Spaces:
Paused
Paused
Commit
·
6670a17
1
Parent(s):
b053c71
update model
Browse files- app.py +0 -54
- text2sql.py +2 -2
app.py
CHANGED
@@ -76,60 +76,6 @@ from utils.db_utils import add_a_record
|
|
76 |
from langdetect.lang_detect_exception import LangDetectException
|
77 |
import os
|
78 |
|
79 |
-
# Suppress excessive warnings from Hugging Face transformers library
|
80 |
-
hf_logging.set_verbosity_error()
|
81 |
-
|
82 |
-
# SchemaItemClassifierInference class for loading the Hugging Face model
|
83 |
-
class SchemaItemClassifierInference:
|
84 |
-
def __init__(self, model_name: str, token=None):
|
85 |
-
"""
|
86 |
-
model_name: Hugging Face repository path, e.g., "Roxanne-WANG/LangSQL"
|
87 |
-
token: Authentication token for Hugging Face (if the model is private)
|
88 |
-
"""
|
89 |
-
# Load the tokenizer and model from Hugging Face, trust remote code if needed
|
90 |
-
self.tokenizer = AutoTokenizer.from_pretrained(
|
91 |
-
model_name,
|
92 |
-
use_auth_token=token, # Pass the token for accessing private models
|
93 |
-
trust_remote_code=True # Trust custom model code from Hugging Face repo
|
94 |
-
)
|
95 |
-
self.model = AutoModelForSequenceClassification.from_pretrained(
|
96 |
-
model_name,
|
97 |
-
use_auth_token=token,
|
98 |
-
trust_remote_code=True
|
99 |
-
)
|
100 |
-
|
101 |
-
def predict(self, text: str):
|
102 |
-
# Tokenize the input text and get predictions from the model
|
103 |
-
inputs = self.tokenizer(
|
104 |
-
text,
|
105 |
-
return_tensors="pt",
|
106 |
-
padding=True,
|
107 |
-
truncation=True
|
108 |
-
)
|
109 |
-
outputs = self.model(**inputs)
|
110 |
-
return outputs.logits
|
111 |
-
|
112 |
-
|
113 |
-
# ChatBot class that interacts with SchemaItemClassifierInference
|
114 |
-
class ChatBot:
|
115 |
-
def __init__(self):
|
116 |
-
# Specify the Hugging Face model name (replace with your model's path)
|
117 |
-
model_name = "Roxanne-WANG/LangSQL"
|
118 |
-
hf_token = os.getenv('HF_TOKEN') # Get token from environment variables
|
119 |
-
|
120 |
-
if hf_token is None:
|
121 |
-
raise ValueError("Hugging Face token is required. Please set HF_TOKEN.")
|
122 |
-
|
123 |
-
# Initialize the schema item classifier with Hugging Face token
|
124 |
-
self.sic = SchemaItemClassifierInference(model_name, token=hf_token)
|
125 |
-
|
126 |
-
def get_response(self, question: str, db_id: str):
|
127 |
-
# Get the model's prediction (logits) for the input question
|
128 |
-
logits = self.sic.predict(question)
|
129 |
-
# For now, return logits as a placeholder for the actual SQL query
|
130 |
-
return logits
|
131 |
-
|
132 |
-
|
133 |
# -------- Streamlit Web Application --------
|
134 |
text2sql_bot = ChatBot()
|
135 |
baidu_api_token = None # Your Baidu API token (if needed for translation)
|
|
|
76 |
from langdetect.lang_detect_exception import LangDetectException
|
77 |
import os
|
78 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
79 |
# -------- Streamlit Web Application --------
|
80 |
text2sql_bot = ChatBot()
|
81 |
baidu_api_token = None # Your Baidu API token (if needed for translation)
|
text2sql.py
CHANGED
@@ -104,14 +104,14 @@ def get_db_id2ddl(db_path):
|
|
104 |
class ChatBot():
|
105 |
def __init__(self) -> None:
|
106 |
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
|
107 |
-
model_name = "seeklhy/codes-
|
108 |
self.tokenizer = AutoTokenizer.from_pretrained(model_name)
|
109 |
self.model = AutoModelForCausalLM.from_pretrained(model_name, device_map = "auto", torch_dtype = torch.float16)
|
110 |
self.max_length = 4096
|
111 |
self.max_new_tokens = 256
|
112 |
self.max_prefix_length = self.max_length - self.max_new_tokens
|
113 |
|
114 |
-
self.sic = SchemaItemClassifierInference("
|
115 |
|
116 |
self.db_id2content_searcher = dict()
|
117 |
for db_id in os.listdir("db_contents_index"):
|
|
|
104 |
class ChatBot():
|
105 |
def __init__(self) -> None:
|
106 |
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
|
107 |
+
model_name = "seeklhy/codes-1b"
|
108 |
self.tokenizer = AutoTokenizer.from_pretrained(model_name)
|
109 |
self.model = AutoModelForCausalLM.from_pretrained(model_name, device_map = "auto", torch_dtype = torch.float16)
|
110 |
self.max_length = 4096
|
111 |
self.max_new_tokens = 256
|
112 |
self.max_prefix_length = self.max_length - self.max_new_tokens
|
113 |
|
114 |
+
self.sic = SchemaItemClassifierInference("Roxanne-WANG/LangSQL", token=os.getenv('HF_TOKEN'))
|
115 |
|
116 |
self.db_id2content_searcher = dict()
|
117 |
for db_id in os.listdir("db_contents_index"):
|