inference_example.py

"""
Example inference script for the fine-tuned Reading Steiner model.
Replace with your actual LoRA adapter path after training.
"""
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen3.5-2B",
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True,
)

# Load LoRA adapter (replace with your actual adapter path)
# model = PeftModel.from_pretrained(base_model, "OmAlve/reading-steiner-qwen3.5-2b")
# Optionally merge adapter into base model for faster inference:
# model = model.merge_and_unload()
model = base_model  # Replace with above after training

# Tokenizer
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3.5-2B", trust_remote_code=True)

# Example input
messages = [
    {"role": "system", "content": (
        "You are Reading Steiner, a web content extraction model. "
        "Given a webpage split into indexed blocks, identify which blocks contain the main content. "
        "Output indices as a Python list of [start, end] intervals."
    )},
    {"role": "user", "content": (
        "URL: https://example.com\n"
        "Title: Example Page\n"
        "Blocks:\n"
        '[1] <div class="nav">Home | About | Contact</div>\n'
        '[2] <div class="sidebar">Trending</div>\n'
        '[3] <p>This is the main article content.</p>\n'
        '[4] <p>More content here.</p>\n'
        '[5] <div class="footer">Copyright 2025</div>'
    )},
]

inputs = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    return_tensors="pt",
    add_generation_prompt=True,
)
inputs = inputs.to(model.device)

# Generate
with torch.no_grad():
    outputs = model.generate(
        inputs,
        max_new_tokens=128,
        temperature=0.1,
        do_sample=False,
        pad_token_id=tokenizer.pad_token_id,
    )

result = tokenizer.decode(outputs[0], skip_special_tokens=True)
print("=== Generated Output ===")
print(result)