app.py

import os
import torch
import gradio as gr
from huggingface_hub import login
from transformers import AutoTokenizer, AutoModelForCausalLM
# === STEP 1: Authenticate with Hugging Face ===
# Make sure you set your HF token as an environment variable or paste it here temporarily
# For security, prefer environment variable (recommended)
#hf_token = "TLpIICgZJrDCTgVTsaaydFFWbWyGKiGAPa"
#login(token="hf_" + hf_token)

# === STEP 2: Load base and adapter models ===
base_model = "meta-llama/Llama-2-7b-chat-hf"
adapter_model = "olacode55/zimble-llama2-finetunedhybride"

tokenizer = AutoTokenizer.from_pretrained(adapter_model)

# Enable memory-efficient loading if needed
device = "cuda" if torch.cuda.is_available() else "cpu"

model = AutoModelForCausalLM.from_pretrained(
    adapter_model,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    device_map="auto",
    low_cpu_mem_usage=True
)

# === STEP 3: Define generation function ===
def generate(prompt):
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=250,
            temperature=0.7,
            top_p=0.9,
            do_sample=True,
        )
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# === STEP 4: Launch Gradio app ===
demo = gr.Interface(
    fn=generate,
    inputs=gr.Textbox(label="Enter your prompt", lines=4, placeholder="Type something..."),
    outputs=gr.Textbox(label="Model output"),
    title="🦙 Zimble LLaMA 2 (Merged)",
    description="Fine-tuned and merged version of LLaMA 2 running on Hugging Face Space"
)

demo.launch()