Update app.py

app.py

@@ -10,7 +10,7 @@ access_token = os.environ.get("HUGGING_FACE_HUB_TOKEN")
 login(token=access_token)
 
 # Define model details
-peft_model_id = "kuyesu22/sunbird-ug-lang-v1.0-llama-2-7b-hf-lora"  # Update with the correct ID for your fine-tuned Llama 2 model
+peft_model_id = "kuyesu22/sunbird-ug-lang-v1.0-llama-2-7b-hf-lora"  # Your fine-tuned Llama 2 model ID
 config = PeftConfig.from_pretrained(peft_model_id)
 
 # Load base model and tokenizer
@@ -22,6 +22,10 @@ model = AutoModelForCausalLM.from_pretrained(
 )
 tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)
 
+# Set the tokenizer's padding token
+if tokenizer.pad_token is None:
+    tokenizer.pad_token = tokenizer.eos_token  # Set EOS token as padding if not already defined
+
 # Load the LoRA fine-tuned model
 model = PeftModel.from_pretrained(model, peft_model_id)
 
@@ -32,7 +36,7 @@ model.eval()
 def make_inference(english_text):
     # Format the prompt based on the language pair
     prompt = f"### English:\n{english_text}\n\n### Runyankole:"
-    batch = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True).to(model.device)
+    batch = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True, max_length=256).to(model.device)
 
     # Generate the translation
     with torch.no_grad():
@@ -56,14 +60,14 @@ def launch_gradio_interface():
     inputs = gr.components.Textbox(lines=2, label="English Text")  # Input text in English
     outputs = gr.components.Textbox(label="Translated Runyankole Text")  # Output in Runyankole
 
-    # Launch Gradio app
+    # Launch Gradio app with public sharing link enabled
     gr.Interface(
         fn=make_inference,
         inputs=inputs,
         outputs=outputs,
         title="Sunbird UG Lang Translator",
         description="Translate English to Runyankole using Llama 2 model fine-tuned with LoRA.",
-    ).launch()
+    ).launch(share=True)  # Set `share=True` to create a public link
 
 # Entry point to run the Gradio app
 if __name__ == "__main__":