Upload 4 files

Dockerfile

@@ -0,0 +1,26 @@
+# Use the official Python 3.10 slim image
+FROM python:3.10-slim
+
+# Set the working directory to /app
+WORKDIR /app
+
+# Copy the requirements file into the container at /app
+COPY requirements.txt .
+
+# Install any needed packages specified in requirements.txt
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy the current directory contents into the container at /app
+COPY . .
+
+# Create a non-root user and switch to it (required by Hugging Face Spaces)
+RUN useradd -m -u 1000 user
+USER user
+ENV HOME=/home/user \
+	PATH=/home/user/.local/bin:$PATH
+
+# Expose port 7860 (Hugging Face Spaces default)
+EXPOSE 7860
+
+# Run uvicorn when the container launches
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]

main.py

@@ -0,0 +1,22 @@
+from fastapi import FastAPI
+from pydantic import BaseModel
+from fastapi.middleware.cors import CORSMiddleware
+
+from model_service import correct_code_with_ai
+app = FastAPI()
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"], # Allows all origins for simplicity.
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+class CodeSnippet(BaseModel):
+    code: str
+
+@app.post("/api/correct")
+def correct_code_endpoint(snippet: CodeSnippet):
+    corrected_code = correct_code_with_ai(snippet.code)
+    return {"corrected_code": corrected_code}
+

model_service.py

@@ -0,0 +1,62 @@
+from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
+
+# Initialize the model pipeline.
+# We use 'Qwen/Qwen2.5-0.5B-Instruct' which is small, fast, and works natively.
+model_id = "Qwen/Qwen2.5-0.5B-Instruct"
+print(f"Loading AI model ({model_id})...")
+
+code_fixer = None
+try:
+    # 1. Try loading from local cache first (offline mode)
+    print("Attempting to load from local cache...")
+    model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True, local_files_only=True)
+    tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True, local_files_only=True)
+    code_fixer = pipeline("text-generation", model=model, tokenizer=tokenizer)
+    print("Success: Loaded model from local cache.")
+except Exception as local_err:
+    print(f"Local cache not found or incomplete: {local_err}")
+    print("Attempting to download model from Hugging Face (requires internet)...")
+    try:
+        # 2. Fallback to downloading (online mode)
+        code_fixer = pipeline("text-generation", model=model_id, trust_remote_code=True)
+        print("Success: Model downloaded and loaded.")
+    except Exception as remote_err:
+        print(f"CRITICAL: Failed to load model. Error: {remote_err}")
+        print("To run locally, ensure you have internet access for the first run to download the model.")
+        code_fixer = None
+
+def correct_code_with_ai(code: str) -> str:
+    """
+    Takes a buggy code snippet and returns a corrected version using the Qwen model.
+    """
+    if not code_fixer:
+        return "# Model failed to load. Check server logs."
+
+    # Frame the input as a chat conversation
+    messages = [
+        {"role": "system", "content": "You are a helpful Python coding assistant. Your task is to fix bugs, suggest better variable naming in form of comments in the provided code. Return ONLY the corrected code, without explanation."},
+        {"role": "user", "content": f"{code}"},
+    ]
+    
+    try:
+        # Generate the response
+        # max_new_tokens controls how much new text is generated.
+        outputs = code_fixer(messages, max_new_tokens=512)
+        
+        # The pipeline for chat-like input typically returns a list of dictionaries.
+        # We need to parse the output to get just the assistant's response.
+        # The structure is usually: [{'generated_text': [...conversation including response...]}]
+        # or sometimes just the generated text depending on pipeline version.
+        
+        result = outputs[0]['generated_text']
+        
+        # If the result is the full conversation list (common in newer transformers for chat)
+        if isinstance(result, list):
+            # The last message should be the assistant's response
+            return result[-1]['content']
+        else:
+            # Fallback if it returns a string
+            return result
+    except Exception as e:
+        print(f"An error occurred during AI correction: {e}")
+        return f"# Unable to correct the code. Error: {str(e)}"

requirements.txt

@@ -0,0 +1,4 @@
+fastapi
+uvicorn
+transformers
+torch