forked from defog-ai/sqlcoder
-
Notifications
You must be signed in to change notification settings - Fork 0
/
inference.py
67 lines (59 loc) · 2.39 KB
/
inference.py
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
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import argparse
def generate_prompt(question, prompt_file="prompt.md", metadata_file="metadata.sql"):
with open(prompt_file, "r") as f:
prompt = f.read()
with open(metadata_file, "r") as f:
table_metadata_string = f.read()
prompt = prompt.format(
user_question=question, table_metadata_string=table_metadata_string
)
return prompt
def get_tokenizer_model(model_name):
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
trust_remote_code=True,
torch_dtype=torch.float16,
device_map="auto",
use_cache=True,
)
return tokenizer, model
def run_inference(question, prompt_file="prompt.md", metadata_file="metadata.sql"):
tokenizer, model = get_tokenizer_model("defog/sqlcoder-7b-2")
prompt = generate_prompt(question, prompt_file, metadata_file)
# make sure the model stops generating at triple ticks
# eos_token_id = tokenizer.convert_tokens_to_ids(["```"])[0]
eos_token_id = tokenizer.eos_token_id
pipe = pipeline(
"text-generation",
model=model,
tokenizer=tokenizer,
max_new_tokens=300,
do_sample=False,
return_full_text=False, # added return_full_text parameter to prevent splitting issues with prompt
num_beams=5, # do beam search with 5 beams for high quality results
)
generated_query = (
pipe(
prompt,
num_return_sequences=1,
eos_token_id=eos_token_id,
pad_token_id=eos_token_id,
)[0]["generated_text"]
.split(";")[0]
.split("```")[0]
.strip()
+ ";"
)
return generated_query
if __name__ == "__main__":
# Parse arguments
_default_question="Do we get more sales from customers in New York compared to customers in San Francisco? Give me the total sales for each city, and the difference between the two."
parser = argparse.ArgumentParser(description="Run inference on a question")
parser.add_argument("-q","--question", type=str, default=_default_question, help="Question to run inference on")
args = parser.parse_args()
question = args.question
print("Loading a model and generating a SQL query for answering your question...")
print(run_inference(question))