Create app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# Use a smaller instruct-tuned model that runs on Hugging Face Spaces
model_name = "tiiuae/falcon-7b-instruct" # Falcon-7B is lighter than Mistral

# Load model and tokenizer with optimizations
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype=torch.float16,
device_map="auto" # Uses available GPU/CPU
)

# AI Response Function
def nithin_ai(question):
inputs = tokenizer(question, return_tensors="pt").input_ids.to(model.device)
outputs = model.generate(inputs, max_length=200)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
return response

# Gradio Chat Interface
iface = gr.Interface(
fn=nithin_ai,
inputs="text",
outputs="text",
title="Nithin AI - Student Doubt Solver",
description="Ask any question related to robotics, science, or math!"
)

iface.launch()

app.py

@@ -2,18 +2,20 @@ import gradio as gr
 from transformers import AutoModelForCausalLM, AutoTokenizer
 import torch
 
-# Load a smaller model that works on Hugging Face free tier
-model_name = "tiiuae/falcon-7b-instruct"  # Use instruct-tuned model
+# Use a smaller instruct-tuned model that runs on Hugging Face Spaces
+model_name = "tiiuae/falcon-7b-instruct"  # Falcon-7B is lighter than Mistral
 
+# Load model and tokenizer with optimizations
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 model = AutoModelForCausalLM.from_pretrained(
     model_name, 
     torch_dtype=torch.float16,  
-    device_map="cpu"  # Change to "auto" if using a GPU
+    device_map="auto"  # Uses available GPU/CPU
 )
 
+# AI Response Function
 def nithin_ai(question):
-    inputs = tokenizer(question, return_tensors="pt").input_ids
+    inputs = tokenizer(question, return_tensors="pt").input_ids.to(model.device)
     outputs = model.generate(inputs, max_length=200)
     response = tokenizer.decode(outputs[0], skip_special_tokens=True)
     return response