Update app.py

app.py

@@ -1,159 +1,163 @@
-import gradio as gr
-import numpy as np
-import random
-from diffusers import DiffusionPipeline
+import os
+import re
+import json
 import torch
-
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model_repo_id = "stabilityai/sdxl-turbo"
-
-if torch.cuda.is_available():
-    torch_dtype = torch.float16
-else:
-    torch_dtype = torch.float32
-
-pipe = DiffusionPipeline.from_pretrained(
-    model_repo_id,
-    torch_dtype=torch_dtype,
-    variant="fp16"
+import torch.nn.functional as F
+from flask import Flask, request, jsonify
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+from pyvi import ViTokenizer
+from PIL import Image
+from torchvision import transforms
+import timm
+import requests
+from flask_cors import CORS
+
+# =====================
+# CONFIG
+# =====================
+os.environ["TRANSFORMERS_CACHE"] = "/tmp/hf_cache"  # tránh vượt storage limit
+TEXT_MODEL_REPO = "phuongsuga/PBL6_AI_Model_Text_Image"
+IMAGE_MODEL_REPO = "phuongsuga/PBL6_AI_Model_Image"
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+THRESHOLD = 0.65
+
+# =====================
+# LOAD TEXT MODEL
+# =====================
+print("🔹 Downloading text model...")
+
+tokenizer = AutoTokenizer.from_pretrained(
+    TEXT_MODEL_REPO,
+    subfolder="text_model",
+    use_fast=False
 )
 
-pipe = pipe.to(device)
-pipe.enable_attention_slicing()
-pipe.to("cuda" if torch.cuda.is_available() else "cpu")
-
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
-
-
-# @spaces.GPU #[uncomment to use ZeroGPU]
-def infer(
-    prompt,
-    negative_prompt,
-    seed,
-    randomize_seed,
-    width,
-    height,
-    guidance_scale,
-    num_inference_steps,
-    progress=gr.Progress(track_tqdm=True),
-):
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
-
-    generator = torch.Generator().manual_seed(seed)
-
-    image = pipe(
-        prompt=prompt,
-        negative_prompt=negative_prompt,
-        guidance_scale=guidance_scale,
-        num_inference_steps=num_inference_steps,
-        width=width,
-        height=height,
-        generator=generator,
-    ).images[0]
-
-    return image, seed
-
-
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
-]
-
-css = """
-#col-container {
-    margin: 0 auto;
-    max-width: 640px;
-}
-"""
-
-with gr.Blocks(css=css) as demo:
-    with gr.Column(elem_id="col-container"):
-        gr.Markdown(" # Text-to-Image Gradio Template")
-
-        with gr.Row():
-            prompt = gr.Text(
-                label="Prompt",
-                show_label=False,
-                max_lines=1,
-                placeholder="Enter your prompt",
-                container=False,
-            )
-
-            run_button = gr.Button("Run", scale=0, variant="primary")
-
-        result = gr.Image(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced Settings", open=False):
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
-
-            seed = gr.Slider(
-                label="Seed",
-                minimum=0,
-                maximum=MAX_SEED,
-                step=1,
-                value=0,
-            )
-
-            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
-
-            with gr.Row():
-                width = gr.Slider(
-                    label="Width",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-                height = gr.Slider(
-                    label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-            with gr.Row():
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0,  # Replace with defaults that work for your model
-                )
-
-                num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
-                    minimum=1,
-                    maximum=50,
-                    step=1,
-                    value=2,  # Replace with defaults that work for your model
-                )
-
-        gr.Examples(examples=examples, inputs=[prompt])
-    gr.on(
-        triggers=[run_button.click, prompt.submit],
-        fn=infer,
-        inputs=[
-            prompt,
-            negative_prompt,
-            seed,
-            randomize_seed,
-            width,
-            height,
-            guidance_scale,
-            num_inference_steps,
-        ],
-        outputs=[result, seed],
-    )
-
+text_model = AutoModelForSequenceClassification.from_pretrained(
+    TEXT_MODEL_REPO,
+    subfolder="text_model/checkpoint-3390"
+)
+text_model.to(DEVICE).eval()
+
+# load label2id.json
+label_url = f"https://huggingface.co/{TEXT_MODEL_REPO}/resolve/main/text_model/label2id.json"
+label2id = requests.get(label_url).json()
+id2label_text = {i: l for l, i in label2id.items()}
+
+# =====================
+# LOAD IMAGE MODEL
+# =====================
+print("🔹 Downloading image model...")
+
+class_names = ["an_toan", "bao_luc", "khieu_dam_doi_truy", "nhay_cam_chinh_tri"]
+
+def build_model(num_classes=4):
+    return timm.create_model("efficientnet_b3", pretrained=False, num_classes=num_classes)
+
+image_model = build_model()
+
+# tải model tạm trong /tmp để không chiếm storage
+image_model_path = "/tmp/efficientnet_b3.pth"
+if not os.path.exists(image_model_path):
+    url = f"https://huggingface.co/{IMAGE_MODEL_REPO}/resolve/main/image_model/efficientnet_b3.pth"
+    torch.hub.download_url_to_file(url, image_model_path)
+
+image_model.load_state_dict(torch.load(image_model_path, map_location=DEVICE))
+image_model.to(DEVICE).eval()
+
+# chuẩn hóa ảnh
+val_transforms = transforms.Compose([
+    transforms.Resize((224, 224)),
+    transforms.ToTensor(),
+    transforms.Normalize([0.485, 0.456, 0.406],
+                         [0.229, 0.224, 0.225])
+])
+
+# =====================
+# UTILS
+# =====================
+def seg_pyvi(text: str) -> str:
+    try:
+        seg = ViTokenizer.tokenize(text)
+        seg = seg.replace(" ", "_")
+    except Exception:
+        seg = text
+    return seg
+
+def split_sentences(text: str):
+    sents = re.split(r'(?<=[.!?])\s+|\n+', text.strip())
+    return [s for s in sents if s.strip()]
+
+def predict_text(text: str):
+    sentences = split_sentences(text)
+    results = []
+    for sent in sentences:
+        seg = seg_pyvi(sent)
+        inputs = tokenizer(seg, truncation=True, padding="max_length",
+                           max_length=128, return_tensors="pt").to(DEVICE)
+        with torch.no_grad():
+            logits = text_model(**inputs).logits
+            probs = F.softmax(logits, dim=-1).cpu().numpy()[0]
+        pred_id = probs.argmax()
+        label = id2label_text[int(pred_id)]
+        prob = float(probs[pred_id])
+        if label != "an_toan" and prob >= THRESHOLD:
+            results.append({"sentence": sent, "label": label, "confidence": prob})
+        else:
+            results.append({
+                "sentence": sent,
+                "label": "an_toan",
+                "confidence": float(probs[label2id["an_toan"]])
+            })
+    return results
+
+def predict_image(pil_image: Image.Image):
+    img = val_transforms(pil_image).unsqueeze(0).to(DEVICE)
+    with torch.no_grad():
+        outputs = image_model(img)
+        probs = F.softmax(outputs, dim=1)[0].cpu().numpy()
+    pred_id = probs.argmax()
+    label = class_names[pred_id]
+    prob = float(probs[pred_id])
+    if label != "an_toan" and prob >= THRESHOLD:
+        return {"label": label, "confidence": prob}
+    else:
+        return {
+            "label": "an_toan",
+            "confidence": float(probs[class_names.index("an_toan")])
+        }
+
+# =====================
+# FLASK APP
+# =====================
+app = Flask(__name__)
+CORS(app)
+
+@app.route("/")
+def home():
+    return jsonify({"message": "✅ AI moderation API is running!"})
+
+@app.route("/analyze", methods=["POST"])
+def analyze():
+    result = {"text_result": [], "image_result": []}
+
+    # 🧠 PHÂN TÍCH TEXT
+    if "content" in request.form:
+        text_input = request.form["content"]
+        result["text_result"] = predict_text(text_input)
+
+    # 🧠 PHÂN TÍCH NHIỀU ẢNH
+    if "image" in request.files:
+        image_files = request.files.getlist("image")
+        for image_file in image_files:
+            image = Image.open(image_file.stream).convert("RGB")
+            image_result = predict_image(image)
+            result["image_result"].append(image_result)
+
+    return jsonify(result)
+
+# =====================
+# RUN APP
+# =====================
 if __name__ == "__main__":
-    demo.launch()
+    port = int(os.environ.get("PORT", 7860))  # Hugging Face Space truyền PORT vào
+    app.run(host="0.0.0.0", port=port)