Create app.py

app.py

@@ -0,0 +1,220 @@
+import streamlit as st
+from huggingface_hub import InferenceClient, login
+from transformers import HfAgent, HuggingFaceHubTool
+import os
+import pickle
+
+# Initialize Hugging Face login
+login(token=os.environ.get("HUGGINGFACE_API_KEY"))
+
+# Initialize HfAgent
+agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder")
+
+# Add custom tools
+date_extraction_tool = HuggingFaceHubTool("dslim/bert-base-NER")
+claim_number_tool = HuggingFaceHubTool("distilbert-base-uncased-finetuned-sst-2-english")
+call_purpose_tool = HuggingFaceHubTool("facebook/bart-large-mnli")
+
+agent.add_tool(date_extraction_tool)
+agent.add_tool(claim_number_tool)
+agent.add_tool(call_purpose_tool)
+
+st.title("Transcription Analysis Agent")
+
+# Base URL for Inference API
+BASE_URL = "https://api-inference.huggingface.co/models/"
+
+model_links = {
+    "LegacyLift🚀": BASE_URL + "mistralai/Mistral-7B-Instruct-v0.3",
+    "ModernMigrate⭐": BASE_URL + "mistralai/Mixtral-8x7B-Instruct-v0.1",
+    "RetroRecode🔄": BASE_URL + "microsoft/Phi-3-mini-4k-instruct"
+}
+
+model_info = {
+    "LegacyLift🚀": {
+        'description': "The LegacyLift model is a **Large Language Model (LLM)** for problem-solving, content writing, and daily tips.",
+        'logo': './11.jpg'
+    },
+    "ModernMigrate⭐": {
+        'description': "The ModernMigrate model excels in coding, logical reasoning, and high-speed inference.",
+        'logo': './2.jpg'
+    },
+    "RetroRecode🔄": {
+        'description': "The RetroRecode model is ideal for critical development, practical knowledge, and serverless inference.",
+        'logo': './3.jpg'
+    },
+}
+
+def format_prompt(message, conversation_history, custom_instructions=None):
+    prompt = f"\[INST\] {custom_instructions} \[/INST\]\n\[CONV_HISTORY\]\n"
+    for role, content in conversation_history:
+        prompt += f"{role.upper()}: {content}\n"
+    prompt += f"\[/CONV_HISTORY\]\n\[INST\] {message} \[/INST\]\n\[RESPONSE\]\n"
+    return prompt
+
+def reset_conversation():
+    st.session_state.conversation = []
+    st.session_state.messages = []
+    st.session_state.chat_state = "reset"
+
+def load_conversation_history():
+    history_file = "conversation_history.pickle"
+    return pickle.load(open(history_file, "rb")) if os.path.exists(history_file) else []
+
+def save_conversation_history(conversation_history):
+    with open("conversation_history.pickle", "wb") as f:
+        pickle.dump(conversation_history, f)
+
+def analyze_transcript(transcript, model):
+    client = InferenceClient(model=model)
+    
+    # Extract dates
+    dates_prompt = f"Extract dates from this text: {transcript}"
+    dates = client.text_generation(dates_prompt, max_new_tokens=50)
+    
+    # Extract claim number
+    claim_prompt = f"Find the claim number in this text: {transcript}"
+    claim_number = client.text_generation(claim_prompt, max_new_tokens=50)
+    
+    # Determine call purpose
+    purpose_prompt = f"What is the main purpose of this call based on the transcript: {transcript}"
+    call_purpose = client.text_generation(purpose_prompt, max_new_tokens=100)
+    
+    return {
+        "dates": dates,
+        "claim_number": claim_number,
+        "call_purpose": call_purpose
+    }
+
+# Sidebar UI
+models = list(model_links.keys())
+selected_model = st.sidebar.selectbox("Select Model", models)
+temp_values = st.sidebar.slider('Select Temperature', 0.0, 1.0, 0.5)
+st.sidebar.button('Reset Chat', on_click=reset_conversation)
+
+st.sidebar.write(f"Chatting with **{selected_model}**")
+st.sidebar.markdown(model_info[selected_model]['description'])
+st.sidebar.image(model_info[selected_model]['logo'])
+
+# Load session state
+if "prev_option" not in st.session_state:
+    st.session_state.prev_option = selected_model
+
+if st.session_state.prev_option != selected_model:
+    st.session_state.messages = []
+    st.session_state.prev_option = selected_model
+
+if "chat_state" not in st.session_state:
+    st.session_state.chat_state = "normal"
+
+if "messages" not in st.session_state:
+    st.session_state.messages = load_conversation_history()
+
+# Main Chat
+repo_id = model_links[selected_model]
+st.subheader(f"{selected_model}")
+
+if st.session_state.chat_state == "normal":
+    for message in st.session_state.messages:
+        with st.chat_message(message["role"]):
+            st.markdown(message["content"])
+
+    if prompt := st.chat_input(f"Hi I'm {selected_model}, How can I help you today?"):
+        custom_instruction = (
+            "Analyze this transcript with precision. Remove commas in claim numbers, preserve exact numbers and "
+            "dates (dd/mm/yy). Extract claim numbers as single entities."
+            "Analyze this transcript with precision. Remove commas in claim numbers, preserve exact numbers and "
+            "dates (dd/mm/yy). Extract claim numbers as single entities."
+            """1. Only include information explicitly stated
+            2. Mark unclear information as "UNCLEAR"
+            
+            3. Preserve exact numbers, dates (in dd/mm/yy format), and claims
+            
+            4. Focus on factual content
+            
+            IMPORTANT REQUIREMENTS:
+            
+            - Format all dates as dd/mm/yy
+            
+            - Extract and list all claim numbers mentioned
+            
+            - Maintain exact numbers and statistics as stated
+            
+            - Do not make assumptions about unclear information
+            
+            Please analyze the following transcript and structure your response as follows:
+            
+            PARTICIPANTS:
+            
+            - List all participants and their roles (if mentioned)
+            
+            CONTEXT:
+            
+            - Meeting purpose
+            
+            - Duration (if mentioned)
+            
+            - Meeting date/time
+            
+            KEY POINTS:
+            
+            - Main topics discussed
+            
+            - Decisions made
+            
+            - Important numbers/metrics mentioned
+            
+            - Claims discussed
+            
+            ACTION ITEMS:
+            
+            - Specific tasks assigned
+            
+            - Who is responsible
+            
+            - Deadlines (in dd/mm/yy format)
+            
+            FOLLOW UP:
+            
+            - Scheduled next meetings
+            
+            - Pending items
+            
+            - Required approvals or confirmations"""
+            # ... (rest of the custom instructions)
+        )
+
+        conversation_history = [(msg["role"], msg["content"]) for msg in st.session_state.messages]
+        formatted_text = format_prompt(prompt, conversation_history, custom_instruction)
+
+        with st.chat_message("user"):
+            st.markdown(prompt)
+
+        st.session_state.messages.append({"role": "user", "content": prompt})
+
+        with st.chat_message("assistant"):
+            try:
+                client = InferenceClient(model=repo_id)
+                response = client.text_generation(
+                    formatted_text,
+                    temperature=temp_values,
+                    max_new_tokens=1024,
+                    stream=True
+                )
+                response_text = ''.join(response)
+                response_text = response_text.replace(",", "")
+                st.markdown(response_text)
+                
+                # Use the agent to analyze the transcript
+                analysis_results = analyze_transcript(prompt, repo_id)
+                st.markdown("### Transcript Analysis")
+                st.json(analysis_results)
+                
+                st.session_state.messages.append({"role": "assistant", "content": response_text})
+                save_conversation_history(st.session_state.messages)
+            except Exception as e:
+                st.error(f"An error occurred: {e}")
+
+elif st.session_state.chat_state == "reset":
+    st.session_state.chat_state = "normal"
+    st.experimental_rerun()