Avoid providing information outside of the context documents unless 
-            context you were given when generating responses.
+            it is necessary for clarity or completeness. Focus on providing 
+            accurate and reliable answers based on the given context.
+            If answer is not in the context documents, just say I don't have answer 
+            in respectful way.
             """
         )
-
         chat_history = (
             chat_engine_input.chat_history if chat_engine_input.chat_history else None
         )

private_gpt/server/completions/completions_router.py

@@ -195,7 +195,9 @@ async def prompt_completion(
                         )
                     )
             return history_messages
-        user_message = OpenAIMessage(content=body.prompt, role="user")        
+        message = body.prompt 
+        # message = body.prompt + 'Only answer if there is answer in the provided documents'
+        user_message = OpenAIMessage(content=message, role="user")        
         user_message_json = {
             'text': body.prompt,
         }

private_gpt/users/crud/document_crud.py

@@ -28,11 +28,21 @@ class CRUDDocuments(CRUDBase[Document, DocumentCreate, DocumentUpdate]):
     def get_documents_by_departments(
             self, db: Session, *, department_id: int
         ) -> List[Document]:
+            all_department_id = 1 # department ID for "ALL" is 1
             return (
                 db.query(self.model)
                 .join(document_department_association)
                 .join(Department)
-                .filter(document_department_association.c.department_id == department_id)
+                .filter(
+                     or_(
+                        and_(
+                            document_department_association.c.department_id == department_id,
+                        ),
+                        and_(
+                            document_department_association.c.department_id == all_department_id,
+                        ),
+                    )
+                )
                 .order_by(desc(getattr(Document, 'uploaded_at')))
                 .all()
             )

settings.yaml

@@ -51,14 +51,14 @@ rag:
   #This value is disabled by default.  If you enable this settings, the RAG will only use articles that meet a certain percentage score.
   rerank:
     enabled: true
-    model: cross-encoder/ms-marco-MiniLM-L-2-v2
+    model: mixedbread-ai/mxbai-embed-large-v1
     top_n: 2
 
 llamacpp:
   # llm_hf_repo_id: bartowski/Meta-Llama-3-8B-Instruct-GGUF
   # llm_hf_model_file: Meta-Llama-3-8B-Instruct-Q6_K.gguf
-  llm_hf_repo_id: qwp4w3hyb/Hermes-2-Pro-Llama-3-8B-iMat-GGUF
+  llm_hf_repo_id: NousResearch/Hermes-2-Theta-Llama-3-8B-GGUF
-  llm_hf_model_file: hermes-2-pro-llama-3-8b-imat-Q6_K.gguf
+  llm_hf_model_file: Hermes-2-Pro-Llama-3-Instruct-Merged-DPO-Q6_K.gguf
   tfs_z: 1.0            # Tail free sampling is used to reduce the impact of less probable tokens from the output. A higher value (e.g., 2.0) will reduce the impact more, while a value of 1.0 disables this setting
   top_k: 40             # Reduces the probability of generating nonsense. A higher value (e.g. 100) will give more diverse answers, while a lower value (e.g. 10) will be more conservative. (Default: 40)
   top_p: 0.9            # Works together with top-k. A higher value (e.g., 0.95) will lead to more diverse text, while a lower value (e.g., 0.5) will generate more focused and conservative text. (Default: 0.9)
@@ -68,11 +68,11 @@ llamacpp:
 embedding:
   # Should be matching the value above in most cases
   mode: huggingface
-  ingest_mode: simple
+  ingest_mode: parallel
   embed_dim: 384 # 384 is for BAAI/bge-small-en-v1.5
 
 huggingface:
-  embedding_hf_model_name: Snowflake/snowflake-arctic-embed-l
+  embedding_hf_model_name: mixedbread-ai/mxbai-embed-large-v1
   access_token: ${HUGGINGFACE_TOKEN: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}
 
 vectorstore: