File size: 6,140 Bytes
e6f4fec |
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 |
import json
import os
import torch
from datasets import Dataset
from transformers import (
AutoTokenizer,
AutoModelForSeq2SeqLM,
Seq2SeqTrainer,
Seq2SeqTrainingArguments,
)
from torch.utils.data import DataLoader
from sklearn.model_selection import train_test_split
from tqdm import tqdm
def load_table_schemas(tables_file):
"""
Load table schemas from the tables.jsonl file.
Args:
tables_file: Path to the tables.jsonl file.
Returns:
A dictionary mapping table IDs to their column names.
"""
table_schemas = {}
with open(tables_file, 'r') as f:
for line in f:
table_data = json.loads(line)
table_id = table_data["id"]
table_columns = table_data["header"]
table_schemas[table_id] = table_columns
return table_schemas
# Step 1: Load and Preprocess WikiSQL Data
def load_wikisql(data_dir):
"""
Load WikiSQL data and prepare it for training.
Args:
data_dir: Path to the WikiSQL dataset directory.
Returns:
List of examples with input and target text.
"""
def parse_file(file_path):
with open(file_path, 'r') as f:
return [json.loads(line) for line in f]
tables_data = parse_file(os.path.join(data_dir, "train.tables.jsonl"))
train_data = parse_file(os.path.join(data_dir, "train.jsonl"))
dev_data = parse_file(os.path.join(data_dir, "dev.jsonl"))
print("====>", train_data[0])
tables_file = "./data/train.tables.jsonl"
table_schemas = load_table_schemas(tables_file)
dev_tables = './data/dev.tables.jsonl'
dev_tables_schema = load_table_schemas(dev_tables)
def format_data(data, type):
formatted = []
for item in data:
table_id = item["table_id"]
table_columns = table_schemas[table_id] if type == 'train' else dev_tables_schema[table_id]
question = item["question"]
sql = item["sql"]
sql_query = sql_to_text(sql, table_columns)
print("SQL Query", sql_query)
formatted.append({"input": f"Question: {question}", "target": sql_query})
return formatted
return format_data(train_data, "train"), format_data(dev_data, "dev")
def sql_to_text(sql, table_columns):
"""
Convert SQL dictionary from WikiSQL to text representation.
Args:
sql: SQL dictionary from WikiSQL (e.g., {"sel": 5, "conds": [[3, 0, "value"]], "agg": 0}).
table_columns: List of column names corresponding to the table.
Returns:
SQL query as a string.
"""
# Aggregation functions mapping
agg_functions = ["", "MAX", "MIN", "COUNT", "SUM", "AVG"]
operators = ["=", ">", "<"]
# Get selected column
sel_column = table_columns[sql["sel"]]
agg_func = agg_functions[sql["agg"]]
select_clause = f"SELECT {agg_func}({sel_column})" if agg_func else f"SELECT {sel_column}"
# Get conditions
if sql["conds"]:
conditions = []
for cond in sql["conds"]:
col_idx, operator, value = cond
col_name = table_columns[col_idx]
conditions.append(f"{col_name} {operators[operator]} '{value}'")
where_clause = " WHERE " + " AND ".join(conditions)
else:
where_clause = ""
# Combine clauses into a full query
return select_clause + where_clause
# Step 2: Tokenize the Data
def tokenize_data(data, tokenizer, max_length=128):
"""
Tokenize the input and target text.
Args:
data: List of examples with "input" and "target".
tokenizer: Pretrained tokenizer.
max_length: Maximum sequence length for the model.
Returns:
Tokenized dataset.
"""
inputs = [item["input"] for item in data]
targets = [item["target"] for item in data]
tokenized = tokenizer(
inputs,
max_length=max_length,
padding="max_length",
truncation=True,
return_tensors="pt",
)
labels = tokenizer(
targets,
max_length=max_length,
padding="max_length",
truncation=True,
return_tensors="pt",
)
tokenized["labels"] = labels["input_ids"]
return tokenized
# Step 3: Load Model and Tokenizer
model_name = "t5-small" # Use "t5-small", "t5-base", or "t5-large"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
# Step 4: Prepare Training and Validation Data
data_dir = "data" # Path to the WikiSQL dataset
train_data, dev_data = load_wikisql(data_dir)
# Tokenize Data
train_dataset = tokenize_data(train_data, tokenizer)
dev_dataset = tokenize_data(dev_data, tokenizer)
# # Convert to Hugging Face Dataset format
train_dataset = Dataset.from_dict(train_dataset)
dev_dataset = Dataset.from_dict(dev_dataset)
# # # Step 5: Define Training Arguments
# training_args = Seq2SeqTrainingArguments(
# output_dir="./t5_sql_finetuned",
# evaluation_strategy="steps",
# save_steps=1000,
# eval_steps=100,
# logging_steps=100,
# per_device_train_batch_size=16,
# per_device_eval_batch_size=16,
# num_train_epochs=3,
# save_total_limit=2,
# learning_rate=5e-5,
# predict_with_generate=True,
# fp16=torch.cuda.is_available(), # Enable mixed precision for faster training
# logging_dir="./logs",
# )
# # # Step 6: Define Trainer
# trainer = Seq2SeqTrainer(
# model=model,
# args=training_args,
# train_dataset=train_dataset,
# eval_dataset=dev_dataset,
# tokenizer=tokenizer,
# )
# # # Step 7: Train the Model
# trainer.train()
# # # Step 8: Save the Model
# trainer.save_model("./t5_sql_finetuned")
# tokenizer.save_pretrained("./t5_sql_finetuned")
# # Step 9: Test the Model
test_question = "Find all orders with product_id greater than 5."
input_text = f"Question: {test_question}"
inputs = tokenizer(input_text, return_tensors="pt", truncation=True, padding=True)
outputs = model.generate(**inputs, max_length=128)
generated_sql = tokenizer.decode(outputs[0], skip_special_tokens=True)
print("Generated SQL:", generated_sql)
|