app.py

import os
from threading import Event, Thread
from transformers import (
    AutoModelForCausalLM,
    AutoTokenizer,
    StoppingCriteria,
    StoppingCriteriaList,
    TextIteratorStreamer,
)
import gradio as gr
import torch

model_name = os.getenv("HF_MODEL_NAME", None)
tok = AutoTokenizer.from_pretrained(model_name)

max_new_tokens = 1024

print(f"Starting to load the model {model_name}")

m = AutoModelForCausalLM.from_pretrained(
    model_name,
    device_map=0,
    #load_in_8bit=True,
)

m.config.pad_token_id = m.config.eos_token_id
m.generation_config.pad_token_id = m.config.eos_token_id

stop_tokens = [";", "###", "Result"]
stop_token_ids = tok.convert_tokens_to_ids(stop_tokens)

print(f"Successfully loaded the model {model_name} into memory")

class StopOnTokens(StoppingCriteria):
    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
        for stop_id in stop_token_ids:
            if input_ids[0][-1] == stop_id:
                return True
        return False

def bot(input_message: str, db_info="", temperature=0.1, top_p=0.9, top_k=0, repetition_penalty=1.08):
    stop = StopOnTokens()

    # Format the user's input message
    messages = f"Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\n\n### Instruction:\n\nConvert text to sql: {input_message} {db_info}\n\n### Response:\n\n"

    input_ids = tok(messages, return_tensors="pt").input_ids
    input_ids = input_ids.to(m.device)
    streamer = TextIteratorStreamer(tok, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        input_ids=input_ids,
        max_new_tokens=max_new_tokens,
        temperature=temperature,
        do_sample=temperature > 0.0,
        top_p=top_p,
        top_k=top_k,
        repetition_penalty=repetition_penalty,
        streamer=streamer,
        stopping_criteria=StoppingCriteriaList([stop]),
    )

    stream_complete = Event()

    def generate_and_signal_complete():
        m.generate(**generate_kwargs)
        stream_complete.set()

    t1 = Thread(target=generate_and_signal_complete)
    t1.start()

    partial_text = ""
    for new_text in streamer:
        partial_text += new_text

    # Split the text by "|", and get the last element in the list which should be the final query
    final_query = partial_text.split("|")[1].strip()
    # Convert SQL to markdown (not required, but just to show how to use the markdown module)
    final_query_markdown = f"```sql\n{final_query}\n```"
    return final_query_markdown

with gr.Blocks() as demo:
    input_text = gr.Textbox(lines=5, placeholder='Input text here...', label='Input Text')
    db_info = gr.Textbox(lines=5, placeholder='Example: | table_01 : column_01 , column_02 | table_02 : column_01 , column_02 | ...', label='Database Info')
    temperature = gr.Slider(label="Temperature", minimum=0.0, maximum=1.0, value=0.1, step=0.1)
    top_p = gr.Slider(label="Top-p (nucleus sampling)", minimum=0.0, maximum=1.0, value=0.9, step=0.01)
    top_k = gr.Slider(label="Top-k", minimum=0, maximum=200, value=0, step=1)
    repetition_penalty = gr.Slider(label="Repetition Penalty", minimum=1.0, maximum=2.0, value=1.08, step=0.1)
    output_box = gr.Markdown(label="Generated SQL Query")
    run_button = gr.Button("Generate SQL Query")
    run_button.click(fn=bot, inputs=[input_text, db_info, temperature, top_p, top_k, repetition_penalty], outputs=output_box)

demo.launch()