Update app.py

app.py

@@ -1,62 +1,27 @@
 from transformers import AutoTokenizer, AutoModelForCausalLM
 import gradio as gr
 import torch
-# Load the Open-Source LLM (e.g., BLOOM or Falcon)
-model_name = "tiiuae/falcon-7b-instruct"  # Replace with your desired model
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForCausalLM.from_pretrained(model_name,
-                                             device_map="auto",  # Automatically allocates model to available devices
-    torch_dtype=torch.float16  # Use reduced precision to save memory
-                                            )
 
-# Ensure the `pad_token_id` is set explicitly to avoid warnings
-if tokenizer.pad_token_id is None:
-    tokenizer.pad_token_id = tokenizer.eos_token_id
+model_name = "tiiuae/falcon-7b-instruct"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    device_map="auto",
+    torch_dtype=torch.float16
+)
 
 def convert_to_spoken_hindi(formal_hindi_text):
-    """
-    Convert formal Hindi text to spoken Hindi using an open-source LLM.
-    """
-    # Define the prompt
-    prompt = (
-        "Convert the following formal Hindi text into conversational spoken Hindi:\n\n"
-        f"Formal Hindi: {formal_hindi_text}\n\n"
-        "Spoken Hindi:"
-    )
-
-    # Tokenize the input and create an attention mask
+    prompt = f"Convert the following formal Hindi text into conversational spoken Hindi:\n\nFormal Hindi: {formal_hindi_text}\n\nSpoken Hindi:"
     inputs = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True)
-    input_ids = inputs["input_ids"]
-    attention_mask = inputs["attention_mask"]
-
-    # Generate the response
     outputs = model.generate(
-        input_ids=input_ids,
-        attention_mask=attention_mask,
-        max_length=150,  # Adjust based on your use case
-        num_beams=5,  # Beam search for diverse generation
-        temperature=0.7,  # Sampling temperature for randomness
-        pad_token_id=tokenizer.pad_token_id  # Avoid warnings
+        inputs["input_ids"], 
+        attention_mask=inputs["attention_mask"], 
+        max_length=150, 
+        num_beams=5, 
+        temperature=0.7
     )
-
-    # Decode the generated text
     spoken_hindi = tokenizer.decode(outputs[0], skip_special_tokens=True)
-
-    # Extract the relevant output (after "Spoken Hindi:")
-    if "Spoken Hindi:" in spoken_hindi:
-        spoken_hindi = spoken_hindi.split("Spoken Hindi:")[-1].strip()
-
-    return spoken_hindi
-
-# Example Input
-formal_hindi_text = "आपका स्वास्थ्य अच्छा रहे, इस बात का ध्यान रखें। क्या आप ठीक से भोजन कर रहे हैं?"
-
-# Convert to Spoken Hindi
-#spoken_hindi_text = convert_to_spoken_hindi(formal_hindi_text)
-
-# Print the results
-print("Formal Hindi Text:", formal_hindi_text)
-#print("Spoken Hindi Text:", spoken_hindi_text)
+    return spoken_hindi.split("Spoken Hindi:")[-1].strip()
 
 iface = gr.Interface(
     fn=convert_to_spoken_hindi,
@@ -64,4 +29,5 @@ iface = gr.Interface(
     outputs="text",
     title="Hindi Text Converter"
 )
+
 iface.launch()