app.py

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

# Load model with proper 4-bit quantization
model_id = "HelpMumHQ/MamaBot-Llama"

print("Loading MamaBot with 4-bit quantization...")

tokenizer = AutoTokenizer.from_pretrained(model_id)

# Proper 4-bit configuration
quantization_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float16
)

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    quantization_config=quantization_config,
    torch_dtype=torch.float16,
    low_cpu_mem_usage=True
)

if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

# Create pipeline with loaded model
pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer
)

print("✅ Model loaded!")

def infer(prompt):
    # Generate response
    output = pipe(
        prompt,
        max_new_tokens=200,
        do_sample=True,
        temperature=0.7,
        pad_token_id=tokenizer.pad_token_id
    )[0]['generated_text']
    return output

# Gradio interface (for testing + exposes API)
gr.Interface(
    infer,
    inputs="text",
    outputs="text",
    title="MamaBot-Llama Inference API",
    description="Enter a maternal health prompt (e.g., 'Explain NT 2.8mm at 12 weeks')."
).launch()