import gradio as gr
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

# --- Configuration ---
MODEL_NAME = "distilbert-base-uncased"
NUM_LABELS = 6
MODEL_PATH = "controlled_bert_model.pth" # The name of the file you uploaded

# --- Load Tokenizer and Model ---
print("Loading tokenizer...")
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)

print("Loading model architecture...")
# First, create the model "shell"
model = AutoModelForSequenceClassification.from_pretrained(
    MODEL_NAME, 
    num_labels=NUM_LABELS
)

print(f"Loading fine-tuned weights from {MODEL_PATH}...")
# Now, load your trained weights into the shell
model.load_state_dict(
    torch.load(MODEL_PATH, map_location=torch.device("cpu"))
)
model.eval() # Set model to evaluation mode
print("Model loaded successfully!")

def classify_log(log_text):
    """
    This function runs the classification using your loaded .pth model.
    """
    inputs = tokenizer(log_text, return_tensors="pt", padding=True, truncation=True)
    with torch.no_grad():
        logits = model(**inputs).logits
    
    scores = torch.softmax(logits, dim=1).squeeze().tolist()
    # Create a dictionary of {label_name: score}
    confidences = {model.config.id2label[i]: score for i, score in enumerate(scores)}
    return confidences

# This creates the Gradio interface and API endpoint
gr.Interface(
    fn=classify_log,
    inputs=gr.Textbox(lines=5, label="Log Entry"),
    outputs=gr.Label(num_top_classes=6, label="Classification Results"),
    title="Infrnce Private Log Classifier API"
).launch()