Spaces:
Running
Running
File size: 14,520 Bytes
dc59b01 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 | # =========================================================
# RLHF TRAINING FOR TEXT2SQL (STABLE PPO VERSION)
# =========================================================
import torch
import torch.nn.functional as F
from datasets import load_dataset
from transformers import AutoTokenizer
from transformers.generation.logits_process import LogitsProcessor, LogitsProcessorList
from trl import PPOTrainer, PPOConfig, AutoModelForSeq2SeqLMWithValueHead
from peft import PeftModel
import os, sys, sqlite3, re, random
sys.path.append(os.path.dirname(os.path.abspath(__file__)))
from execution_reward import execution_reward, extract_tables, extract_columns
try:
import sqlparse # gate PPO updates on parsable SQL only
except Exception: # pragma: no cover
sqlparse = None
# ======================================================
# DEVICE
# ======================================================
os.environ.setdefault("TOKENIZERS_PARALLELISM", "false")
device = "mps" if torch.backends.mps.is_available() else "cuda" if torch.cuda.is_available() else "cpu"
print("Using device:", device)
# ======================================================
# TRAINING SETTINGS
# ======================================================
NUM_EPOCHS = 15
LOG_EVERY = 20
USE_SCHEMA = True
SCHEMA_WARMUP_EPOCHS = 2
MAX_SCHEMA_CHARS = 1500
MAX_OUTPUT_TOKENS = 64 # π Speed up: Reduced max tokens
ROLLOUTS_PER_EPOCH = 1024
# ======================================================
# PATHS
# ======================================================
PROJECT_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
RL_MODEL_PATH = os.path.join(PROJECT_ROOT, "outputs/rlhf_text2sql")
output_dir = RL_MODEL_PATH
DB_ROOT = os.path.join(PROJECT_ROOT, "data/database")
# Explicit resume checkpoint
RESUME_CHECKPOINT = os.path.join(PROJECT_ROOT, "checkpoints/milestone_before_more_dbs")
ADAPTER_PATH = os.path.abspath(os.path.join(PROJECT_ROOT, "checkpoints/sft_adapter_codet5"))
FALLBACK_ADAPTER_PATH = ADAPTER_PATH
FALLBACK_ADAPTER_PATH_2 = os.path.join(PROJECT_ROOT, "checkpoints")
BASE_MODEL = os.environ.get("BASE_MODEL", "Salesforce/codet5-base")
# ======================================================
# LOAD MODEL (LoRA)
# ======================================================
def find_valid_adapter(path_candidates):
# π SAFETY & RESUME: Check for existing milestone first
if os.path.exists(os.path.join(RESUME_CHECKPOINT, "adapter_config.json")):
print(f"\nβ
Resuming RL training from checkpoint: {RESUME_CHECKPOINT}\n")
return RESUME_CHECKPOINT
for p in path_candidates:
if p and os.path.exists(os.path.join(p, "adapter_config.json")):
return os.path.abspath(p)
return None
print("Loading base:", BASE_MODEL)
ADAPTER_PATH = find_valid_adapter([
ADAPTER_PATH,
FALLBACK_ADAPTER_PATH,
FALLBACK_ADAPTER_PATH_2,
])
if ADAPTER_PATH is None:
raise RuntimeError("β No valid LoRA adapter found!")
print("Loading adapter:", ADAPTER_PATH)
# ======================================================
# TOKENIZER
# ======================================================
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, use_fast=False)
if tokenizer.pad_token is None:
tokenizer.pad_token = tokenizer.eos_token
# ======================================================
# LOAD PPO MODEL
# ======================================================
model = AutoModelForSeq2SeqLMWithValueHead.from_pretrained(
BASE_MODEL,
torch_dtype=torch.float32
).to(device)
# π RESUME: Load adapter dynamically and ensure it's trainable
model.pretrained_model = PeftModel.from_pretrained(
model.pretrained_model,
ADAPTER_PATH,
is_trainable=True
)
# ======================================================
# LOAD REFERENCE MODEL (FROZEN)
# ======================================================
ref_model = AutoModelForSeq2SeqLMWithValueHead.from_pretrained(
BASE_MODEL,
torch_dtype=torch.float32
).to(device)
ref_model.pretrained_model = PeftModel.from_pretrained(
ref_model.pretrained_model,
ADAPTER_PATH,
is_trainable=False
)
ref_model.eval()
for p in ref_model.parameters():
p.requires_grad = False
# ======================================================
# TRAINABLE PARAMS β ONLY LoRA + VALUE HEAD
# ======================================================
for name, p in model.named_parameters():
if "lora_" in name or "v_head" in name:
p.requires_grad = True
else:
p.requires_grad = False
model.train()
trainable = sum(p.numel() for p in model.parameters() if p.requires_grad)
total = sum(p.numel() for p in model.parameters())
print(f"Trainable params: {trainable}/{total} ({100*trainable/total:.2f}%)")
model.config.use_cache = False
ref_model.config.use_cache = False
# ======================================================
# DATASET
# ======================================================
print("Loading Spider subset...")
random.seed(0)
TRAIN_DBS = [
# already trained
"flight_1","student_assessment","store_1","bike_1","book_2","chinook_1",
"academic","aircraft","car_1","cinema","club_1","csu_1",
# medium difficulty (NEW)
"college_1","college_2","company_1","company_employee",
"customer_complaints","department_store","employee_hire_evaluation",
"museum_visit","products_for_hire","restaurant_1",
"school_finance","shop_membership","small_bank_1",
"soccer_1","student_1","tvshow","voter_1","world_1"
]
dataset = load_dataset("spider", split="train")
dataset = dataset.filter(lambda x: x["db_id"] in TRAIN_DBS)
def valid_example(x):
return 5 <= len(x["question"].split()) <= 40
dataset = dataset.filter(valid_example)
print("Filtered dataset size:", len(dataset))
def sample_example():
return dataset[random.randrange(len(dataset))]
# ======================================================
# DB UTILITIES
# ======================================================
def get_db_path(db_id):
return os.path.join(DB_ROOT, db_id, f"{db_id}.sqlite")
# π SPEED OPTIMIZATION: Cache schema so we don't spam disk IO
_SCHEMA_CACHE = {}
def get_db_schema_cached(db_path):
if db_path in _SCHEMA_CACHE:
return _SCHEMA_CACHE[db_path]
schema_text = ""
try:
conn = sqlite3.connect(db_path)
cursor = conn.cursor()
tables = cursor.execute("SELECT name FROM sqlite_master WHERE type='table';").fetchall()
for table in tables:
table_name = table[0]
columns = cursor.execute(f"PRAGMA table_info({table_name});").fetchall()
col_names = [col[1] for col in columns]
schema_text += f"{table_name}({', '.join(col_names)}) "
conn.close()
except:
pass
_SCHEMA_CACHE[db_path] = schema_text
return schema_text
# ======================================================
# PROMPT
# ======================================================
def trim_schema(schema: str, max_chars: int = 1200) -> str:
if schema is None:
return ""
schema = str(schema)
if len(schema) <= max_chars:
return schema
return schema[:max_chars]
def build_prompt(question: str, schema: str, use_schema: bool) -> str:
if not use_schema:
return f"### Question:\n{question}\n### SQL:"
schema = trim_schema(schema, max_chars=MAX_SCHEMA_CHARS)
return f"### Database Schema:\n{schema}\n### Question:\n{question}\n### SQL:"
# ======================================================
# PPO CONFIG (STABLE POLICY LEARNING)
# ======================================================
ppo_config = PPOConfig(
learning_rate=5e-6,
batch_size=8,
mini_batch_size=2,
gradient_accumulation_steps=2,
ppo_epochs=1,
init_kl_coef=0.2,
target_kl=0.02,
adap_kl_ctrl=True,
cliprange=0.1,
cliprange_value=0.1,
whiten_rewards=False,
kl_penalty="kl",
max_grad_norm=0.5,
)
trainer = PPOTrainer(
config=ppo_config,
model=model,
ref_model=ref_model,
tokenizer=tokenizer,
)
try:
model.device = torch.device(device)
except Exception:
pass
# ======================================================
# GENERATION CONFIG
# ======================================================
# π SPEED OPTIMIZATION: generation limits and randomness bypass
generation_kwargs = dict(
max_new_tokens=MAX_OUTPUT_TOKENS,
do_sample=True, # TRL Requires do_sample=True
temperature=1.0, # Disabled randomness logic
top_p=1.0, # Disabled randomness logic
top_k=0, # Disabled randomness logic
pad_token_id=tokenizer.pad_token_id,
eos_token_id=tokenizer.eos_token_id,
)
# ======================================================
# TRAIN LOOP (BATCHED & OPTIMIZED)
# ======================================================
print("Starting RL training π (CodeT5 PPO Stable)")
best_reward = -1e9
global_ppo_step = 0
model.train()
for epoch in range(1, NUM_EPOCHS + 1):
epoch_reward_sum = 0
valid_sql_count = 0
total_seen = 0
# Process in exact chunks matching batch_size to avoid buffer remnants
for step in range(0, ROLLOUTS_PER_EPOCH, ppo_config.batch_size):
batch_prompts = []
batch_meta = [] # Store tuple of (question, gold_sql, db_path, db_id)
# π BATCH PREPARATION
for _ in range(ppo_config.batch_size):
example = sample_example()
question = example["question"]
gold_sql = example["query"]
db_id = example["db_id"]
db_path = get_db_path(db_id)
schema = get_db_schema_cached(db_path)
prompt = build_prompt(question, schema, use_schema=True)
batch_prompts.append(prompt)
batch_meta.append((question, gold_sql, db_path, db_id))
# π SPEED OPTIMIZATION: Padded Batch Tokenization (Multiple of 8)
encoded_inputs = tokenizer(
batch_prompts,
return_tensors="pt",
padding=True,
truncation=True,
max_length=512,
pad_to_multiple_of=8
).to(device)
# TRL expects lists of 1D tensors
query_tensors = [encoded_inputs.input_ids[i] for i in range(ppo_config.batch_size)]
# π SPEED OPTIMIZATION: Disable gradients for generation pass
with torch.no_grad():
response_tensors = trainer.generate(
query_tensors,
**generation_kwargs
)
batch_rewards = []
batch_responses_text = []
# π BATCH SQL REWARD EXECUTION (Strictly CPU strings)
for i in range(ppo_config.batch_size):
response = tokenizer.decode(response_tensors[i], skip_special_tokens=True)
batch_responses_text.append(response)
question, gold_sql, db_path, db_id = batch_meta[i]
total_seen += 1
# ---------- BASIC SQL FILTER ----------
if "select" not in response.lower():
batch_rewards.append(torch.tensor(-1.0, dtype=torch.float32).to(device))
continue
# ---------- EXECUTION REWARD ----------
reward = execution_reward(response, db_path, gold_sql)
if reward is None:
batch_rewards.append(torch.tensor(-1.0, dtype=torch.float32).to(device))
continue
reward = float(reward)
# ---------- TABLE BONUS ----------
pred_tables = extract_tables(response)
gold_tables = extract_tables(gold_sql)
if len(gold_tables) > 0:
reward += 0.25 * (len(pred_tables & gold_tables) / len(gold_tables))
# ---------- COLUMN BONUS ----------
pred_cols = extract_columns(response)
gold_cols = extract_columns(gold_sql)
if len(gold_cols) > 0:
reward += 0.15 * (len(pred_cols & gold_cols) / len(gold_cols))
# ---------- CLAMP ----------
reward = max(-1.0, min(1.0, reward))
batch_rewards.append(torch.tensor(reward, dtype=torch.float32).to(device))
epoch_reward_sum += reward
valid_sql_count += 1
# ---------- PPO UPDATE ----------
try:
trainer.step(
query_tensors,
response_tensors,
batch_rewards
)
global_ppo_step += 1
except Exception as e:
print("β οΈ PPO skipped:", e)
continue
# π AUTO CHECKPOINT SAVING: Every 200 PPO Updates
if global_ppo_step > 0 and global_ppo_step % 200 == 0:
step_save_path = os.path.join(PROJECT_ROOT, f"checkpoints/rl_step_{global_ppo_step}")
os.makedirs(step_save_path, exist_ok=True)
# Saves ONLY the adapter, keeping disk usage tiny!
model.save_pretrained(step_save_path)
tokenizer.save_pretrained(step_save_path)
print(f"\nπΎ [AUTO-SAVE] Checkpoint saved at PPO step {global_ppo_step} -> {step_save_path}")
# ---------- LOG ----------
if step % (LOG_EVERY * ppo_config.batch_size) == 0 and valid_sql_count > 0:
print("\n---------------------------")
print(f"Epoch {epoch}/{NUM_EPOCHS} Step {step}/{ROLLOUTS_PER_EPOCH} | Global Update {global_ppo_step}")
print("Avg Reward:", round(epoch_reward_sum/valid_sql_count,3))
print("Valid SQL:", valid_sql_count,"/",total_seen)
# Print sample from latest batch
sample_idx = random.randint(0, ppo_config.batch_size - 1)
print("DB:", batch_meta[sample_idx][3])
print("Q:", batch_meta[sample_idx][0])
print("SQL:", batch_responses_text[sample_idx])
print("Reward:", round(batch_rewards[sample_idx].item(), 3))
# ---------- SAVE BEST MODEL (INSIDE EPOCH) ----------
avg_reward = epoch_reward_sum / max(valid_sql_count, 1)
if avg_reward > best_reward:
best_reward = avg_reward
save_path = os.path.join(PROJECT_ROOT, "checkpoints/best_rlhf_model")
os.makedirs(save_path, exist_ok=True)
model.save_pretrained(save_path)
tokenizer.save_pretrained(save_path)
print(f"\nβ
Saved BEST RLHF model for Epoch {epoch} (reward {best_reward:.3f})") |