Spaces:
Sleeping
Sleeping
| """ | |
| docker build -t wm-detector . | |
| docker run -p 7860:7860 -v $(pwd)/data:/app/data wm-detector | |
| """ | |
| from flask import Flask, render_template, request | |
| import torch | |
| import numpy as np | |
| from src.detector import MarylandDetector, AutoTokenizer | |
| app = Flask(__name__) | |
| # Minimal setup: pick a detector (example: MarylandDetector) | |
| model_id = "meta-llama/Llama-3.2-1B-Instruct" | |
| tokenizer = AutoTokenizer.from_pretrained(model_id) | |
| detector = MarylandDetector(tokenizer=tokenizer) | |
| def tokenize_text(text): | |
| return tokenizer.encode(text, add_special_tokens=False) | |
| def compute_scores(tokens): | |
| # Convert tokens to the detector's format | |
| score_list = [] | |
| for i in range(len(tokens)): | |
| if i < detector.ngram: | |
| score_list.append(0) | |
| continue | |
| ngram_tokens = tokens[i-detector.ngram:i] | |
| curr_score = detector.score_tok(ngram_tokens, tokens[i]).sum().item() | |
| score_list.append(curr_score) | |
| # Compute final p-value (example uses sum of scores) | |
| final_pvalue = detector.get_pvalue(sum(score_list), len(score_list), 1e-10) | |
| return score_list, final_pvalue | |
| def index(): | |
| tokens, colors, pvalue = [], [], None | |
| if request.method == "POST": | |
| user_text = request.form.get("user_text", "") | |
| tokens = tokenize_text(user_text) | |
| score_list, pvalue = compute_scores(tokens) | |
| # Convert token IDs to text | |
| displayed_tokens = tokenizer.convert_ids_to_tokens(tokens) | |
| # Assign a simple color scale based on score | |
| max_score = max(score_list) if score_list else 1 | |
| colors = [f"rgba(255, 0, 0, {s/max_score})" if max_score!=0 else "white" | |
| for s in score_list] | |
| return render_template("index.html", | |
| tokens=displayed_tokens, | |
| colors=colors, | |
| pvalue=pvalue) | |
| return render_template("index.html", tokens=tokens, colors=colors, pvalue=pvalue) | |
| if __name__ == "__main__": | |
| app.run(host='0.0.0.0', port=7860) | |