File size: 8,389 Bytes
1e3f569 dd9d480 1e3f569 dd9d480 1e3f569 eaeb469 1e3f569 eaeb469 1e3f569 83c6516 1e3f569 dd9d480 1e3f569 dd9d480 83c6516 1e3f569 ef53845 1e3f569 dd9d480 1e3f569 e4fb671 1e3f569 83c6516 1e3f569 dd9d480 1e3f569 dd9d480 1e3f569 dd9d480 1e3f569 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 |
import os
import gradio as gr
import sqlparse
import requests
from time import sleep
import re
def format(text):
# Split the text by "|", and get the last element in the list which should be the final query
try:
final_query = text.split("|")[1].strip()
except Exception:
final_query = text
try:
# Attempt to format SQL query using sqlparse
formatted_query = sqlparse.format(final_query, reindent=True, keyword_case='upper')
except Exception:
# If formatting fails, use the original, unformatted query
formatted_query = final_query
# Convert SQL to markdown (not required, but just to show how to use the markdown module)
final_query_markdown = f"{formatted_query}"
return final_query_markdown
def generate(input_message: str, db_info="", temperature=0.3, top_p=0.9, top_k=0, repetition_penalty=1.08):
# 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"
url = "https://e9f4be879d38-8269039109365193683.ngrok-free.app/api/v1/generate"
payload = {
"prompt": messages,
"temperature": temperature,
"top_p": top_p,
"top_k": top_k,
"top_a": 0,
"n": 1,
"max_context_length": 2048,
"max_length": 512,
"rep_pen": repetition_penalty,
"sampler_order": [6,0,1,3,4,2,5],
"stop_sequence": ["###", "Result"],
}
headers = {
"Content-Type": "application/json",
"ngrok-skip-browser-warning": "1" # added this line
}
for _ in range(3): # Try 3 times
try:
response = requests.post(url, json=payload, headers=headers)
response_text = response.json()["results"][0]["text"]
response_text = response_text.replace("\n", "").replace("\t", " ")
if response_text and response_text[-1] == ".":
response_text = response_text[:-1]
return format(response_text)
except Exception as e:
print(f'Error occurred: {str(e)}')
print('Waiting for 10 seconds before retrying...')
sleep(10)
# Gradio UI Code
with gr.Blocks(theme='gradio/soft') as demo:
header = gr.HTML("""
<h1 style="text-align: center">SQL Skeleton WizardCoder Demo</h1>
<h3 style="text-align: center">🕷️☠️🧙♂️ Generate SQL queries from Natural Language 🕷️☠️🧙♂️</h3>
<p>Note: Should take around 30-60s to generate.</p>
""")
output_box = gr.Code(label="Generated SQL", lines=2, interactive=True)
input_text = gr.Textbox(lines=3, placeholder='Write your question here...', label='NL Input')
db_info = gr.Textbox(lines=4, placeholder='Example: | table_01 : column_01 , column_02 | table_02 : column_01 , column_02 | ...', label='Database Info')
with gr.Accordion("Hyperparameters", open=False):
temperature = gr.Slider(label="Temperature", minimum=0.0, maximum=1.0, value=0.3, 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=0.01)
run_button = gr.Button("Generate SQL", variant="primary")
run_button.click(fn=generate, inputs=[input_text, db_info, temperature, top_p, top_k, repetition_penalty], outputs=output_box, api_name="txt2sql")
with gr.Accordion("Examples", open=True):
examples = gr.Examples([
["What is the average, minimum, and maximum age of all singers from France?", "| stadium : stadium_id , location , name , capacity , highest , lowest , average | singer : singer_id , name , country , song_name , song_release_year , age , is_male | concert : concert_id , concert_name , theme , stadium_id , year | singer_in_concert : concert_id , singer_id | concert.stadium_id = stadium.stadium_id | singer_in_concert.singer_id = singer.singer_id | singer_in_concert.concert_id = concert.concert_id |"],
["Show location and name for all stadiums with a capacity between 5000 and 10000.", "| stadium : stadium_id , location , name , capacity , highest , lowest , average | singer : singer_id , name , country , song_name , song_release_year , age , is_male | concert : concert_id , concert_name , theme , stadium_id , year | singer_in_concert : concert_id , singer_id | concert.stadium_id = stadium.stadium_id | singer_in_concert.singer_id = singer.singer_id | singer_in_concert.concert_id = concert.concert_id |"],
["What are the number of concerts that occurred in the stadium with the largest capacity ?", "| stadium : stadium_id , location , name , capacity , highest , lowest , average | singer : singer_id , name , country , song_name , song_release_year , age , is_male | concert : concert_id , concert_name , theme , stadium_id , year | singer_in_concert : concert_id , singer_id | concert.stadium_id = stadium.stadium_id | singer_in_concert.singer_id = singer.singer_id | singer_in_concert.concert_id = concert.concert_id |"],
["How many male singers performed in concerts in the year 2023?", "| stadium : stadium_id , location , name , capacity , highest , lowest , average | singer : singer_id , name , country , song_name , song_release_year , age , is_male | concert : concert_id , concert_name , theme , stadium_id , year | singer_in_concert : concert_id , singer_id | concert.stadium_id = stadium.stadium_id | singer_in_concert.singer_id = singer.singer_id | singer_in_concert.concert_id = concert.concert_id |"],
["List the names of all singers who performed in a concert with the theme 'Rock'", "| stadium : stadium_id , location , name , capacity , highest , lowest , average | singer : singer_id , name , country , song_name , song_release_year , age , is_male | concert : concert_id , concert_name , theme , stadium_id , year | singer_in_concert : concert_id , singer_id | concert.stadium_id = stadium.stadium_id | singer_in_concert.singer_id = singer.singer_id | singer_in_concert.concert_id = concert.concert_id |"]
], inputs=[input_text, db_info, temperature, top_p, top_k, repetition_penalty], fn=generate, cache_examples=False if os.name == 'posix' or os.name == 'nt' else True, outputs=output_box)
quantized_model = "richardr1126/spider-skeleton-wizard-coder-ggml"
merged_model = "richardr1126/spider-skeleton-wizard-coder-merged"
initial_model = "WizardLM/WizardCoder-15B-V1.0"
lora_model = "richardr1126/spider-skeleton-wizard-coder-qlora"
dataset = "richardr1126/spider-skeleton-context-instruct"
footer = gr.HTML(f"""
<p>🛠️ If you want you can <strong>duplicate this Space</strong>, then change the HF_MODEL_REPO spaces env varaible to use any GGML model.</p>
<p>🌐 Leveraging the <a href='https://huggingface.co/{quantized_model}'><strong>4-bit GGML version</strong></a> of <a href='https://huggingface.co/{merged_model}'><strong>{merged_model}</strong></a> model.</p>
<p>🔗 How it's made: <a href='https://huggingface.co/{initial_model}'><strong>{initial_model}</strong></a> was finetuned to create <a href='https://huggingface.co/{lora_model}'><strong>{lora_model}</strong></a>, then merged together to create <a href='https://huggingface.co/{merged_model}'><strong>{merged_model}</strong></a>.</p>
<p>📉 Fine-tuning was performed using QLoRA techniques on the <a href='https://huggingface.co/datasets/{dataset}'><strong>{dataset}</strong></a> dataset. You can view training metrics on the <a href='https://huggingface.co/{lora_model}'><strong>QLoRa adapter HF Repo</strong></a>.</p>
""")
readme_content = requests.get(f"https://huggingface.co/{merged_model}/raw/main/README.md").text
readme_content = re.sub('---.*?---', '', readme_content, flags=re.DOTALL) #Remove YAML front matter
with gr.Accordion("📖 Model Readme", open=True):
readme = gr.Markdown(
readme_content,
)
demo.queue(concurrency_count=1, max_size=10).launch(debug=True) |