Create app.py

app.py

@@ -0,0 +1,40 @@
+import gradio as gr
+import torch
+import pickle
+from model import predict_with_beam_search, PosFormerImprovedWithIAC, PAD_TOKEN, SOS_TOKEN, EOS_TOKEN
+
+# --- Загрузка модели и vocab ---
+DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+with open("vocab.pkl", "rb") as f:
+    vocab = pickle.load(f)
+
+model = PosFormerImprovedWithIAC(
+    vocab=vocab,
+    vocab_size=len(vocab.itos),
+    pad_token_id=vocab.stoi[PAD_TOKEN],
+    structure_symbols_set={PAD_TOKEN, SOS_TOKEN, EOS_TOKEN, UNK_TOKEN, "{","}","^","_"},
+    id_to_token_map={i:s for s,i in vocab.stoi.items()}
+).to(DEVICE)
+
+checkpoint = torch.load("best_model.pth", map_location=DEVICE)
+model.load_state_dict(checkpoint["model_state_dict"])
+model.eval()
+
+# --- Функция для Gradio ---
+def inference(image, beam_size=5):
+    return predict_with_beam_search(model, image, vocab, beam_size, DEVICE)
+
+# --- Gradio Interface ---
+iface = gr.Interface(
+    fn=inference,
+    inputs=[
+        gr.inputs.Image(type="pil", label="Upload Image"),
+        gr.inputs.Slider(minimum=1, maximum=20, default=10, label="Beam Size")
+    ],
+    outputs=gr.outputs.Textbox(label="Predicted LaTeX"),
+    title="Handwritten Formula Recognition",
+    description="Загрузите изображение, и модель вернёт LaTeX-строку."
+)
+
+if __name__ == "__main__":
+    iface.launch()