push

app.py

@@ -1,83 +1,26 @@
-import logging
+import gradio as gr
 import torch
 from transformers import AutoTokenizer, T5ForConditionalGeneration
-import gradio as gr
 
-# ----------------------
-# Model configuration
-# ----------------------
 MODEL_NAME = "google/byt5-small"
 
-logging.basicConfig(level=logging.INFO)
-logging.info("Loading ByT5 model...")
-
-# Load tokenizer and model
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
 model = T5ForConditionalGeneration.from_pretrained(MODEL_NAME)
 model.eval()
 
-logging.info("Model loaded successfully")
-
-# ----------------------
-# Prompt engineering / fine-tuning-ready function
-# ----------------------
 def text_to_ipa(text: str) -> str:
-    """
-    Converts Scottish Gaelic text into IPA using ByT5.
-    You can customize this prompt for better results.
-    """
-    prompt = f"""
-You are a Scottish Gaelic tutor. Convert Gaelic text into IPA.
-Only respond with the IPA transcription.
-
-Text: beurla
-IPA: /bjɤːRLə/
-
-Text: faclair
-IPA: /faxgLɛrʲ/
-
-Text: {text}
-IPA:
-"""
-
-    # Tokenize input
-    inputs = tokenizer(
-        prompt,
-        return_tensors="pt",
-        truncation=True,
-        max_length=512,
-    )
-
-    # Generate IPA
+    prompt = f"Text: {text}\nIPA:"
+    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512)
     with torch.no_grad():
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=64,
-            do_sample=False,
-            num_beams=4,
-            repetition_penalty=1.1,
-            early_stopping=True,
-        )
+        outputs = model.generate(**inputs, max_new_tokens=64)
+    return tokenizer.decode(outputs[0], skip_special_tokens=True)
 
-    return tokenizer.decode(outputs[0], skip_special_tokens=True).strip()
-
-# ----------------------
-# API setup (no UI)
-# ----------------------
+# ✅ This exposes /api/predict/
 iface = gr.Interface(
     fn=text_to_ipa,
     inputs=gr.Textbox(),
     outputs=gr.Textbox(),
-    api_name="predict"  # This exposes /api/predict
+    api_name="predict"  # ← critical
 )
 
-# Enable async requests
-iface.queue()
-
-# Launch server
-iface.launch(
-    server_name="0.0.0.0",
-    server_port=7860,
-    show_error=True,
-    # share=True,  # optional: public URL for testing
-)
+iface.launch(server_name="0.0.0.0", server_port=7860, share=True)