File size: 13,093 Bytes
2a26d3b |
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 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 |
# 1. git clone -b v0.5.5-tablegpt-merged https://github.com/zTaoplus/vllm.git
# install tablegpt vllm
## apply diff file (recommended in case of use only)
# 1. pip install vllm==0.5.5
# 2. cd vllm
# 3. git diff 09c7792610ada9f88bbf87d32b472dd44bf23cc2 HEAD -- vllm | patch -p1 -d "$(pip show vllm | grep Location | awk '{print $2}')"
## build from source (dev recommended)
## Note: Building from source may take 10-30 minutes and requires access to GitHub or other repositories. Make sure to configure an HTTP/HTTPS proxy.
## cd vllm && pip install -e . [-v]. The -v flag is optional and can be used to display verbose logs.
# see https://github.com/zTaoplus/TableGPT-hf to view the model-related configs.
import os
import json
from tqdm import tqdm
from vllm import LLM
from vllm.sampling_params import SamplingParams
from transformers import AutoTokenizer
from text2sql.src.gpt_request import (
generate_schema_prompt,
generate_comment_prompt,
cot_wizard,
decouple_question_schema,
generate_sql_file,
parser_sql
) # text2sql.src.
import warnings
import random
# 忽略所有警告
warnings.filterwarnings("ignore")
# import pandas as pd
# from typing import Literal, List, Optional
# from io import StringIO
import sqlite3
# DEFAULT_SYS_MSG = "You are a helpful assistant."
# ENCODER_TYPE = "contrastive"
def get_table_info(db_path, enum_num=None):
# extract create ddls
'''
:param root_place:
:param db_name:
:return:
'''
full_schema_prompt_list = []
conn = sqlite3.connect(db_path)
# Create a cursor object
cursor = conn.cursor()
cursor.execute("SELECT name FROM sqlite_master WHERE type='table'")
tables = cursor.fetchall()
all_tables_info = []
# 表截断
# if len(tables) > 16:
# tables = random.sample(tables, 16)
for table in tables:
if table == 'sqlite_sequence':
continue
# 前几行枚举值
cur_table = "`{}`".format(table[0])
cursor.execute("SELECT * FROM {} LIMIT {}".format(cur_table, enum_num))
row_ls = cursor.fetchall()
cursor.execute(f"PRAGMA table_info({cur_table});")
column_name_tp_ls = cursor.fetchall()
all_columns_info = []
for column_name_tp in column_name_tp_ls:
pos_id = column_name_tp[0] # 字段位置
col_name = column_name_tp[1] # 字段名
col_type = column_name_tp[2] # 字段类型
# 字段枚举值
contains_nan = False
enum_values = []
for row in row_ls:
value = row[pos_id]
if value is None:
contains_nan = True
enum_values.append(str(value))
if len(enum_values) == 0:
enum_values = ["None"]
single_columns_info = {
"name": col_name,
"dtype": col_type,
"values": enum_values,
"contains_nan": contains_nan,
"is_unique": False
}
all_columns_info.append(single_columns_info)
# 列截断
# if len(all_columns_info) > 32:
# all_columns_info = random.sample(all_columns_info, 32)
single_table_info = {"columns": all_columns_info}
all_tables_info.append(single_table_info)
return all_tables_info
def generate_combined_prompts_one_encoder(db_path, question, knowledge=None):
schema_prompt = generate_schema_prompt(db_path, num_rows=None) # This is the entry to collect values
comment_prompt = generate_comment_prompt(question, knowledge)
# encoder_prompt = get_encoder_prompt(table_info)
combined_prompts = schema_prompt + '\n\n' + comment_prompt + cot_wizard() + '\nSELECT '
return combined_prompts
def get_encoder_prompt(table_info):
encode_prompt = "".join(
f"table_{i} as follow:\n<TABLE_CONTENT>\n"
for i in range(len(table_info)))
return encode_prompt
def get_messages_one(db_path, question, knowledge=None):
table_info = get_table_info(db_path, enum_num=3) # 采用几行枚举值
prompt = generate_combined_prompts_one_encoder(db_path, question, knowledge=knowledge)
messages = [
{
"role": "system",
"content": "You are a helpul assistant."
}
]
content = []
for i in range(len(table_info)):
if i == len(table_info) -1:
# 最后一个
content.extend([
{
"type": "text",
"text": f"table_{i} as follow: \n",
},
{"type": "table", "table": table_info[i]},
{
"type": "text",
"text": prompt,
}
]
)
else:
content.extend([
{
"type": "text",
"text": f"table_{i} as follow: \n",
},
{"type": "table", "table": table_info[i]},
]
)
messages.append(
{
"role": "user",
"content": content
}
)
# print("*"*100)
# print(json.dumps(messages, ensure_ascii=False))
# print("*"*100)
# exit()
return messages
def calculate_table_num():
import os
db_dir = "/home/jyuan/LLM/evaluation/table_related_benchmarks/evalset/bird_data/dev_databases"
table_nums = []
cols_nums = []
for file in os.listdir(db_dir):
if "." in file:
continue
db_path = os.path.join(db_dir, file, f"{file}.sqlite")
# print(db_path)
conn = sqlite3.connect(db_path)
# Create a cursor object
cursor = conn.cursor()
cursor.execute("SELECT name FROM sqlite_master WHERE type='table'")
tables = cursor.fetchall()
table_num = len(tables)
table_nums.append(table_num)
for table in tables:
if table == 'sqlite_sequence':
continue
# cursor.execute("SELECT sql FROM sqlite_master WHERE type='table' AND name='{}';".format(table[0]))
# create_prompt = cursor.fetchone()[0]
# schemas[table[0]] = create_prompt
# 前几行枚举值
cur_table = "`{}`".format(table[0])
cursor.execute(f"PRAGMA table_info({cur_table});")
column_name_tp_ls = cursor.fetchall()
if len(column_name_tp_ls) == 115:
print(db_path)
cols_nums.append(len(column_name_tp_ls))
cols_nums = sorted(cols_nums, reverse=True)
print(f"max table: {max(table_nums)}, max columns: {max(cols_nums)}")
print(cols_nums[:10])
def llm_generate_result_encoder(model_name_or_path, gpus_num, messages_ls):
# 批量推理
print("model", model_name_or_path)
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
print(
"load tokenizer {} from {} over.".format(
tokenizer.__class__, model_name_or_path
)
)
model = LLM(
model=model_name_or_path,
max_model_len=12000,
max_num_seqs=1,
dtype="bfloat16",
limit_mm_per_prompt={"table": 16},
gpu_memory_utilization=0.9,
tensor_parallel_size=gpus_num,
)
p = SamplingParams(temperature=0, max_tokens=1024)
outputs = model.chat(messages=messages_ls, sampling_params=p)
generated_res = []
for i, output in enumerate(tqdm(outputs)):
text = output.outputs[0].text
sql = parser_sql(text)
generated_res.append(sql)
return generated_res
def col_nums_max(message):
content = message[1]["content"]
table_nums = 0
col_nums_ls = []
for dic in content:
if dic["type"] == "table":
table_nums += 1
col_num = len(dic["table"]["columns"])
col_nums_ls.append(col_num)
return int(max(col_nums_ls) + 1), table_nums
def llm_generate_result_encoder_one(model_name_or_path, gpus_num, messages_ls):
# 单条推理
print("model", model_name_or_path)
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
print(
"load tokenizer {} from {} over.".format(
tokenizer.__class__, model_name_or_path
)
)
model = LLM(
model=model_name_or_path,
max_model_len=12000,
max_num_seqs=1,
dtype="bfloat16",
limit_mm_per_prompt={"table": 20},
gpu_memory_utilization=0.9
)
p = SamplingParams(temperature=0, max_tokens=1024)
error_ls = []
generated_res = []
for i, messages in enumerate(messages_ls):
# if i != 89:
# continue
try:
# max_col, table_nums = col_nums_max(messages)
# # print("="*100, max_col, table_nums)
# model.llm_engine.model_config.hf_config.encoder_config.max_cols = max_col
# model.llm_engine.model_config.multimodal_config.limit_per_prompt = {"table": table_nums}
outputs = model.chat(messages=messages, sampling_params=p)
text = outputs[0].outputs[0].text
sql = parser_sql(text)
except Exception as e:
error_ls.append(i)
sql = ""
generated_res.append(sql)
if len(error_ls) != 0:
json.dump({"error_id": error_ls}, open("table_related_benchmarks/text2sql/output/error_ls.json", 'w'), indent=4)
return generated_res
def collect_response_from_gpt_encoder(model_path, gpus_num, db_path_list, question_list, knowledge_list=None):
'''
:param db_path: str
:param question_list: []
:return: dict of responses collected from llm
'''
responses_dict = {}
response_list = []
messages_ls = []
for i in tqdm(range(len(question_list)), desc="get prompt"):
# print('--------------------- processing {}th question ---------------------'.format(i))
# print('the question is: {}'.format(question))
question = question_list[i]
db_path = db_path_list[i]
if knowledge_list:
messages = get_messages_one(db_path, question, knowledge=knowledge_list[i])
else:
messages = get_messages_one(db_path, question)
messages_ls.append(messages)
outputs_sql = llm_generate_result_encoder(model_path, gpus_num, messages_ls)
for i in tqdm(range(len(outputs_sql)), desc="postprocess result"):
question = question_list[i]
sql = outputs_sql[i]
db_id = db_path_list[i].split('/')[-1].split('.sqlite')[0]
sql = sql + '\t----- bird -----\t' + db_id # to avoid unpredicted \t appearing in codex results
response_list.append(sql)
return response_list
def generate_main_encoder(eval_data, args):
question_list, db_path_list, knowledge_list = decouple_question_schema(
datasets=eval_data, db_root_path=args.db_root_path)
assert len(question_list) == len(db_path_list) == len(knowledge_list)
if args.use_knowledge == 'True':
responses = collect_response_from_gpt_encoder(model_path=args.model_path, gpus_num=args.gpus_num, db_path_list=db_path_list, question_list=question_list, knowledge_list=knowledge_list)
else:
responses = collect_response_from_gpt_encoder(model_path=args.model_path, gpus_num=args.gpus_num, db_path_list=db_path_list, question_list=question_list, knowledge_list=None)
if args.chain_of_thought == 'True':
output_name = os.path.join(args.data_output_path, f'predict_{args.mode}_cot.json')
else:
output_name = os.path.join(args.data_output_path, f'predict_{args.mode}.json')
# pdb.set_trace()
generate_sql_file(sql_lst=responses, output_path=output_name)
print('successfully collect results from {} for {} evaluation; Use knowledge: {}; Use COT: {}; Use encoder: {}'.format(args.model_path, args.mode, args.use_knowledge, args.chain_of_thought, args.use_encoder))
print(f'output: {output_name}')
# 返回推理数据保存路径
return output_name
def test_single():
db_path = "/home/jyuan/LLM/evaluation/table_related_benchmarks/evalset/spider_data/test_database/aan_1/aan_1.sqlite"
question = "How many authors do we have?"
messages = get_messages_one(db_path, question, knowledge=None)
# print("*"*100)
# print(messages)
# print("*"*100)
model_name_or_path = "/data4/workspace/yss/models/longlin_encoder_model/contrastive"
model = LLM(
model=model_name_or_path,
max_model_len=8192,
max_num_seqs=16,
dtype="bfloat16",
limit_mm_per_prompt={"table": 10},
gpu_memory_utilization=0.9,
tensor_parallel_size=1,
)
p = SamplingParams(temperature=0, max_tokens=1024)
res = model.chat(messages=messages, sampling_params=p)
print(res[0].outputs[0].text)
if __name__ == "__main__":
# test_single()
calculate_table_num() |