app.py

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

model_id = "google/gemma-3-1b-it"

tokenizer = AutoTokenizer.from_pretrained(model_id)

model = AutoModelForCausalLM.from_pretrained(
model_id,
torch_dtype=torch.float32,
low_cpu_mem_usage=True,
device_map=“cpu”
)

model.eval()

def predict(message, history):
messages = []
for turn in history:
messages.append({“role”: “user”, “content”: turn[0]})
messages.append({“role”: “assistant”, “content”: turn[1]})
messages.append({“role”: “user”, “content”: message[-1000:]})

```
tokenized = tokenizer.apply_chat_template(
    messages,
    return_tensors="pt",
    add_generation_prompt=True
)
input_ids = tokenized.to("cpu")

with torch.no_grad():
    output = model.generate(
        input_ids=input_ids,
        max_new_tokens=256,
        do_sample=True,
        temperature=0.7,
        top_p=0.9,
        use_cache=True
    )

new_tokens = output[0][input_ids.shape[-1]:]
return tokenizer.decode(new_tokens, skip_special_tokens=True).strip()
```

demo = gr.ChatInterface(
fn=predict,
title=“Gemma 3 1B (CPU)”,
description=“google/gemma-3-1b-it — runs on HF free tier CPU (~4GB RAM)”
)

if **name** == “**main**”:
demo.launch()