inference.py

#!/usr/bin/env python3
"""
Simple inference script for the merged Phi-4-mini 128K model.
"""

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

def generate_response(prompt, max_new_tokens=500):
    model_path = "/data/phi4_merged_128k"
    
    # Load model and tokenizer
    print("Loading model...")
    model = AutoModelForCausalLM.from_pretrained(
        model_path,
        torch_dtype=torch.bfloat16,
        device_map="auto",
        trust_remote_code=True
    )
    tokenizer = AutoTokenizer.from_pretrained(model_path)
    
    # Tokenize input
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    
    # Generate
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=max_new_tokens,
            temperature=0.7,
            do_sample=True,
            top_p=0.95
        )
    
    # Decode and return
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response

if __name__ == "__main__":
    # Example usage
    prompt = "What is the capital of France?"
    response = generate_response(prompt)
    print(f"Response: {response}")