Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -6,6 +6,7 @@ from httpx import Client
|
|
6 |
from huggingface_hub import HfApi
|
7 |
from huggingface_hub.utils import logging
|
8 |
from llama_cpp import Llama
|
|
|
9 |
|
10 |
load_dotenv()
|
11 |
|
@@ -24,10 +25,13 @@ headers = {
|
|
24 |
logger = logging.get_logger(__name__)
|
25 |
client = Client(headers=headers)
|
26 |
api = HfApi(token=HF_TOKEN)
|
|
|
|
|
27 |
llama = Llama(
|
28 |
model_path="DuckDB-NSQL-7B-v0.1-q8_0.gguf",
|
29 |
n_ctx=2048,
|
30 |
)
|
|
|
31 |
|
32 |
def get_first_parquet(dataset: str):
|
33 |
resp = client.get(f"{BASE_DATASETS_SERVER_URL}/parquet?dataset={dataset}")
|
@@ -59,9 +63,9 @@ def text2sql(dataset_name, query_input):
|
|
59 |
print(first_parquet_url)
|
60 |
con = duckdb.connect()
|
61 |
con.execute("INSTALL 'httpfs'; LOAD httpfs;")
|
|
|
62 |
con.execute(f"CREATE TABLE data as SELECT * FROM '{first_parquet_url}' LIMIT 1;")
|
63 |
result = con.sql("SELECT sql FROM duckdb_tables() where table_name ='data';").df()
|
64 |
-
con.close()
|
65 |
|
66 |
ddl_create = result.iloc[0,0]
|
67 |
text = f"""### Instruction:
|
@@ -73,7 +77,7 @@ def text2sql(dataset_name, query_input):
|
|
73 |
### Question:
|
74 |
{query_input}
|
75 |
|
76 |
-
### Response (use duckdb shorthand if possible):
|
77 |
"""
|
78 |
|
79 |
print(text)
|
@@ -81,7 +85,17 @@ def text2sql(dataset_name, query_input):
|
|
81 |
# sql_output = query_remote_model(text)
|
82 |
|
83 |
sql_output = query_local_model(text)
|
84 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
85 |
|
86 |
|
87 |
with gr.Blocks() as demo:
|
@@ -92,5 +106,6 @@ with gr.Blocks() as demo:
|
|
92 |
query_input = gr.Textbox("How many rows there are?", label="Ask something about your data")
|
93 |
btn = gr.Button("Generate SQL")
|
94 |
query_output = gr.Textbox(label="Output SQL", interactive= False)
|
95 |
-
|
|
|
96 |
demo.launch()
|
|
|
6 |
from huggingface_hub import HfApi
|
7 |
from huggingface_hub.utils import logging
|
8 |
from llama_cpp import Llama
|
9 |
+
import pandas as pd
|
10 |
|
11 |
load_dotenv()
|
12 |
|
|
|
25 |
logger = logging.get_logger(__name__)
|
26 |
client = Client(headers=headers)
|
27 |
api = HfApi(token=HF_TOKEN)
|
28 |
+
|
29 |
+
print("About to load DuckDB-NSQL-7B model")
|
30 |
llama = Llama(
|
31 |
model_path="DuckDB-NSQL-7B-v0.1-q8_0.gguf",
|
32 |
n_ctx=2048,
|
33 |
)
|
34 |
+
print("DuckDB-NSQL-7B model has been loaded")
|
35 |
|
36 |
def get_first_parquet(dataset: str):
|
37 |
resp = client.get(f"{BASE_DATASETS_SERVER_URL}/parquet?dataset={dataset}")
|
|
|
63 |
print(first_parquet_url)
|
64 |
con = duckdb.connect()
|
65 |
con.execute("INSTALL 'httpfs'; LOAD httpfs;")
|
66 |
+
# could get from parquet instead?
|
67 |
con.execute(f"CREATE TABLE data as SELECT * FROM '{first_parquet_url}' LIMIT 1;")
|
68 |
result = con.sql("SELECT sql FROM duckdb_tables() where table_name ='data';").df()
|
|
|
69 |
|
70 |
ddl_create = result.iloc[0,0]
|
71 |
text = f"""### Instruction:
|
|
|
77 |
### Question:
|
78 |
{query_input}
|
79 |
|
80 |
+
### Response (use duckdb shorthand if possible) replace table name with {first_parquet_url} in the generated sql query:
|
81 |
"""
|
82 |
|
83 |
print(text)
|
|
|
85 |
# sql_output = query_remote_model(text)
|
86 |
|
87 |
sql_output = query_local_model(text)
|
88 |
+
|
89 |
+
try:
|
90 |
+
query_result = con.sql(sql_output).df()
|
91 |
+
except Exception as error:
|
92 |
+
query_result = pd.DataFrame([{"error": f"❌ Could not execute SQL query {error=}"}])
|
93 |
+
finally:
|
94 |
+
con.close()
|
95 |
+
return {
|
96 |
+
query_output:sql_output,
|
97 |
+
df:query_result
|
98 |
+
}
|
99 |
|
100 |
|
101 |
with gr.Blocks() as demo:
|
|
|
106 |
query_input = gr.Textbox("How many rows there are?", label="Ask something about your data")
|
107 |
btn = gr.Button("Generate SQL")
|
108 |
query_output = gr.Textbox(label="Output SQL", interactive= False)
|
109 |
+
df = gr.DataFrame(datatype="markdown")
|
110 |
+
btn.click(text2sql, inputs=[dataset_name, query_input], outputs=[query_output,df])
|
111 |
demo.launch()
|