Updated to incl SERPER_API and Openai

app.py

@@ -1,25 +1,20 @@
-from datasets import load_dataset
+import gradio as gr
+from huggingface_hub import InferenceClient
 from transformers import BertTokenizer, BertForSequenceClassification, TrainingArguments, Trainer
 import openai
+import os
 import faiss
 import numpy as np
+import requests
 
-# Set up OpenAI API key for GPT-4
-openai.api_key = "your_openai_api_key"
+# Load OpenAI and Serper API keys from Hugging Face secrets
+openai.api_key = os.getenv("OPENAI_API_KEY")  # Ensure the OpenAI API key is pulled correctly
+serper_api_key = os.getenv("SERPER_API_KEY")  # Ensure the Serper API key is pulled correctly
 
-# Load PubMedBERT tokenizer and model
+# Load PubMedBERT tokenizer and model for FDA-related processing
 tokenizer = BertTokenizer.from_pretrained("microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract")
 model = BertForSequenceClassification.from_pretrained("microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract", num_labels=2)
 
-# Load the FDA dataset from Hugging Face
-dataset = load_dataset("pretzinger/cdx-cleared-approved", split="train")
-
-# Tokenize the dataset
-def tokenize_function(example):
-    return tokenizer(example["text"], padding="max_length", truncation=True)
-
-tokenized_dataset = dataset.map(tokenize_function, batched=True)
-
 # FAISS setup for vector search (embedding-based memory)
 dimension = 768  # PubMedBERT embedding size
 index = faiss.IndexFlatL2(dimension)
@@ -40,43 +35,96 @@ def search_memory(query):
     D, I = index.search(query_embedding, k=1)  # Retrieve most similar past conversation
     return I
 
-# Function to handle FDA-related queries with PubMedBERT
+# Function to handle FDA-specific queries with PubMedBERT
 def handle_fda_query(query):
-    # If query requires specific FDA info, process it with PubMedBERT
     inputs = tokenizer(query, return_tensors="pt", padding="max_length", truncation=True)
     outputs = model(**inputs)
     logits = outputs.logits
-    # Process logits for classification or output a meaningful response
     response = "Processed FDA-related query via PubMedBERT"
     return response
 
-# Function to handle general queries using GPT-4
+# Function to handle general queries using GPT-4o
 def handle_openai_query(prompt):
     response = openai.Completion.create(
-        engine="gpt-4",  # Ensuring GPT-4 usage
+        engine="gpt-4o",  # Using GPT-4o as per instruction
         prompt=prompt,
         max_tokens=100
     )
     return response.choices[0].text.strip()
 
-# Main assistant function that delegates to either OpenAI or PubMedBERT
-def assistant(query):
-    # First, determine if query needs FDA-specific info
-    openai_response = handle_openai_query(f"Is this query FDA-related: {query}")
+# Web search with Serper API
+def web_search(query):
+    url = f"https://google.serper.dev/search"
+    headers = {
+        "X-API-KEY": serper_api_key
+    }
+    params = {
+        "q": query
+    }
+    response = requests.get(url, headers=headers, params=params)
+    return response.json()
+
+# Main assistant function that delegates to either OpenAI, PubMedBERT, or Serper (web search)
+def respond(
+    message,
+    history: list[tuple[str, str]],
+    system_message,
+    max_tokens,
+    temperature,
+    top_p,
+):
+    # Prepare the context for OpenAI and PubMedBERT
+    messages = [{"role": "system", "content": system_message}]
+
+    for val in history:
+        if val[0]:
+            messages.append({"role": "user", "content": val[0]})
+        if val[1]:
+            messages.append({"role": "assistant", "content": val[1]})
+
+    messages.append({"role": "user", "content": message})
+
+    # Check if the query is related to FDA
+    openai_response = handle_openai_query(f"Is this query FDA-related: {message}")
     
     if "FDA" in openai_response or "regulatory" in openai_response:
         # Search past conversations/memory using FAISS
-        memory_index = search_memory(query)
+        memory_index = search_memory(message)
         if memory_index:
             return f"Found relevant past memory: {past_conversation}"  # Return past context from memory
 
         # If no memory match, proceed with PubMedBERT
-        return handle_fda_query(query)
-    
-    # General conversational handling with OpenAI (GPT-4)
-    return openai_response
+        return handle_fda_query(message)
+
+    # If query asks for a web search, perform web search
+    if "search the web" in message.lower():
+        return web_search(message)
+
+    # General conversational handling with GPT-4o
+    response = ""
+    for message in client.chat_completion(
+        messages,
+        max_tokens=max_tokens,
+        stream=True,
+        temperature=temperature,
+        top_p=top_p,
+    ):
+        token = message.choices[0].delta.content
+
+        response += token
+        yield response
+
+
+# Create Gradio ChatInterface for interaction
+demo = gr.ChatInterface(
+    respond,
+    additional_inputs=[
+        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
+        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
+        gr.Slider(minimum=0.1, maximum 4.0, value=0.7, step=0.1, label="Temperature"),
+        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)")
+    ],
+)
 
-# Example Usage
-query = "What is required for PMA approval for companion diagnostics?"
-response = assistant(query)
-print(response)
+if __name__ == "__main__":
+    demo.launch()