Update README.md
Browse files
README.md
CHANGED
@@ -61,21 +61,22 @@ Can be fine-tuned further for specific databases or Arabic dialect adaptations.
|
|
61 |
- Ensure compatibility with specific database schemas.
|
62 |
|
63 |
## How to Get Started with the Model
|
64 |
-
|
65 |
```python
|
66 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
67 |
import torch
|
|
|
68 |
|
69 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
70 |
base_model_id = "Qwen/Qwen2.5-1.5B-Instruct"
|
71 |
finetuned_model_id = "OsamaMo/Arabic_Text-To-SQL_using_Qwen2.5-1.5B"
|
72 |
|
|
|
73 |
model = AutoModelForCausalLM.from_pretrained(
|
74 |
base_model_id,
|
75 |
device_map="auto",
|
76 |
torch_dtype=torch.bfloat16
|
77 |
)
|
78 |
-
|
79 |
model.load_adapter(finetuned_model_id)
|
80 |
|
81 |
tokenizer = AutoTokenizer.from_pretrained(base_model_id)
|
@@ -86,25 +87,150 @@ def generate_resp(messages):
|
|
86 |
tokenize=False,
|
87 |
add_generation_prompt=True
|
88 |
)
|
89 |
-
|
90 |
model_inputs = tokenizer([text], return_tensors="pt").to(device)
|
91 |
-
|
92 |
generated_ids = model.generate(
|
93 |
model_inputs.input_ids,
|
94 |
max_new_tokens=1024,
|
95 |
-
do_sample=False,
|
96 |
)
|
97 |
-
|
98 |
generated_ids = [
|
99 |
output_ids[len(input_ids):]
|
100 |
for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
101 |
]
|
102 |
-
|
103 |
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
104 |
-
|
105 |
return response
|
106 |
```
|
107 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
108 |
## Training Details
|
109 |
|
110 |
### Training Data
|
|
|
61 |
- Ensure compatibility with specific database schemas.
|
62 |
|
63 |
## How to Get Started with the Model
|
64 |
+
### Load Model
|
65 |
```python
|
66 |
from transformers import AutoModelForCausalLM, AutoTokenizer
|
67 |
import torch
|
68 |
+
import re
|
69 |
|
70 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
71 |
base_model_id = "Qwen/Qwen2.5-1.5B-Instruct"
|
72 |
finetuned_model_id = "OsamaMo/Arabic_Text-To-SQL_using_Qwen2.5-1.5B"
|
73 |
|
74 |
+
# Load the base model and adapter for fine-tuning
|
75 |
model = AutoModelForCausalLM.from_pretrained(
|
76 |
base_model_id,
|
77 |
device_map="auto",
|
78 |
torch_dtype=torch.bfloat16
|
79 |
)
|
|
|
80 |
model.load_adapter(finetuned_model_id)
|
81 |
|
82 |
tokenizer = AutoTokenizer.from_pretrained(base_model_id)
|
|
|
87 |
tokenize=False,
|
88 |
add_generation_prompt=True
|
89 |
)
|
|
|
90 |
model_inputs = tokenizer([text], return_tensors="pt").to(device)
|
|
|
91 |
generated_ids = model.generate(
|
92 |
model_inputs.input_ids,
|
93 |
max_new_tokens=1024,
|
94 |
+
do_sample=False, temperature= False,
|
95 |
)
|
|
|
96 |
generated_ids = [
|
97 |
output_ids[len(input_ids):]
|
98 |
for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
|
99 |
]
|
|
|
100 |
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
|
|
101 |
return response
|
102 |
```
|
103 |
|
104 |
+
|
105 |
+
|
106 |
+
### Example Usage
|
107 |
+
```python
|
108 |
+
|
109 |
+
# Production-ready system message for SQL generation
|
110 |
+
system_message = (
|
111 |
+
"You are a highly advanced Arabic text-to-SQL converter. Your mission is to Understand first the db schema and reltions between it and then accurately transform Arabic "
|
112 |
+
"natural language queries into SQL queries with precision and clarity.\n"
|
113 |
+
)
|
114 |
+
|
115 |
+
def get_sql_query(db_schema, arabic_query):
|
116 |
+
# Construct the instruction message including the DB schema and the Arabic query
|
117 |
+
instruction_message = "\n".join([
|
118 |
+
"## DB-Schema:",
|
119 |
+
db_schema,
|
120 |
+
"",
|
121 |
+
"## User-Prompt:",
|
122 |
+
arabic_query,
|
123 |
+
"# Output SQL:",
|
124 |
+
"```SQL"
|
125 |
+
])
|
126 |
+
|
127 |
+
messages = [
|
128 |
+
{"role": "system", "content": system_message},
|
129 |
+
{"role": "user", "content": instruction_message}
|
130 |
+
]
|
131 |
+
|
132 |
+
response = generate_resp(messages)
|
133 |
+
|
134 |
+
# Extract the SQL query from the response using a regex to capture text within the ```sql markdown block
|
135 |
+
match = re.search(r"```sql\s*(.*?)\s*```", response, re.DOTALL | re.IGNORECASE)
|
136 |
+
if match:
|
137 |
+
sql_query = match.group(1).strip()
|
138 |
+
return sql_query
|
139 |
+
else:
|
140 |
+
return response.strip()
|
141 |
+
|
142 |
+
# Example usage:
|
143 |
+
example_db_schema = r'''{
|
144 |
+
'Pharmcy':
|
145 |
+
CREATE TABLE `purchase` (
|
146 |
+
`BARCODE` varchar(20) NOT NULL,
|
147 |
+
`NAME` varchar(50) NOT NULL,
|
148 |
+
`TYPE` varchar(20) NOT NULL,
|
149 |
+
`COMPANY_NAME` varchar(20) NOT NULL,
|
150 |
+
`QUANTITY` int NOT NULL,
|
151 |
+
`PRICE` double NOT NULL,
|
152 |
+
`AMOUNT` double NOT NULL,
|
153 |
+
PRIMARY KEY (`BARCODE`),
|
154 |
+
KEY `fkr3` (`COMPANY_NAME`),
|
155 |
+
CONSTRAINT `fkr3` FOREIGN KEY (`COMPANY_NAME`) REFERENCES `company` (`NAME`) ON DELETE CASCADE ON UPDATE CASCADE
|
156 |
+
) ENGINE=InnoDB DEFAULT CHARSET=latin1
|
157 |
+
|
158 |
+
CREATE TABLE `sales` (
|
159 |
+
`BARCODE` varchar(20) NOT NULL,
|
160 |
+
`NAME` varchar(50) NOT NULL,
|
161 |
+
`TYPE` varchar(10) NOT NULL,
|
162 |
+
`DOSE` varchar(10) NOT NULL,
|
163 |
+
`QUANTITY` int NOT NULL,
|
164 |
+
`PRICE` double NOT NULL,
|
165 |
+
`AMOUNT` double NOT NULL,
|
166 |
+
`DATE` varchar(15) NOT NULL
|
167 |
+
) ENGINE=InnoDB DEFAULT CHARSET=latin1
|
168 |
+
|
169 |
+
CREATE TABLE `users` (
|
170 |
+
`ID` int NOT NULL,
|
171 |
+
`NAME` varchar(50) NOT NULL,
|
172 |
+
`DOB` varchar(20) NOT NULL,
|
173 |
+
`ADDRESS` varchar(100) NOT NULL,
|
174 |
+
`PHONE` varchar(20) NOT NULL,
|
175 |
+
`SALARY` double NOT NULL,
|
176 |
+
`PASSWORD` varchar(20) NOT NULL,
|
177 |
+
PRIMARY KEY (`ID`)
|
178 |
+
) ENGINE=InnoDB DEFAULT CHARSET=latin1
|
179 |
+
|
180 |
+
CREATE TABLE `history_sales` (
|
181 |
+
`USER_NAME` varchar(20) NOT NULL,
|
182 |
+
`BARCODE` varchar(20) NOT NULL,
|
183 |
+
`NAME` varchar(50) NOT NULL,
|
184 |
+
`TYPE` varchar(10) NOT NULL,
|
185 |
+
`DOSE` varchar(10) NOT NULL,
|
186 |
+
`QUANTITY` int NOT NULL,
|
187 |
+
`PRICE` double NOT NULL,
|
188 |
+
`AMOUNT` double NOT NULL,
|
189 |
+
`DATE` varchar(15) NOT NULL,
|
190 |
+
`TIME` varchar(20) NOT NULL
|
191 |
+
) ENGINE=InnoDB DEFAULT CHARSET=latin1
|
192 |
+
|
193 |
+
CREATE TABLE `expiry` (
|
194 |
+
`PRODUCT_NAME` varchar(50) NOT NULL,
|
195 |
+
`PRODUCT_CODE` varchar(20) NOT NULL,
|
196 |
+
`DATE_OF_EXPIRY` varchar(10) NOT NULL,
|
197 |
+
`QUANTITY_REMAIN` int NOT NULL
|
198 |
+
) ENGINE=InnoDB DEFAULT CHARSET=latin1
|
199 |
+
|
200 |
+
CREATE TABLE `drugs` (
|
201 |
+
`NAME` varchar(50) NOT NULL,
|
202 |
+
`TYPE` varchar(20) NOT NULL,
|
203 |
+
`BARCODE` varchar(20) NOT NULL,
|
204 |
+
`DOSE` varchar(10) NOT NULL,
|
205 |
+
`CODE` varchar(10) NOT NULL,
|
206 |
+
`COST_PRICE` double NOT NULL,
|
207 |
+
`SELLING_PRICE` double NOT NULL,
|
208 |
+
`EXPIRY` varchar(20) NOT NULL,
|
209 |
+
`COMPANY_NAME` varchar(50) NOT NULL,
|
210 |
+
`PRODUCTION_DATE` date NOT NULL,
|
211 |
+
`EXPIRATION_DATE` date NOT NULL,
|
212 |
+
`PLACE` varchar(20) NOT NULL,
|
213 |
+
`QUANTITY` int NOT NULL,
|
214 |
+
PRIMARY KEY (`BARCODE`)
|
215 |
+
) ENGINE=InnoDB DEFAULT CHARSET=latin1
|
216 |
+
|
217 |
+
CREATE TABLE `company` (
|
218 |
+
`NAME` varchar(50) NOT NULL,
|
219 |
+
`ADDRESS` varchar(50) NOT NULL,
|
220 |
+
`PHONE` varchar(20) NOT NULL,
|
221 |
+
PRIMARY KEY (`NAME`)
|
222 |
+
) ENGINE=InnoDB DEFAULT CHARSET=latin1
|
223 |
+
|
224 |
+
Answer the following questions about this schema:
|
225 |
+
}'''
|
226 |
+
|
227 |
+
example_arabic_query = "اريد الباركود الخاص بدواء يبداء اسمه بحرف 's'"
|
228 |
+
|
229 |
+
sql_result = get_sql_query(example_db_schema, example_arabic_query)
|
230 |
+
print("استعلام SQL الناتج:")
|
231 |
+
print(sql_result)
|
232 |
+
```
|
233 |
+
|
234 |
## Training Details
|
235 |
|
236 |
### Training Data
|