v0.2.8.6: Baseline 03 - fix require einops for nomic - update Gradio embed components, add queue - update README - attempt fixing GenAI role:assistant with modify_history_in_place()

README.md

@@ -27,7 +27,7 @@ owner: research-semmyk
 #---
 
 #short_description: PDF & HTML parser to markdown
-version: 0.2.0
+version: 0.2.8.6
 readme: README.md
 requires-python: ">=3.12"
 #dependencies: []
@@ -96,13 +96,15 @@ gradio app_gradio_lightrag.py --demo-name=gradio_ui
 - For Colab: install requirements and run the app cell.
 
 ## Usage
-- Select your data folder (default: `dataset/data/docs`)
-- Choose LLM backend (OpenAI or Ollama). GenAI has a bug yieling error: role: 'assistant' instead of 'user' when updating history.
+- Browse/Select your data folder (default: `dataset/data/docs`)
+- Choose LLM backend (OpenAI or Ollama). [fix: GenAI has a bug yieling error: role:'assistant' instead of 'user' when updating history].
 - Activate the RAG constructor
 - Click 'Index Documents' to build the KG entities
 - Click 'Query' to get answers
 -- Enter your query and select query mode
-- Click 'Show Knowledge Graph' to visualise the KG
+- Click 'Show Knowledge Graph' to visualise the KG  
+
+NB: If using HuggingFace, log in first before browsing/selecting/uploading files and setting LLM parameters.
 
 ## Notes
 - Only markdown files are supported for ingestion (images in `/images` subfolder are ignored for now). <br>NB: other formats will be enabled later: pdf, txt, html...

app.py

@@ -89,7 +89,7 @@ def gradio_ui(app_logic: LightRAGApp):
                 with gr.Row():    
                     llm_backend_cb = gr.Radio(["OpenAI", "Ollama", "GenAI"], value="OpenAI", label="LLM Backend: OpenAI, Local or GenAI")
                     llm_model_name_tb = gr.Textbox(value=os.getenv("LLM_MODEL", "openai/gpt-oss-120b"), label="LLM Model Name", show_copy_button=True)  #.split('/')[1], label="LLM Model Name") "meta-llama/Llama-4-Maverick-17B-128E-Instruct")),  #image-Text-to-Text  #"openai/gpt-oss-120b",
-                with gr.Row():
+                #with gr.Row():
                     with gr.Row():  #elem_classes="password-box"):
                         #openai_key_tb = gr.Textbox(value=os.getenv("OPENAI_API_KEY", "jan-ai"), label="OpenAI API Key", 
                         #                           type="password", elem_classes="password-box", container=False, interactive=True, info="OpenAI API Key") #, show_copy_button=True)
@@ -98,15 +98,16 @@ def gradio_ui(app_logic: LightRAGApp):
                         toggle_btn_openai_key = gr.Button(
                                     value="👁️",  # Initial eye icon
                                     elem_classes="icon-button", size="sm")  #, min_width=50)
-                with gr.Row():    
+                with gr.Row():
                     openai_baseurl_tb = gr.Textbox(value=os.getenv("OPENAI_API_BASE", "https://router.huggingface.co/v1"), label="OpenAI baseurl", show_copy_button=True)
                     ollama_host_tb = gr.Textbox(value=os.getenv("OLLAMA_HOST", "http://localhost:1234/v1"), label="Ollama Host", show_copy_button=True)
                     #ollama_host_tb = gr.Textbox(value=os.getenv("OPENAI_API_EMBED_BASE", ""), label="Ollama Host")
-                with gr.Row():    
-                    openai_baseurl_embed_tb = gr.Textbox(value=os.getenv("OPENAI_API_EMBED_BASE", "http://localhost:1234/v1"), label="LLM Embed baseurl", show_copy_button=True)
-                    llm_model_embed_tb = gr.Textbox(value=os.getenv("LLM_MODEL_EMBED","text-embedding-bge-m3"), label="LLM Embedding Model", show_copy_button=True) #.split('/')[1], label="Embedding Model")
                 with gr.Row():
-                    embed_backend_dd = gr.Dropdown(choices=["Transformer", "Provider"], value="Provider", label="Embedding Type")
+                    embed_backend_dd = gr.Dropdown(choices=["Transformer", "Provider"], value="Transformer", label="Embedding Type")
+                    openai_baseurl_embed_tb = gr.Textbox(placeholder=os.getenv("OPENAI_API_EMBED_BASE", "http://localhost:1234/v1"), label="LLM Embed baseurl", show_copy_button=True)
+                    
+                with gr.Row():
+                    llm_model_embed_tb = gr.Textbox(placeholder=os.getenv("LLM_MODEL_EMBED","text-embedding-bge-m3"), label="LLM Embedding Model", show_copy_button=True) #.split('/')[1], label="Embedding Model")
                     with gr.Row():  #elem_classes="password-box"):
                         openai_key_embed_tb = gr.Textbox(value=os.getenv("OPENAI_API_KEY_EMBED", "jan-ai"), label="LLM API Key Embed",   #lm-studio
                                                 type="password", elem_classes="password-box", container=False, interactive=True, info="LLM API Key Embed") #, show_copy_button=True)
@@ -114,7 +115,7 @@ def gradio_ui(app_logic: LightRAGApp):
                                     value="👁️",  # Initial eye icon
                                     elem_classes="icon-button", size="sm")  #, min_width=50)
                     #openai_key_embed_tb = gr.Textbox(value=os.getenv("OPENAI_API_KEY_EMBED", "jan-ai"), label="OpenAI API Key Embed", type="password", show_copy_button=True)  #("OLLAMA_API_KEY", ""), label="OpenAI API Key Embed", type="password")
-        
+                
         # Step 1: Section 2
         with gr.Row():
                 with gr.Column():
@@ -130,7 +131,7 @@ def gradio_ui(app_logic: LightRAGApp):
                 working_dir_reset_cb = gr.Checkbox(value=False, label="Reset working files?")
         with gr.Accordion("🤗 HuggingFace Client Control", open=True):  #, open=False):
             # HuggingFace controls
-            hf_login_logout_btn = gr.LoginButton(value="Sign in to HuggingFace 🤗", logout_value="Logout of HF: ({}) 🤗", variant="huggingface")
+            hf_login_logout_btn = gr.LoginButton( variant="huggingface", value="Sign in to HuggingFace 🤗", logout_value="Logout of HF: ({}) 🤗 \n    [NB: check LLM settings & reload files (upload folder)]",)
 
         gr.Markdown("---")   #gr.HTML("<hr>")
         
@@ -191,7 +192,7 @@ def gradio_ui(app_logic: LightRAGApp):
                 openai_baseurl_tb: gr.update(value=os.getenv("OPENAI_API_BASE", "https://router.huggingface.co/v1")),
                 ollama_host_tb: gr.update(value=os.getenv("OLLAMA_HOST", "http://localhost:1234/v1")), #"http://localhost:11434"
                 openai_baseurl_embed_tb: gr.update(value=os.getenv("OPENAI_API_EMBED_BASE", "http://localhost:1234/v1")),   #"http://localhost:1234/v1/embeddings"
-                llm_model_embed_tb: gr.update(value=os.getenv("LLM_MODEL_EMBED","nomic-embed-text")),
+                llm_model_embed_tb: gr.update(value=os.getenv("LLM_MODEL_EMBED","nomic-embed-text")),   #"nomic-ai/nomic-embed-text-v1.5"
                 openai_key_embed_tb: gr.update(value=os.getenv("OPENAI_API_KEY_EMBED", "jan-ai"))
                 }
             elif llm_backend == "GenAI":
@@ -201,7 +202,7 @@ def gradio_ui(app_logic: LightRAGApp):
                 openai_baseurl_tb: gr.update(value=os.getenv("GEMINI_API_BASE", "https://generativelanguage.googleapis.com/v1beta/openai/"), label="GenAI baaseurl"),
                 ollama_host_tb: gr.update(value=os.getenv("OLLAMA_HOST", "http://localhost:11434")), #"http://localhost:1234/v1"
                 openai_baseurl_embed_tb: gr.update(value=os.getenv("OPENAI_API_EMBED_BASE", "http://localhost:1234/v1")),   #"http://localhost:1234/v1/embeddings"
-                llm_model_embed_tb: gr.update(value=os.getenv("LLM_MODEL_EMBED","all-MiniLM-L6-v2")),
+                llm_model_embed_tb: gr.update(value=os.getenv("LLM_MODEL_EMBED", "nomic-ai/nomic-embed-text-v1.5")), #"all-MiniLM-L6-v2")),
                 openai_key_embed_tb: gr.update(value=os.getenv("OPENAI_API_KEY_EMBED", "jan-ai"))
                 }
             elif llm_backend == "OpenAI":
@@ -215,6 +216,24 @@ def gradio_ui(app_logic: LightRAGApp):
                     openai_key_embed_tb: gr.update(value=os.getenv("OPENAI_API_KEY_EMBED", ""))
                 }
         
+        # Change Handling: update Ollama
+        def update_embedding_backend(embedding_backend):
+            """ Update LLM settings fields with ollama values"""
+            # Get model name excluding the model provider: # llm_model_name.rpartition("/")[-1]
+            
+            if embedding_backend == "Provider":
+                return {
+                openai_baseurl_embed_tb: gr.update(value=os.getenv("OPENAI_API_EMBED_BASE", "http://localhost:1234/v1")),   #"http://localhost:1234/v1/embeddings"
+                llm_model_embed_tb: gr.update(value=os.getenv("LLM_MODEL_EMBED","nomic-embed-text")),
+                openai_key_embed_tb: gr.update(value=os.getenv("OPENAI_API_KEY_EMBED", "jan-ai"))
+                }
+            elif embedding_backend == "Transformer":
+                return { 
+                openai_baseurl_embed_tb: gr.update(value=None, placeholder=os.getenv("OPENAI_API_EMBED_BASE", "http://localhost:1234/v1")),   #"http://localhost:1234/v1/embeddings"
+                llm_model_embed_tb: gr.update(value=None, placeholder=os.getenv("LLM_MODEL_EMBED", "nomic-ai/nomic-embed-text-v1.5")),   #(value="all-MiniLM-L6-v2"),
+                openai_key_embed_tb: gr.update(value=None, placeholder="jan-ai")
+                }
+        
         # Change Handling: Update password reveal state - reusable function for toggling password visibility
         def toggle_password(current_state):
             """ Change state
@@ -235,8 +254,9 @@ def gradio_ui(app_logic: LightRAGApp):
         # Update gr.State values on HF login change.
         def custom_do_logout(openai_key, oauth_token: gr.OAuthToken | None=None,):
             #'''  ##SMY: TO DELETE
+            st_openai_key_get = os.getenv("OPENAI_API_KEY", default="") #""  ##SMY:  # UnboundLocalError: not catching
             try:
-                if oauth_token:
+                if oauth_token or oauth_token is not None:  ##SMY: hack: is not None!
                     st_openai_key_get= update_state_stored_value(oauth_token.token)   ##SMY: currently not used optimally
             except AttributeError:
                 st_openai_key_get= get_login_token(openai_key)  #(openai_key_tb)
@@ -265,6 +285,9 @@ def gradio_ui(app_logic: LightRAGApp):
         llm_backend_cb.change(show_progress="hidden", fn=update_ollama, inputs=llm_backend_cb,  #inputs=None, 
                               outputs=[llm_model_name_tb, openai_key_tb, openai_baseurl_tb, ollama_host_tb, openai_baseurl_embed_tb, llm_model_embed_tb, openai_key_embed_tb])
 
+        embed_backend_dd.change(show_progress="hidden", fn=update_embedding_backend, inputs=embed_backend_dd, 
+                                outputs=[openai_baseurl_embed_tb,llm_model_embed_tb, openai_key_embed_tb],)
+        
         ### Button handlers
 
         #hf_login_logout_btn.click(update_state_stored_value, inputs=openai_key_tb, outputs=st_openai_key)
@@ -340,7 +363,8 @@ if __name__ == "__main__":
         # Instantiate LightRAG and launch Gradio
         try:
             app_logic = LightRAGApp()
-            gradio_ui(app_logic).launch()    #(server_port=7866)
+            # Launch Gradio with queue enable to enable >60s timeout
+            gradio_ui(app_logic).queue().launch()    #(server_port=7866)
         except Exception as e:
             print(f"An error occurred: {e}")
         finally:

app_gradio_lightrag.py

@@ -361,13 +361,63 @@ class LightRAGApp:
                                **kwargs) -> Union[str, types.Content]:
         """ Create GenAI client and complete a prompt """
         # https://github.com/googleapis/python-genai/tree/main
-
+        
         # 1. Combine prompts: system prompt, history, and user prompt
         if not history_messages or history_messages is None:
             history_messages = []
 
+        ## SMY: role mapping: attempting to map assistant with user  #role:assistant pydantic error
+        '''   ##SMY working code: disused. See modify_history_in_place()
+        ## SMY: new dictionary by unpacking (**) the existing message dictionary, explicitly set "role" key to "user"
+        history_messages1 = [{**message, 'role': 'user'} if message.get('role', ) == 'assistant' else message
+                            for message in history_messages]
+        for message in history_messages1: print(f"history len {len(history_messages1)} : \n {'\n'.join(f'{key}: {value[:25]}' for key, value in message.items())}") if isinstance(message, dict) else print(f"\n {str(message)[:25]}")
+        '''
+        '''
+        #idiomatic way to handle an in-place mutation within a list comprehension
+        #setattr(m, 'role', 'user') mutates the object in-place by changing its role attribute
+        #(..., m)[1] tricks returning the mutated object m after the setattr operation is completed, necessary for the list comprehension to work correctly
+        history_messages2 = [
+            (lambda m: (setattr(m, 'role', 'user'), m)[1] if message.get('role', ) == 'assistant' else m)(message)
+            for message in history_messages ]
+        for message in history_messages2: print(f"history len {len(history_messages2)} : \n {'\n'.join(f'{key}: {value[:25]}' for key, value in message.items())}") if isinstance(message, dict) else print(f"\n {str(message)[:25]}")
+        '''
+        ## one-liner to change all 'model' roles to 'user': List comprehensions are not intended for in-place mutations
+        #[(setattr(message, 'role', 'user'), message) for message in history_messages if hasattr(message, 'role') and message.role == 'assistant']
+
+        
+        def modify_history_in_place(history_messages):
+            """
+            Modifies the history_messages list in-place, converting 'assistant' roles to 'user'.
+            
+            Args:
+                history_messages: A list that may contain a mix of dicts and GenAI Content objects.
+            """
+            #history_messages_dict = history_messages    #debug
+            ## enumerating to avoid potential "off-by-one" errors
+            for index, message in enumerate(history_messages):
+                # Handle the custom GenAI Content object using its API
+                if hasattr(message, 'to_dict'):
+                    msg_dict = message.to_dict()
+                    #history_messages[index] = {**msg_dict, 'role': 'user'} if msg_dict.get(key='role', default_value='user') == 'assistant' else msg_dict
+                    
+                    ## SMY: {'role': []'user', 'model'], 'content': 'content_text'}
+                    role = 'user' if msg_dict.get('role', 'user') == 'assistant' else msg_dict.get('role', 'user')  #msg_dict.get(key='role', default_value='user') 
+                    parts = [types.Part.from_text(text=msg_dict.get('content', ''))]
+                    history_messages[index] = types.Content(role=role, parts=parts)
+                # Handle standard Python dictionaries
+                elif isinstance(message, dict):
+                    #history_messages[index] = {**message, 'role': 'user'} if message.get('role') == 'assistant' else message
+
+                    role = 'user' if message.get('role', 'user') == 'assistant' else message.get('role', 'user')
+                    parts = [types.Part.from_text(text=message.get('content', ''))]
+                    history_messages[index] = types.Content(role=role, parts=parts)
+            ##debug
+            #for message in history_messages: print(f"history len {len(history_messages)} : \n {'\n'.join(f'{key}: {value[:50]}' for key, value in message.items())}") if isinstance(message, dict) else print(f"\n {str(message)[:50]}")
+        modify_history_in_place(history_messages)
+            
+
         # prepare message
-        #messages: list[dict[str, Any]] = []
         messages: list[types.Content] = []
 
         if system_prompt:   ##See system_instruction
@@ -375,6 +425,7 @@ class LightRAGApp:
         new_user_content =  types.Content(role="user", parts=[types.Part.from_text(text=prompt)])
         history_messages.append(new_user_content)
 
+
         logger.debug(f"Sending messages to Gemini: Model: {self.llm_model_name.rpartition('/')[-1]} \n~ Message: {prompt}")
         logger_kg.log(level=20, msg=f"Sending messages to Gemini: Model: {self.llm_model_name.rpartition('/')[-1]} \n~ Message: {prompt}")
         
@@ -396,7 +447,7 @@ class LightRAGApp:
                     #automatic_function_calling=types.AutomaticFunctionCallingConfig(disable=False),
                     system_instruction=["You are an expert in Knowledge graph.",
                                         "You are well versed in entities, relations, objects and ontology reasoning",
-                                        "Your mission/task is to create/construct knowledge Graph"], #system_prompt,                    
+                                        "Your mission/task is to create/construct knowledge Graph, otherwise, query the Knowledge Graph when instructed"], #system_prompt,                    
                 )
             )
             ## GenAI keeps giving pydantic error relating to 'role': 'assistant'  #wierd

requirements.txt

@@ -14,6 +14,7 @@ nest_asyncio>=1.6.0     ##SMY:    #HF Spaces modulenotfounderror: No module name
 pyvis>=0.3.1            ##SMY: >=0.3.0  --> 0.3.2
 networkx>=3.4.2         ##SMY: 3.2.1  (3.4.2)
 sentence-transformers   ##SMY:
+einops>=0.8.1           ##SMY: sentence-transformer on HF Space - nomic-ai
 hf_xet                  ##SMY:    #HF Xet Storage downloader
 
 plotly>=6.0.1           ##SMY: 6.0.1