Upload 3 files

best_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:015516272417da498a4797a4dcb97d006bdba009a2583d6739853968bcce434b
+size 860265152

inference.py

@@ -0,0 +1,51 @@
+
+# inference.py  ——  drop this next to best_model.pt for your webapp
+import re, json, torch, torch.nn.functional as F
+from pathlib import Path
+from PIL import Image
+from torchvision import transforms
+from transformers import BertTokenizer
+# Import your model class here:
+# from model import MultimodalSentimentModel
+
+DEVICE  = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+META    = json.load(open("deployment/model_meta.json"))
+CONFIG  = META["config"]
+
+tokenizer = BertTokenizer.from_pretrained(CONFIG["BERT_MODEL"])
+
+img_transform = transforms.Compose([
+    transforms.Resize((CONFIG["IMAGE_SIZE"], CONFIG["IMAGE_SIZE"])),
+    transforms.ToTensor(),
+    transforms.Normalize(META["img_mean"], META["img_std"]),
+])
+
+def load_model():
+    model = MultimodalSentimentModel(CONFIG).to(DEVICE)
+    ckpt  = torch.load("deployment/best_model.pt", map_location=DEVICE)
+    model.load_state_dict(ckpt["model_state"])
+    model.eval()
+    return model
+
+def predict(model, text: str, image_path: str) -> dict:
+    text = re.sub(r"http\S+", "", text)
+    text = re.sub(r"@\w+", "", text)
+    text = re.sub(r"#(\w+)", r"\1", text).strip() or "no text"
+
+    enc = tokenizer(text, max_length=CONFIG["MAX_TEXT_LEN"],
+                    padding="max_length", truncation=True, return_tensors="pt")
+    input_ids      = enc["input_ids"].to(DEVICE)
+    attention_mask = enc["attention_mask"].to(DEVICE)
+
+    img = img_transform(Image.open(image_path).convert("RGB")).unsqueeze(0).to(DEVICE)
+
+    with torch.no_grad():
+        logits = model(input_ids, attention_mask, img)
+        probs  = F.softmax(logits, dim=-1).cpu().numpy()[0]
+
+    pred_idx = probs.argmax()
+    return {
+        "label"       : META["label_names"][pred_idx],
+        "confidence"  : float(probs[pred_idx]),
+        "probabilities": {n: float(p) for n, p in zip(META["label_names"], probs)},
+    }

model_meta.json

@@ -0,0 +1,43 @@
+{
+  "config": {
+    "ZIP_PATH": "MVSA_Single.zip",
+    "EXTRACT_DIR": "./mvsa_single",
+    "OUTPUT_DIR": "./outputs",
+    "BERT_MODEL": "bert-base-uncased",
+    "MAX_TEXT_LEN": 128,
+    "IMAGE_SIZE": 224,
+    "HIDDEN_DIM": 512,
+    "DROPOUT": 0.3,
+    "NUM_CLASSES": 3,
+    "BATCH_SIZE": 16,
+    "NUM_EPOCHS": 20,
+    "LR": 2e-05,
+    "WEIGHT_DECAY": 0.0001,
+    "PATIENCE": 4,
+    "GRAD_CLIP": 1.0,
+    "TEST_SIZE": 0.15,
+    "VAL_SIZE": 0.15
+  },
+  "label_map": {
+    "negative": 0,
+    "neutral": 1,
+    "positive": 2
+  },
+  "label_names": [
+    "Negative",
+    "Neutral",
+    "Positive"
+  ],
+  "img_mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "img_std": [
+    0.229,
+    0.224,
+    0.225
+  ],
+  "test_acc": 0.6963064295485636,
+  "test_f1": 0.6968211025395353
+}