File size: 9,819 Bytes
8b187bb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
MiniMind Training Utilities
Standard training loop with mixed precision and gradient accumulation.
"""

import os
import math
import time
from typing import Optional, Dict, Any
from pathlib import Path
from dataclasses import dataclass

import torch
import torch.nn as nn
from torch.utils.data import DataLoader
from torch.cuda.amp import GradScaler, autocast

import sys
sys.path.insert(0, str(Path(__file__).parent.parent))
from configs.model_config import Mind2Config


@dataclass
class TrainingConfig:
    """Training configuration."""
    # Optimization
    learning_rate: float = 3e-4
    min_learning_rate: float = 3e-5
    weight_decay: float = 0.1
    beta1: float = 0.9
    beta2: float = 0.95
    grad_clip: float = 1.0
    warmup_steps: int = 1000

    # Training
    num_epochs: int = 3
    batch_size: int = 8
    gradient_accumulation_steps: int = 4
    max_steps: Optional[int] = None

    # Mixed precision
    use_amp: bool = True
    amp_dtype: str = "float16"  # float16 or bfloat16

    # Checkpointing
    save_steps: int = 1000
    eval_steps: int = 500
    output_dir: str = "./outputs"
    resume_from: Optional[str] = None

    # Logging
    log_steps: int = 10
    wandb_project: Optional[str] = None


class Mind2Trainer:
    """Trainer for MiniMind models."""

    def __init__(
        self,
        model: nn.Module,
        train_dataloader: DataLoader,
        eval_dataloader: Optional[DataLoader] = None,
        config: Optional[TrainingConfig] = None,
    ):
        self.model = model
        self.train_dataloader = train_dataloader
        self.eval_dataloader = eval_dataloader
        self.config = config or TrainingConfig()

        self.device = next(model.parameters()).device
        self.global_step = 0
        self.epoch = 0

        # Setup optimizer
        self.optimizer = self._create_optimizer()
        self.scheduler = self._create_scheduler()

        # Mixed precision
        self.scaler = GradScaler() if self.config.use_amp else None
        self.amp_dtype = torch.float16 if self.config.amp_dtype == "float16" else torch.bfloat16

        # Output directory
        Path(self.config.output_dir).mkdir(parents=True, exist_ok=True)

    def _create_optimizer(self) -> torch.optim.Optimizer:
        """Create AdamW optimizer with weight decay."""
        decay_params = []
        no_decay_params = []

        for name, param in self.model.named_parameters():
            if not param.requires_grad:
                continue
            if "bias" in name or "norm" in name or "layernorm" in name:
                no_decay_params.append(param)
            else:
                decay_params.append(param)

        optimizer_groups = [
            {"params": decay_params, "weight_decay": self.config.weight_decay},
            {"params": no_decay_params, "weight_decay": 0.0},
        ]

        return torch.optim.AdamW(
            optimizer_groups,
            lr=self.config.learning_rate,
            betas=(self.config.beta1, self.config.beta2),
        )

    def _create_scheduler(self):
        """Create cosine annealing scheduler with warmup."""
        total_steps = self._get_total_steps()

        def lr_lambda(step):
            if step < self.config.warmup_steps:
                return step / max(1, self.config.warmup_steps)
            progress = (step - self.config.warmup_steps) / max(1, total_steps - self.config.warmup_steps)
            return max(
                self.config.min_learning_rate / self.config.learning_rate,
                0.5 * (1.0 + math.cos(math.pi * progress))
            )

        return torch.optim.lr_scheduler.LambdaLR(self.optimizer, lr_lambda)

    def _get_total_steps(self) -> int:
        if self.config.max_steps:
            return self.config.max_steps
        steps_per_epoch = len(self.train_dataloader) // self.config.gradient_accumulation_steps
        return steps_per_epoch * self.config.num_epochs

    def train(self) -> Dict[str, float]:
        """Main training loop."""
        self.model.train()
        total_steps = self._get_total_steps()

        print(f"Starting training for {total_steps} steps")
        print(f"  Batch size: {self.config.batch_size}")
        print(f"  Gradient accumulation: {self.config.gradient_accumulation_steps}")
        print(f"  Effective batch size: {self.config.batch_size * self.config.gradient_accumulation_steps}")

        running_loss = 0.0
        start_time = time.time()

        for epoch in range(self.config.num_epochs):
            self.epoch = epoch

            for step, batch in enumerate(self.train_dataloader):
                loss = self._training_step(batch)
                running_loss += loss

                if (step + 1) % self.config.gradient_accumulation_steps == 0:
                    self._optimizer_step()
                    self.global_step += 1

                    # Logging
                    if self.global_step % self.config.log_steps == 0:
                        avg_loss = running_loss / self.config.log_steps
                        elapsed = time.time() - start_time
                        tokens_per_sec = (
                            self.config.batch_size * self.config.gradient_accumulation_steps *
                            batch["input_ids"].shape[1] * self.config.log_steps / elapsed
                        )
                        print(
                            f"Step {self.global_step}/{total_steps} | "
                            f"Loss: {avg_loss:.4f} | "
                            f"LR: {self.scheduler.get_last_lr()[0]:.2e} | "
                            f"Tokens/s: {tokens_per_sec:.0f}"
                        )
                        running_loss = 0.0
                        start_time = time.time()

                    # Evaluation
                    if self.eval_dataloader and self.global_step % self.config.eval_steps == 0:
                        eval_loss = self.evaluate()
                        print(f"Eval Loss: {eval_loss:.4f}")
                        self.model.train()

                    # Save checkpoint
                    if self.global_step % self.config.save_steps == 0:
                        self.save_checkpoint()

                    if self.config.max_steps and self.global_step >= self.config.max_steps:
                        break

            if self.config.max_steps and self.global_step >= self.config.max_steps:
                break

        self.save_checkpoint(final=True)
        return {"final_loss": running_loss}

    def _training_step(self, batch: Dict[str, torch.Tensor]) -> float:
        """Single training step."""
        input_ids = batch["input_ids"].to(self.device)
        attention_mask = batch.get("attention_mask", None)
        if attention_mask is not None:
            attention_mask = attention_mask.to(self.device)
        labels = batch["labels"].to(self.device)

        if self.config.use_amp:
            with autocast(dtype=self.amp_dtype):
                loss, _, _, _ = self.model(input_ids, attention_mask, labels)
                loss = loss / self.config.gradient_accumulation_steps
            self.scaler.scale(loss).backward()
        else:
            loss, _, _, _ = self.model(input_ids, attention_mask, labels)
            loss = loss / self.config.gradient_accumulation_steps
            loss.backward()

        return loss.item() * self.config.gradient_accumulation_steps

    def _optimizer_step(self):
        """Optimizer step with gradient clipping."""
        if self.config.use_amp:
            self.scaler.unscale_(self.optimizer)

        torch.nn.utils.clip_grad_norm_(self.model.parameters(), self.config.grad_clip)

        if self.config.use_amp:
            self.scaler.step(self.optimizer)
            self.scaler.update()
        else:
            self.optimizer.step()

        self.scheduler.step()
        self.optimizer.zero_grad()

    @torch.no_grad()
    def evaluate(self) -> float:
        """Evaluate model on eval dataset."""
        self.model.eval()
        total_loss = 0.0
        num_batches = 0

        for batch in self.eval_dataloader:
            input_ids = batch["input_ids"].to(self.device)
            attention_mask = batch.get("attention_mask")
            if attention_mask is not None:
                attention_mask = attention_mask.to(self.device)
            labels = batch["labels"].to(self.device)

            loss, _, _, _ = self.model(input_ids, attention_mask, labels)
            total_loss += loss.item()
            num_batches += 1

        return total_loss / max(1, num_batches)

    def save_checkpoint(self, final: bool = False):
        """Save model checkpoint."""
        checkpoint_name = "final" if final else f"step_{self.global_step}"
        checkpoint_path = Path(self.config.output_dir) / checkpoint_name

        checkpoint_path.mkdir(parents=True, exist_ok=True)

        torch.save(self.model.state_dict(), checkpoint_path / "model.pt")
        torch.save(self.optimizer.state_dict(), checkpoint_path / "optimizer.pt")
        torch.save({
            "global_step": self.global_step,
            "epoch": self.epoch,
            "config": self.config,
        }, checkpoint_path / "trainer_state.pt")

        print(f"Checkpoint saved to {checkpoint_path}")

    def load_checkpoint(self, checkpoint_path: str):
        """Load model checkpoint."""
        path = Path(checkpoint_path)
        self.model.load_state_dict(torch.load(path / "model.pt", map_location=self.device))
        self.optimizer.load_state_dict(torch.load(path / "optimizer.pt", map_location=self.device))

        state = torch.load(path / "trainer_state.pt", map_location=self.device)
        self.global_step = state["global_step"]
        self.epoch = state["epoch"]

        print(f"Checkpoint loaded from {checkpoint_path}")