sql_query_generator/generator.py

from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig

def load_model(model_name="chatdb/natural-sql-7b"):
    """
    Loads the SQL generation model with 8-bit precision.
    """
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    quantization_config = BitsAndBytesConfig(
        load_in_8bit=True,  # Enable 8-bit loading
        llm_int8_threshold=6.0  # Fine-tune threshold if needed
    )
    model = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=quantization_config)
    return tokenizer, model

def generate_sql(question, prompt_inputs, tokenizer, model, device="cpu"):
    """
    Generates an SQL query based on the question and schema.
    """
    prompt = prompt_inputs["formatted_prompt"]
    inputs = tokenizer(prompt, return_tensors="pt").to(device)

    outputs = model.generate(
        **inputs,
        max_new_tokens=128,
    )

    return tokenizer.decode(outputs[0], skip_special_tokens=True)