Add support for colab
Browse files- test_pretrained.ipynb +52 -135
test_pretrained.ipynb
CHANGED
@@ -7,6 +7,56 @@
|
|
7 |
"# Run pre-trained DeepSeek Coder 1.3B Model on Chat-GPT 4o generated dataset"
|
8 |
]
|
9 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
{
|
11 |
"cell_type": "markdown",
|
12 |
"metadata": {},
|
@@ -33,10 +83,6 @@
|
|
33 |
}
|
34 |
],
|
35 |
"source": [
|
36 |
-
"import pandas as pd \n",
|
37 |
-
"import warnings\n",
|
38 |
-
"warnings.filterwarnings(\"ignore\")\n",
|
39 |
-
"\n",
|
40 |
"# Load dataset and check length\n",
|
41 |
"df = pd.read_csv(\"./train-data/sql_train.tsv\", sep='\\t')\n",
|
42 |
"print(\"Total dataset examples: \" + str(len(df)))\n",
|
@@ -62,9 +108,6 @@
|
|
62 |
"metadata": {},
|
63 |
"outputs": [],
|
64 |
"source": [
|
65 |
-
"from transformers import AutoTokenizer, AutoModelForCausalLM\n",
|
66 |
-
"import torch\n",
|
67 |
-
"\n",
|
68 |
"# Set device to cuda if available, otherwise CPU\n",
|
69 |
"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
|
70 |
"\n",
|
@@ -74,22 +117,6 @@
|
|
74 |
"model.generation_config.pad_token_id = tokenizer.pad_token_id"
|
75 |
]
|
76 |
},
|
77 |
-
{
|
78 |
-
"cell_type": "markdown",
|
79 |
-
"metadata": {},
|
80 |
-
"source": [
|
81 |
-
"## Create prompt to setup the model for better performance"
|
82 |
-
]
|
83 |
-
},
|
84 |
-
{
|
85 |
-
"cell_type": "code",
|
86 |
-
"execution_count": 19,
|
87 |
-
"metadata": {},
|
88 |
-
"outputs": [],
|
89 |
-
"source": [
|
90 |
-
"from src.prompts.prompt import input_text"
|
91 |
-
]
|
92 |
-
},
|
93 |
{
|
94 |
"cell_type": "markdown",
|
95 |
"metadata": {},
|
@@ -144,8 +171,6 @@
|
|
144 |
}
|
145 |
],
|
146 |
"source": [
|
147 |
-
"import sqlite3 as sql\n",
|
148 |
-
"\n",
|
149 |
"# Create connection to sqlite3 database\n",
|
150 |
"connection = sql.connect('./nba-data/nba.sqlite')\n",
|
151 |
"cursor = connection.cursor()\n",
|
@@ -193,115 +218,12 @@
|
|
193 |
}
|
194 |
],
|
195 |
"source": [
|
196 |
-
"import math\n",
|
197 |
-
"from src.evaluation.compare_result import compare_result_two\n",
|
198 |
-
"\n",
|
199 |
-
"def compare_result(sample_query, sample_result, query_output):\n",
|
200 |
-
" # Clean model output to only have the query output\n",
|
201 |
-
" if query_output[0:7] == \"SQLite:\":\n",
|
202 |
-
" query = query_output[7:]\n",
|
203 |
-
" elif query_output[0:4] == \"SQL:\":\n",
|
204 |
-
" query = query_output[4:]\n",
|
205 |
-
" else:\n",
|
206 |
-
" query = query_output\n",
|
207 |
-
" \n",
|
208 |
-
" # Try to execute query, if it fails, then this is a failure of the model\n",
|
209 |
-
" try:\n",
|
210 |
-
" # Execute query and obtain result\n",
|
211 |
-
" cursor.execute(query)\n",
|
212 |
-
" rows = cursor.fetchall()\n",
|
213 |
-
"\n",
|
214 |
-
" # Strip all whitespace before comparing queries since there may be differences in spacing, newlines, tabs, etc.\n",
|
215 |
-
" query = query.replace(\" \", \"\").replace(\"\\n\", \"\").replace(\"\\t\", \"\")\n",
|
216 |
-
" sample_query = sample_query.replace(\" \", \"\").replace(\"\\n\", \"\").replace(\"\\t\", \"\")\n",
|
217 |
-
" query_match = (query == sample_query)\n",
|
218 |
-
"\n",
|
219 |
-
" # If the queries match, the results clearly also match\n",
|
220 |
-
" if query_match:\n",
|
221 |
-
" return True, True, True\n",
|
222 |
-
"\n",
|
223 |
-
" # Check if this is a multi-line query\n",
|
224 |
-
" if \"|\" in sample_result or \"(\" in sample_result:\n",
|
225 |
-
" #print(rows)\n",
|
226 |
-
" # Create list of results by stripping separators and splitting on them\n",
|
227 |
-
" if \"(\" in sample_result:\n",
|
228 |
-
" sample_result = sample_result.replace(\"(\", \"\").replace(\")\", \"\")\n",
|
229 |
-
" result_list = sample_result.split(\",\") \n",
|
230 |
-
" else:\n",
|
231 |
-
" result_list = sample_result.split(\"|\") \n",
|
232 |
-
"\n",
|
233 |
-
" # Strip all results in list\n",
|
234 |
-
" for i in range(len(result_list)):\n",
|
235 |
-
" result_list[i] = str(result_list[i]).strip()\n",
|
236 |
-
" \n",
|
237 |
-
" # Loop through model result and see if it matches training example\n",
|
238 |
-
" result = False\n",
|
239 |
-
" for row in rows:\n",
|
240 |
-
" for r in row:\n",
|
241 |
-
" for res in result_list:\n",
|
242 |
-
" try:\n",
|
243 |
-
" if math.isclose(float(r), float(res), abs_tol=0.5):\n",
|
244 |
-
" return True, query_match, True\n",
|
245 |
-
" except:\n",
|
246 |
-
" if r in res or res in r:\n",
|
247 |
-
" return True, query_match, True\n",
|
248 |
-
" \n",
|
249 |
-
" # Check if the model returned a sum of examples as opposed to the whole thing\n",
|
250 |
-
" if len(rows) == 1:\n",
|
251 |
-
" for r in rows[0]:\n",
|
252 |
-
" if r == str(len(result_list)):\n",
|
253 |
-
" return True, query_match, True\n",
|
254 |
-
" \n",
|
255 |
-
" return True, query_match, result\n",
|
256 |
-
" # Else the sample result is a single value or string\n",
|
257 |
-
" else:\n",
|
258 |
-
" #print(rows)\n",
|
259 |
-
" result = False\n",
|
260 |
-
" # Loop through model result and see if it contains the sample result\n",
|
261 |
-
" for row in rows:\n",
|
262 |
-
" for r in row:\n",
|
263 |
-
" # Check by string\n",
|
264 |
-
" if str(r) in str(sample_result):\n",
|
265 |
-
" try:\n",
|
266 |
-
" if math.isclose(float(r), float(sample_result), abs_tol=0.5):\n",
|
267 |
-
" return True, query_match, True\n",
|
268 |
-
" except:\n",
|
269 |
-
" return True, query_match, True\n",
|
270 |
-
" # Check by number, using try incase the cast as float fails\n",
|
271 |
-
" try:\n",
|
272 |
-
" if math.isclose(float(r), float(sample_result), abs_tol=0.5):\n",
|
273 |
-
" return True, query_match, True\n",
|
274 |
-
" except:\n",
|
275 |
-
" pass\n",
|
276 |
-
"\n",
|
277 |
-
" # Check if the model returned a list of examples instead of a total sum (both acceptable)\n",
|
278 |
-
" try:\n",
|
279 |
-
" if len(rows) > 1 and len(rows) == int(sample_result):\n",
|
280 |
-
" return True, query_match, True\n",
|
281 |
-
" if len(rows[0]) > 1 and rows[0][1] is not None and len(rows[0]) == int(sample_result):\n",
|
282 |
-
" return True, query_match, True\n",
|
283 |
-
" except:\n",
|
284 |
-
" pass\n",
|
285 |
-
"\n",
|
286 |
-
" # Compare results and return\n",
|
287 |
-
" return True, query_match, result\n",
|
288 |
-
" except:\n",
|
289 |
-
" return False, False, False\n",
|
290 |
-
"\n",
|
291 |
"# Obtain sample\n",
|
292 |
"sample = df.sample(n=1)\n",
|
293 |
-
"sample_dic = {\n",
|
294 |
-
" \"natural_query\": \"How many home games did the Miami Heat play in the 2021 season?\",\n",
|
295 |
-
" \"sql_query\": \"SELECT COUNT(*) FROM game WHERE team_name_home = 'Miami Heat' AND season_id = '22021';\",\n",
|
296 |
-
" \"result\": 41.0\n",
|
297 |
-
"}\n",
|
298 |
"\n",
|
299 |
-
"sample = pd.DataFrame([sample_dic])\n",
|
300 |
-
"\"\"\"\n",
|
301 |
"print(sample[\"natural_query\"].values[0])\n",
|
302 |
"print(sample[\"sql_query\"].values[0])\n",
|
303 |
"print(sample[\"result\"].values[0])\n",
|
304 |
-
"\"\"\"\n",
|
305 |
"\n",
|
306 |
"# Create message with sample query and run model\n",
|
307 |
"message=[{ 'role': 'user', 'content': input_text + sample[\"natural_query\"].values[0]}]\n",
|
@@ -312,15 +234,10 @@
|
|
312 |
"query_output = tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)\n",
|
313 |
"print(query_output)\n",
|
314 |
"\n",
|
315 |
-
"result = compare_result(sample[\"sql_query\"].values[0], sample[\"result\"].values[0], query_output)\n",
|
316 |
"print(\"Statement valid? \" + str(result[0]))\n",
|
317 |
"print(\"SQLite matched? \" + str(result[1]))\n",
|
318 |
-
"print(\"Result matched? \" + str(result[2]))
|
319 |
-
"\n",
|
320 |
-
"result_two = compare_result_two(cursor, sample[\"sql_query\"].values[0], sample[\"result\"].values[0], query_output)\n",
|
321 |
-
"print(\"Statement valid? \" + str(result_two[0]))\n",
|
322 |
-
"print(\"SQLite matched? \" + str(result_two[1]))\n",
|
323 |
-
"print(\"Result matched? \" + str(result_two[2]))"
|
324 |
]
|
325 |
},
|
326 |
{
|
|
|
7 |
"# Run pre-trained DeepSeek Coder 1.3B Model on Chat-GPT 4o generated dataset"
|
8 |
]
|
9 |
},
|
10 |
+
{
|
11 |
+
"cell_type": "code",
|
12 |
+
"execution_count": 22,
|
13 |
+
"metadata": {},
|
14 |
+
"outputs": [],
|
15 |
+
"source": [
|
16 |
+
"import pandas as pd \n",
|
17 |
+
"import warnings\n",
|
18 |
+
"warnings.filterwarnings(\"ignore\")\n",
|
19 |
+
"from transformers import AutoTokenizer, AutoModelForCausalLM\n",
|
20 |
+
"import torch\n",
|
21 |
+
"import sys\n",
|
22 |
+
"import sqlite3 as sql\n",
|
23 |
+
"from huggingface_hub import snapshot_download"
|
24 |
+
]
|
25 |
+
},
|
26 |
+
{
|
27 |
+
"cell_type": "code",
|
28 |
+
"execution_count": 23,
|
29 |
+
"metadata": {},
|
30 |
+
"outputs": [],
|
31 |
+
"source": [
|
32 |
+
"is_google_colab=False"
|
33 |
+
]
|
34 |
+
},
|
35 |
+
{
|
36 |
+
"cell_type": "code",
|
37 |
+
"execution_count": 24,
|
38 |
+
"metadata": {},
|
39 |
+
"outputs": [],
|
40 |
+
"source": [
|
41 |
+
"if is_google_colab:\n",
|
42 |
+
" hugging_face_path = snapshot_download(\n",
|
43 |
+
" repo_id=\"USC-Applied-NLP-Group/SQL-Generation\",\n",
|
44 |
+
" repo_type=\"model\", \n",
|
45 |
+
" allow_patterns=[\"src/*\"], \n",
|
46 |
+
" )\n",
|
47 |
+
" sys.path.append(hugging_face_path)"
|
48 |
+
]
|
49 |
+
},
|
50 |
+
{
|
51 |
+
"cell_type": "code",
|
52 |
+
"execution_count": 25,
|
53 |
+
"metadata": {},
|
54 |
+
"outputs": [],
|
55 |
+
"source": [
|
56 |
+
"from src.prompts.prompt import input_text\n",
|
57 |
+
"from src.evaluation.compare_result import compare_result"
|
58 |
+
]
|
59 |
+
},
|
60 |
{
|
61 |
"cell_type": "markdown",
|
62 |
"metadata": {},
|
|
|
83 |
}
|
84 |
],
|
85 |
"source": [
|
|
|
|
|
|
|
|
|
86 |
"# Load dataset and check length\n",
|
87 |
"df = pd.read_csv(\"./train-data/sql_train.tsv\", sep='\\t')\n",
|
88 |
"print(\"Total dataset examples: \" + str(len(df)))\n",
|
|
|
108 |
"metadata": {},
|
109 |
"outputs": [],
|
110 |
"source": [
|
|
|
|
|
|
|
111 |
"# Set device to cuda if available, otherwise CPU\n",
|
112 |
"device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
|
113 |
"\n",
|
|
|
117 |
"model.generation_config.pad_token_id = tokenizer.pad_token_id"
|
118 |
]
|
119 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
120 |
{
|
121 |
"cell_type": "markdown",
|
122 |
"metadata": {},
|
|
|
171 |
}
|
172 |
],
|
173 |
"source": [
|
|
|
|
|
174 |
"# Create connection to sqlite3 database\n",
|
175 |
"connection = sql.connect('./nba-data/nba.sqlite')\n",
|
176 |
"cursor = connection.cursor()\n",
|
|
|
218 |
}
|
219 |
],
|
220 |
"source": [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
221 |
"# Obtain sample\n",
|
222 |
"sample = df.sample(n=1)\n",
|
|
|
|
|
|
|
|
|
|
|
223 |
"\n",
|
|
|
|
|
224 |
"print(sample[\"natural_query\"].values[0])\n",
|
225 |
"print(sample[\"sql_query\"].values[0])\n",
|
226 |
"print(sample[\"result\"].values[0])\n",
|
|
|
227 |
"\n",
|
228 |
"# Create message with sample query and run model\n",
|
229 |
"message=[{ 'role': 'user', 'content': input_text + sample[\"natural_query\"].values[0]}]\n",
|
|
|
234 |
"query_output = tokenizer.decode(outputs[0][len(inputs[0]):], skip_special_tokens=True)\n",
|
235 |
"print(query_output)\n",
|
236 |
"\n",
|
237 |
+
"result = compare_result(cursor, sample[\"sql_query\"].values[0], sample[\"result\"].values[0], query_output)\n",
|
238 |
"print(\"Statement valid? \" + str(result[0]))\n",
|
239 |
"print(\"SQLite matched? \" + str(result[1]))\n",
|
240 |
+
"print(\"Result matched? \" + str(result[2]))"
|
|
|
|
|
|
|
|
|
|
|
241 |
]
|
242 |
},
|
243 |
{
|