Upload folder using huggingface_hub

README.md

@@ -155,3 +155,27 @@ $ python main.py # or gradio main.py
 
 # Acknowledgments
 This is a project built during the Vertex sprints held by Google's ML Developer Programs team. We are thankful to be granted good amount of GCP credits to do this project. 
+# AdaptSum
+
+AdaptSum stands for Adaptive Summarization. This project focuses on developing an LLM-powered system for dynamic summarization. Instead of generating entirely new summaries with each update, the system intelligently identifies and modifies only the necessary parts of the existing summary. This approach aims to create a more efficient and fluid summarization process within a continuous chat interaction with an LLM.
+
+# Instructions
+
+1. Install dependencies
+```shell
+$ pip install requirements.txt
+```
+
+2. Setup Gemini API Key
+```shell
+$ export GEMINI_API_KEY=xxxxx
+```
+> note that GEMINI API KEY should be obtained from Google AI Studio. Vertex AI is not supported at the moment (this is because Gemini SDK does not provide file uploading functionality for Vertex AI usage now).
+
+3. Run Gradio app
+```shell
+$ python main.py # or gradio main.py
+```
+
+# Acknowledgments
+This is a project built during the Vertex sprints held by Google's ML Developer Programs team. We are thankful to be granted good amount of GCP credits to do this project. 

app.py

@@ -27,10 +27,11 @@ def find_attached_file(filename, attached_files):
             return file
     return None
 
-async def echo(message, history, state, persona):
+async def echo(message, history, state, persona, use_generated_summaries):
     attached_file = None
     system_instruction = Template(prompt_tmpl['summarization']['system_prompt']).safe_substitute(persona=persona)
-    
+    use_generated_summaries = True if use_generated_summaries == "Yes" else False
+
     if message['files']:
         path_local = message['files'][0]
         filename = os.path.basename(path_local)
@@ -38,31 +39,38 @@ async def echo(message, history, state, persona):
         attached_file = find_attached_file(filename, state["attached_files"])
         if attached_file is None: 
             path_gcp = await client.files.upload(path=path_local)
-            path_wrap = types.Part.from_uri(
-                file_uri=path_gcp.uri, mime_type=path_gcp.mime_type
-            )
             state["attached_files"].append({
                 "name": filename,
                 "path_local": path_local,
                 "gcp_entity": path_gcp,
-                "path_gcp": path_wrap,
-                "mime_type": path_gcp.mime_type,
+                "path_gcp": path_gcp.name,
+                "mime_type=": path_gcp.mime_type,
                 "expiration_time": path_gcp.expiration_time,
             })
-            attached_file = path_wrap
+            attached_file = path_gcp
 
-    user_message_parts = [types.Part.from_text(text=message['text'])]
-    if attached_file: user_message_parts.append(attached_file)
-    user_message = [types.Content(role='user', parts=user_message_parts)]
-    
-    state['messages'] = state['messages'] + user_message
-    print(state['messages'])
+    user_message = [message['text']]
+    if attached_file: user_message.append(attached_file)
+
+    chat_history = state['messages']
+    chat_history = chat_history + user_message
+    state['messages'] = chat_history
 
     response_chunks = ""
+    model_contents = ""
+    if use_generated_summaries:
+        if "summary_history" in state and len(state["summary_history"]):
+            model_contents += state["summary_history"][-1]
+        else:
+            model_contents = state['messages']
+    else:
+        model_contents = state['messages']
     model_content_stream = await client.models.generate_content_stream(
         model=args.model, 
-        contents=state['messages'], 
-        config=types.GenerateContentConfig(seed=args.seed),
+        contents=model_contents, 
+        config=types.GenerateContentConfig(
+            system_instruction=system_instruction, seed=args.seed
+        ),
     )
     async for chunk in model_content_stream:
         response_chunks += chunk.text
@@ -80,10 +88,6 @@ async def echo(message, history, state, persona):
             ),
             gr.DownloadButton(visible=False)
         )        
-
-    state['messages'] = state['messages'] + [
-        types.Content(role='model', parts=[types.Part.from_text(text=response_chunks)])
-    ]
     
     # make summary
     response = await client.models.generate_content(
@@ -92,7 +96,7 @@ async def echo(message, history, state, persona):
             Template(
                 prompt_tmpl['summarization']['prompt']
             ).safe_substitute(
-                previous_summary=state['summary'], 
+                previous_summary=state['summary'],
                 latest_conversation=str({"user": message['text'], "assistant": response_chunks})
             )
         ],
@@ -233,14 +237,20 @@ def main(args):
                 label="Summary Persona", 
                 info="Control the tonality of the conversation.",
                 min_width="auto",
-            )        
+            )      
+            use_generated_summaries = gr.Dropdown(
+                ["No", "Yes"], 
+                label="Feed back the generated summaries", 
+                info="Set this to 'Yes' to ONLY feed the generated summaries back to the model instead of the whole conversation.",
+                min_width="auto",
+            )      
 
         with gr.Column("chat-window", elem_id="chat-window"):
             gr.ChatInterface(
                 multimodal=True,
                 type="messages", 
                 fn=echo, 
-                additional_inputs=[state, persona],
+                additional_inputs=[state, persona, use_generated_summaries],
                 additional_outputs=[state, last_user_msg, summary_diff, summary_md, summary_num, download_summary_md],
             )