Update app.py

app.py

@@ -1,13 +1,4 @@
 import gradio as gr
-import onnxruntime as ort
-from transformers import AutoTokenizer, pipeline
-import numpy as np
-
-MODEL_FILE = "./model.onnx"
-session = ort.InferenceSession(MODEL_FILE)
-tokenizer = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-fr")
-
-labels = [
     "general", "pharma", "legal", "technical", "UI", "medicine", "it", "marketing", 
     "e-commerce", "programming", "website", "html", "keywords", "food commerce", 
     "personal development", "literature", "poetry", "physics", "chemistry", "biology", 
@@ -36,4 +27,35 @@ def translate_text(input_text):
     decoder_start_token_id = tokenizer.cls_token_id or tokenizer.pad_token_id
     decoder_input_ids = np.array([[decoder_start_token_id]], dtype=np.int64)
 
+    for _ in range(512):
+        outputs = session.run(
+            None,
+            {
+                "input_ids": input_ids,
+                "attention_mask": attention_mask,
+                "decoder_input_ids": decoder_input_ids,
+            }
+        )
+
+        logits = outputs[0]
+        next_token_id = np.argmax(logits[:, -1, :], axis=-1).item()
+        decoder_input_ids = np.concatenate(
+            [decoder_input_ids, np.array([[next_token_id]], dtype=np.int64)], axis=1
+        )
+
+        if next_token_id == tokenizer.eos_token_id:
+            break
+
+    return tokenizer.decode(decoder_input_ids[0], skip_special_tokens=True)
+
+def process_request(input_text):
+    context = detect_context(input_text)
+    translation = translate_text(input_text)
+    return translation, context
+
+gr.Interface(
+    fn=process_request,
+    inputs="text",
+    outputs="text",
+    live=True
 ).launch()