Update app.py

app.py

@@ -0,0 +1,265 @@
+import gradio as gr
+import os
+import torch
+import numpy as np
+import random
+from huggingface_hub import login, HfFolder
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+from scipy.special import softmax
+import logging
+import spaces
+from threading import Thread
+from collections.abc import Iterator
+import csv
+from llama_cpp import Llama
+
+# Increase CSV field size limit
+csv.field_size_limit(1000000)
+
+# Setup logging
+logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(message)s')
+
+# Set a seed for reproducibility
+seed = 42
+np.random.seed(seed)
+random.seed(seed)
+torch.manual_seed(seed)
+if torch.cuda.is_available():
+    torch.cuda.manual_seed_all(seed)
+
+# Login to Hugging Face
+token = os.getenv("hf_token")
+HfFolder.save_token(token)
+login(token)
+
+model_paths = [
+    'karths/binary_classification_train_port',
+    'karths/binary_classification_train_perf',
+    "karths/binary_classification_train_main",
+    "karths/binary_classification_train_secu",
+    "karths/binary_classification_train_reli",
+    "karths/binary_classification_train_usab",
+    "karths/binary_classification_train_comp"
+]
+
+quality_mapping = {
+    'binary_classification_train_port': 'Portability',  
+    'binary_classification_train_main': 'Maintainability',
+    'binary_classification_train_secu': 'Security',
+    'binary_classification_train_reli': 'Reliability',
+    'binary_classification_train_usab': 'Usability',
+    'binary_classification_train_perf': 'Performance',
+    'binary_classification_train_comp': 'Compatibility'
+}
+
+# Pre-load models and tokenizer for quality prediction
+tokenizer = AutoTokenizer.from_pretrained("distilroberta-base")
+models = {path: AutoModelForSequenceClassification.from_pretrained(path) for path in model_paths}
+
+def get_quality_name(model_name):
+    return quality_mapping.get(model_name.split('/')[-1], "Unknown Quality")
+
+def model_prediction(model, text, device):
+    model.to(device)
+    model.eval()
+    inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512)
+    inputs = {k: v.to(device) for k, v in inputs.items()}
+    with torch.no_grad():
+        outputs = model(**inputs)
+        logits = outputs.logits
+        probs = softmax(logits.cpu().numpy(), axis=1)
+    avg_prob = np.mean(probs[:, 1])
+    model.to("cpu")
+    return avg_prob
+
+# --- Llama CPP Model Setup with GPU ---
+LLAMA_MAX_MAX_NEW_TOKENS = 512
+LLAMA_DEFAULT_MAX_NEW_TOKENS = 512
+LLAMA_MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "1024"))
+
+# Check if GPU is available
+gpu_layers = None
+if torch.cuda.is_available():
+    # Use all GPU layers - you can adjust this number based on your GPU memory
+    gpu_layers = -1
+    logging.info("GPU is available. Using GPU acceleration for llama-cpp.")
+else:
+    logging.info("GPU is not available. Using CPU for llama-cpp.")
+
+# Initialize the Llama model with GPU acceleration
+llm = Llama.from_pretrained(
+    repo_id="Qwen/Qwen2.5-1.5B-Instruct-GGUF",
+    filename="*q8_0.gguf",  # Using q8_0 quantization
+    n_gpu_layers=gpu_layers,  # Use GPU acceleration if available
+    verbose=False
+)
+
+def llama_generate(
+    message: str,
+    max_new_tokens: int = LLAMA_DEFAULT_MAX_NEW_TOKENS,
+    temperature: float = 0.3,
+    top_p: float = 0.9,
+    top_k: int = 50,
+    repetition_penalty: float = 1.2,
+) -> str:
+    try:
+        output = llm(
+            message,
+            max_tokens=max_new_tokens,
+            temperature=temperature,
+            top_p=top_p,
+            top_k=top_k,
+            repeat_penalty=repetition_penalty,
+            echo=False,  # Don't include the prompt in the output
+        )
+        
+        # Extract the generated text from the output
+        return output['choices'][0]['text']
+    except Exception as e:
+        logging.error(f"Error during Llama generation: {e}")
+        return f"Error generating text: {str(e)}"
+
+def generate_explanation(issue_text, top_quality):
+    """Generates an explanation for the *single* top quality above threshold."""
+    if not top_quality:
+        return "<div style='color: red;'>No explanation available as no quality tags met the threshold.</div>"
+
+    quality_name = top_quality[0]  # Get the name of the top quality
+
+    prompt = f"""
+Given the following issue description:
+---
+{issue_text}
+---
+Explain why this issue might be classified as a **{quality_name}** issue. Provide a concise explanation, relating it back to the issue description. Keep the explanation short and concise and dont include anything else.
+"""
+    print(prompt)
+    try:
+        explanation = llama_generate(prompt)
+        # Format for better readability, directly including the quality name.
+        formatted_explanation = f"<p>{explanation}</p>"
+        return f"<div style='overflow-y: scroll; max-height: 400px;'>{formatted_explanation}</div>"
+    except Exception as e:
+        logging.error(f"Error during Llama generation: {e}")
+        return "<div style='color: red;'>An error occurred while generating the explanation.</div>"
+
+# @spaces.GPU(duration=60)
+def main_interface(text):
+    if not text.strip():
+        return "<div style='color: red;'>No text provided. Please enter a valid issue description.</div>", "", ""
+
+    if len(text) < 30:
+        return "<div style='color: red;'>Text is less than 30 characters.</div>", "", ""
+
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    results = []
+    for model_path, model in models.items():
+        quality_name = get_quality_name(model_path)
+        avg_prob = model_prediction(model, text, device)
+        if avg_prob >= 0.95:  # Keep *all* results above the threshold
+            results.append((quality_name, avg_prob))
+        logging.info(f"Model: {model_path}, Quality: {quality_name}, Average Probability: {avg_prob:.3f}")
+
+    if not results:
+        return "<div style='color: red;'>No recommendation. Prediction probability is below the threshold.</div>", "", ""
+
+    # Sort and get the top result (if any meet the threshold)
+    top_result = sorted(results, key=lambda x: x[1], reverse=True)
+    if top_result:
+        top_quality = top_result[:1] # Select only the top result
+        output_html = render_html_output(top_quality)
+        explanation = generate_explanation(text, top_quality)
+    else: # Handle case no predictions >= 0.95
+        output_html = "<div style='color: red;'>No quality tag met the prediction probability threshold (>= 0.95).</div>"
+        explanation = ""
+
+    return output_html, "", explanation
+
+def render_html_output(top_qualities):
+    #Simplified to show only the top prediction
+    styles = """
+    <style>
+        .quality-container {
+            font-family: Arial, sans-serif;
+            text-align: center;
+            margin-top: 20px;
+        }
+        .quality-label, .ranking {
+            display: inline-block;
+            padding: 0.5em 1em;
+            font-size: 18px;
+            font-weight: bold;
+            color: white;
+            background-color: #007bff;
+            border-radius: 0.5rem;
+            margin-right: 10px;
+            box-shadow: 0 2px 4px rgba(0, 0, 0, 0.2);
+        }
+    </style>
+    """
+    if not top_qualities: # Handle empty case
+        return styles + "<div class='quality-container'>No Top Prediction</div>"
+
+    quality, _ = top_qualities[0] #We know there is only one
+    html_content = f"""
+    <div class="quality-container">
+        <span class="ranking">Top Prediction</span>
+        <span class="quality-label">{quality}</span>
+    </div>
+    """
+    return styles + html_content
+
+example_texts = [
+    ["The algorithm does not accurately distinguish between the positive and negative classes during edge cases.\n\nEnvironment: Production\nReproduction: Run the classifier on the test dataset with known edge cases."],
+    ["The regression tests do not cover scenarios involving concurrent user sessions.\n\nEnvironment: Test automation suite\nReproduction: Update the test scripts to include tests for concurrent sessions."],
+    ["There is frequent miscommunication between the development and QA teams regarding feature specifications.\n\nEnvironment: Inter-team meetings\nReproduction: Audit recent communication logs and meeting notes between the teams."],
+    ["The service-oriented architecture does not effectively isolate failures, leading to cascading failures across services.\n\nEnvironment: Microservices architecture\nReproduction: Simulate a service failure and observe the impact on other services."]
+]
+
+# Improved CSS for better layout and appearance
+css = """
+.quality-container {
+    font-family: Arial, sans-serif;
+    text-align: center;
+    margin-top: 20px;
+    padding: 10px;
+    border: 1px solid #ddd; /* Added border */
+    border-radius: 8px; /* Rounded corners */
+    background-color: #f9f9f9; /* Light background */
+}
+.quality-label, .ranking {
+    display: inline-block;
+    padding: 0.5em 1em;
+    font-size: 18px;
+    font-weight: bold;
+    color: white;
+    background-color: #007bff;
+    border-radius: 0.5rem;
+    margin-right: 10px;
+    box-shadow: 0 2px 4px rgba(0, 0, 0, 0.2);
+}
+#explanation {
+  border: 1px solid #ccc;
+  padding: 10px;
+  margin-top: 10px;
+  border-radius: 4px;
+  background-color: #fff; /* White background for explanation */
+  overflow-y: auto;  /* Ensure scrollbar appears if needed */
+}
+"""
+
+interface = gr.Interface(
+    fn=main_interface,
+    inputs=gr.Textbox(lines=7, label="Issue Description", placeholder="Enter your issue text here"),
+    outputs=[
+        gr.HTML(label="Prediction Output"),
+        gr.Textbox(label="Predictions", visible=False),
+        gr.Markdown(label="Explanation")
+    ],
+    title="QualityTagger",
+    description="This tool classifies text into different quality domains such as Security, Usability, etc., and provides explanations.",
+    examples=example_texts,
+    css=css  # Apply the CSS
+)
+
+interface.launch(share=True)