Spaces:
Runtime error
Runtime error
File size: 3,109 Bytes
a00be78 033af05 a00be78 13e0d1b 033af05 a00be78 13e0d1b 033af05 a00be78 033af05 13e0d1b a00be78 13e0d1b 44cb622 13e0d1b a00be78 44cb622 a00be78 44cb622 a00be78 13e0d1b a00be78 13e0d1b a00be78 13e0d1b a00be78 13e0d1b a00be78 13e0d1b a00be78 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 |
from gradio_huggingfacehub_search import HuggingfaceHubSearch
from huggingface_hub import HfApi
import pandas as pd
import gradio as gr
import duckdb
import requests
import llama_cpp
import instructor
from pydantic import BaseModel
BASE_DATASETS_SERVER_URL = "https://datasets-server.huggingface.co"
view_name = "dataset_view"
hf_api = HfApi()
conn = duckdb.connect()
llama = llama_cpp.Llama(
model_path="Hermes-2-Pro-Llama-3-8B-Q8_0.gguf",
n_gpu_layers=-1,
chat_format="chatml",
n_ctx=2048,
verbose=False,
)
create = instructor.patch(
create=llama.create_chat_completion_openai_v1,
mode=instructor.Mode.JSON_SCHEMA,
)
class SQLResponse(BaseModel):
sql: str
def get_dataset_ddl(dataset_id: str) -> str:
response = requests.get(f"{BASE_DATASETS_SERVER_URL}/parquet?dataset={dataset_id}")
response.raise_for_status() # Check if the request was successful
first_parquet = response.json().get("parquet_files", [])[0]
first_parquet_url = first_parquet.get("url")
if not first_parquet_url:
raise ValueError("No valid URL found for the first parquet file.")
conn.execute(
f"CREATE OR REPLACE VIEW {view_name} as SELECT * FROM read_parquet('{first_parquet_url}');"
)
dataset_ddl = conn.execute(f"PRAGMA table_info('{view_name}');").fetchall()
column_data_types = ",\n\t".join(
[f"{column[1]} {column[2]}" for column in dataset_ddl]
)
sql_ddl = """
CREATE TABLE {} (
{}
);
""".format(
view_name, column_data_types
)
return sql_ddl
def generate_sql(dataset_id: str, query: str) -> str:
ddl = get_dataset_ddl(dataset_id)
system_prompt = f"""
You are an expert SQL assistant with access to the following DuckDB Table:
```sql
{ddl}
```
Please assist the user by writing a SQL query that answers the user's question.
"""
resp: SQLResponse = create(
model="Hermes-2-Pro-Llama-3-8B",
messages=[
{"role": "system", "content": system_prompt},
{
"role": "user",
"content": query,
},
],
response_model=SQLResponse,
)
return resp.sql
def query_dataset(dataset_id: str, query: str) -> tuple[pd.DataFrame, str]:
sql_query = generate_sql(dataset_id, query)
df = conn.execute(sql_query).fetchdf()
markdown_output = f"""```sql\n{sql_query}```"""
return df, markdown_output
with gr.Blocks() as demo:
gr.Markdown("# Query your HF Datasets with Natural Language 📈📊")
dataset_id = HuggingfaceHubSearch(
label="Hub Dataset ID",
placeholder="Find your favorite dataset...",
search_type="dataset",
value="jamescalam/world-cities-geo",
)
user_query = gr.Textbox("", label="Ask anything...")
btn = gr.Button("Ask 🪄")
df = gr.DataFrame()
sql_query = gr.Markdown(label="Output SQL Query")
btn.click(
query_dataset,
inputs=[dataset_id, user_query],
outputs=[df, sql_query],
)
if __name__ == "__main__":
demo.launch()
|