Update app.py

app.py

@@ -1,56 +1,123 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
+import os
+from openai import OpenAI
+import pandas as pd
+import numpy as np
+import json
+from typing import List, Dict
+import time
+from haystack.utils import Secret  # For securely storing the OpenAI API key
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("meta-llama/Llama-3.2-3B-Instruct")
+# Setup OpenAI client
+client = OpenAI(api_key=Secret.from_token(os.getenv('OPENAI_KEY')))
 
 
-def respond(
-    message,
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+def query_openai(messages: List[Dict[str, str]],
+                model: str = "gpt-4o-mini",
+                temperature: float = 0.7) -> str:
+    try:
+        response = client.chat.completions.create(
+            model=model,
+            messages=messages,
+            temperature=temperature
+        )
+        return response.choices[0].message.content
+    except Exception as e:
+        print(f"Error querying OpenAI: {e}")
+        return None
 
-    messages.append({"role": "user", "content": message})
+def evaluate_prompt(prompt_template: dict, review: str) -> pd.DataFrame:
+    messages = [
+        {"role": "system", "content": prompt_template['system']},
+        {"role": "user", "content": prompt_template['user'].replace('{{REVIEW}}', review)}
+    ]
+    response = query_openai(messages)
 
-    response = ""
+    # Extract classification from XML tags
+    import re
+    classification_match = re.search(r'<classification>(.*?)</classification>', response, re.IGNORECASE)
+    predicted = classification_match.group(1).strip().lower() if classification_match else "unknown"
 
-    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
+    return {
+        'predicted': predicted,
+        'response': response
+    }
 
-        response += token
-        yield response
+prompt_v3 = {
+    'system': """You are Roger Ebert and you are here to help us understand the sentiment of movie reviews.""",
+
+    'user': """Classify the following review as <classification>positive</classification> or <classification>negative</classification> (please stick to these labels), using a step-by-step analysis.
+
+    Output Format:
+    1. List key positive sentiment indicators
+    2. List key negative sentiment indicators
+    3. List anything that has a sentiment but is not relevant to the movie itself
+    4. Provide reasoning for your classification decision
+    5. End with classification in XML tags:
+    <classification>positive</classification> or <classification>negative</classification>
+    
+    Example 1:
+    
+    Input: I loved this movie. However the main actor didn't fit the role. But the plot totally saved it.
+    Output: <classification>positive</classification> 
+    
+    Example 2:
+    
+    Input: The movie was actually not THAT bad, especially plot-wise, but the doughy (and hairy!) actor they chose for the leading role was a little chintzy in the acting department. I would have chosen someone else. The idea of "going to America" was very ingenious, and the main character questioning everything that he'd ever known made him somewhat likable, but not very much so when there's a pasty blob for a leading actor.<br /><br />The storyline was interesting. It brings about the question of how the subject of cloning will be handled in the future. Certainly cloning wouldn't be allowed for the purposes in the movie, but it's still a valid argument even for today. Clones ARE still people... right?<br /><br />The movie wasn't particularly special, but it still is a lot better than some of the cheese released during the 70s. Let us not forget the "Giant Spider Invasion." I give it a 4, since it didn't TOTALLY stink, but the MST3K version makes this movie a 10. (I still like Dr. Super Mario!) You'll like this movie, but it won't be your favorite.
+    Output: <classification>negative</classification>
+    
+    Review to classify: {{REVIEW}}"""
+}
+    
+# def respond(
+#     message,
+#     system_message,
+#     max_tokens,
+#     temperature,
+#     top_p,
+# ):
+#     messages = [{"role": "system", "content": system_message}]
+
+#     messages.append({"role": "user", "content": message})
+
+#     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
+
+def respond(review):
+    result = evaluate_prompt(prompt_v3, review) 
+    return result['predicted'] 
 
 demo = gr.Interface(
     respond,
     inputs=[
         gr.Textbox(label="Movie review")  # Set label for the input
     ],
-    additional_inputs=[
-        gr.Textbox(value="You are a movie review classifier. You respond to given movie reviews with a predicted star rating 0-5 (inclusive) for that review with no explanation.", 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)",
-        ),
-    ],
+    # additional_inputs=[
+        # gr.Textbox(value="You are a movie review classifier. You respond to given movie reviews with a predicted star rating 0-5 (inclusive) for that review with no explanation.", 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)",
+        # ),
+    # ],
     outputs=[
-        gr.Textbox(label="Predicted Star rating")  # Set label for the output
+        gr.Textbox(label="Predicted Sentiment")  # Set label for the output
     ],
 )