added gpt model & benchmark (#95)

benchmark/cifar/train.py

@@ -11,12 +11,7 @@ from colossalai.core import global_context as gpc
 from colossalai.logging import get_dist_logger
 from colossalai.nn import Accuracy, CrossEntropyLoss
 from colossalai.nn.lr_scheduler import CosineAnnealingWarmupLR
-from colossalai.trainer import Trainer
-from colossalai.trainer.hooks import (AccuracyHook, LogMemoryByEpochHook,
-                                      LogMetricByEpochHook,
-                                      LogMetricByStepHook,
-                                      LogTimingByEpochHook, LossHook,
-                                      LRSchedulerHook, ThroughputHook)
+from colossalai.trainer import Trainer, hooks
 from colossalai.utils import MultiTimer, get_dataloader
 from model_zoo.vit import vit_lite_depth7_patch4_32
 from torchvision import transforms
@@ -100,22 +95,22 @@ def train_cifar():
     trainer = Trainer(engine=engine, logger=logger, timer=timer)
     logger.info("Trainer is built", ranks=[0])
 
-    hooks = [
-        LogMetricByEpochHook(logger=logger),
-        LogMetricByStepHook(),
-        # LogTimingByEpochHook(timer=timer, logger=logger),
-        # LogMemoryByEpochHook(logger=logger),
-        AccuracyHook(accuracy_func=Accuracy()),
-        LossHook(),
-        ThroughputHook(),
-        LRSchedulerHook(lr_scheduler=lr_scheduler, by_epoch=False)
+    hook_list = [
+        hooks.LogMetricByEpochHook(logger=logger),
+        hooks.LogMetricByStepHook(),
+        # hooks.LogTimingByEpochHook(timer=timer, logger=logger),
+        # hooks.LogMemoryByEpochHook(logger=logger),
+        hooks.AccuracyHook(accuracy_func=Accuracy()),
+        hooks.LossHook(),
+        hooks.ThroughputHook(),
+        hooks.LRSchedulerHook(lr_scheduler=lr_scheduler, by_epoch=False)
     ]
 
     logger.info("Train start", ranks=[0])
     trainer.fit(train_dataloader=train_dataloader,
                 test_dataloader=test_dataloader,
                 epochs=gpc.config.NUM_EPOCHS,
-                hooks=hooks,
+                hooks=hook_list,
                 display_progress=True,
                 test_interval=1)
 

benchmark/gpt2/configs/gpt2_1d.py

+from colossalai.amp import AMP_TYPE
+
+VOCAB_SIZE = 50304
+SEQ_LENGTH = 1024
+
+TOTAL_BATCH_SIZE = 256
+LEARNING_RATE = 0.00015
+WEIGHT_DECAY = 1e-2
+
+TENSOR_PARALLEL_SIZE = 2
+TENSOR_PARALLEL_MODE = '1d'
+
+NUM_EPOCHS = 60
+WARMUP_EPOCHS = int(NUM_EPOCHS * 0.36)
+
+parallel = dict(
+    pipeline=1,
+    tensor=dict(mode=TENSOR_PARALLEL_MODE, size=TENSOR_PARALLEL_SIZE),
+)
+
+fp16 = dict(mode=AMP_TYPE.TORCH, )
+
+gradient_accumulation = 2
+
+BATCH_SIZE = TOTAL_BATCH_SIZE // gradient_accumulation
+
+clip_grad_norm = 1.0
+
+LOG_PATH = f"./gpt2_{TENSOR_PARALLEL_MODE}_tp{TENSOR_PARALLEL_SIZE}_bs{BATCH_SIZE}_lr{LEARNING_RATE}_accum{gradient_accumulation}_clip_grad{clip_grad_norm}/"

benchmark/gpt2/configs/gpt2_2d.py

+from colossalai.amp import AMP_TYPE
+
+VOCAB_SIZE = 50304
+SEQ_LENGTH = 1024
+
+TOTAL_BATCH_SIZE = 256
+LEARNING_RATE = 0.00015
+WEIGHT_DECAY = 1e-2
+
+TENSOR_PARALLEL_SIZE = 4
+TENSOR_PARALLEL_MODE = '2d'
+
+NUM_EPOCHS = 60
+WARMUP_EPOCHS = int(NUM_EPOCHS * 0.36)
+
+parallel = dict(
+    pipeline=1,
+    tensor=dict(mode=TENSOR_PARALLEL_MODE, size=TENSOR_PARALLEL_SIZE),
+)
+
+fp16 = dict(mode=AMP_TYPE.TORCH, )
+
+gradient_accumulation = 1
+
+BATCH_SIZE = TOTAL_BATCH_SIZE // gradient_accumulation
+
+clip_grad_norm = 1.0
+
+LOG_PATH = f"./gpt2_{TENSOR_PARALLEL_MODE}_tp{TENSOR_PARALLEL_SIZE}_bs{BATCH_SIZE}_lr{LEARNING_RATE}_accum{gradient_accumulation}_clip_grad{clip_grad_norm}/"

benchmark/gpt2/configs/gpt2_2p5d.py

+from colossalai.amp import AMP_TYPE
+
+VOCAB_SIZE = 50304
+SEQ_LENGTH = 1024
+
+TOTAL_BATCH_SIZE = 256
+LEARNING_RATE = 0.00015
+WEIGHT_DECAY = 1e-2
+
+TENSOR_PARALLEL_SIZE = 4
+DEPTH = 1
+TENSOR_PARALLEL_MODE = '2.5d'
+
+NUM_EPOCHS = 60
+WARMUP_EPOCHS = int(NUM_EPOCHS * 0.36)
+
+parallel = dict(
+    pipeline=1,
+    tensor=dict(mode=TENSOR_PARALLEL_MODE, size=TENSOR_PARALLEL_SIZE, depth=DEPTH),
+)
+
+fp16 = dict(mode=AMP_TYPE.TORCH, )
+
+gradient_accumulation = 1
+
+BATCH_SIZE = TOTAL_BATCH_SIZE // gradient_accumulation
+
+clip_grad_norm = 1.0
+
+LOG_PATH = f"./gpt2_{TENSOR_PARALLEL_MODE}_tp{TENSOR_PARALLEL_SIZE}_bs{BATCH_SIZE}_lr{LEARNING_RATE}_accum{gradient_accumulation}_clip_grad{clip_grad_norm}/"

benchmark/gpt2/configs/gpt2_3d.py

+from colossalai.amp import AMP_TYPE
+
+VOCAB_SIZE = 50304
+SEQ_LENGTH = 1024
+
+TOTAL_BATCH_SIZE = 256
+LEARNING_RATE = 0.00015
+WEIGHT_DECAY = 1e-2
+
+TENSOR_PARALLEL_SIZE = 8
+TENSOR_PARALLEL_MODE = '3d'
+
+NUM_EPOCHS = 60
+WARMUP_EPOCHS = int(NUM_EPOCHS * 0.36)
+
+parallel = dict(
+    pipeline=1,
+    tensor=dict(mode=TENSOR_PARALLEL_MODE, size=TENSOR_PARALLEL_SIZE),
+)
+
+fp16 = dict(mode=AMP_TYPE.TORCH, )
+
+gradient_accumulation = 1
+
+BATCH_SIZE = TOTAL_BATCH_SIZE // gradient_accumulation
+
+clip_grad_norm = 1.0
+
+LOG_PATH = f"./gpt2_{TENSOR_PARALLEL_MODE}_tp{TENSOR_PARALLEL_SIZE}_bs{BATCH_SIZE}_lr{LEARNING_RATE}_accum{gradient_accumulation}_clip_grad{clip_grad_norm}/"

benchmark/gpt2/configs/gpt2_vanilla.py

+from colossalai.amp import AMP_TYPE
+
+VOCAB_SIZE = 50304
+SEQ_LENGTH = 1024
+
+TOTAL_BATCH_SIZE = 256
+LEARNING_RATE = 0.00015
+WEIGHT_DECAY = 1e-2
+
+TENSOR_PARALLEL_SIZE = 1
+TENSOR_PARALLEL_MODE = None
+
+NUM_EPOCHS = 60
+WARMUP_EPOCHS = int(NUM_EPOCHS * 0.36)
+
+parallel = dict(
+    pipeline=1,
+    tensor=dict(mode=TENSOR_PARALLEL_MODE, size=TENSOR_PARALLEL_SIZE),
+)
+
+fp16 = dict(mode=AMP_TYPE.TORCH, )
+
+gradient_accumulation = 1
+
+BATCH_SIZE = TOTAL_BATCH_SIZE // gradient_accumulation
+
+clip_grad_norm = 1.0
+
+LOG_PATH = f"./gpt2_{TENSOR_PARALLEL_MODE}_tp{TENSOR_PARALLEL_SIZE}_bs{BATCH_SIZE}_lr{LEARNING_RATE}_accum{gradient_accumulation}_clip_grad{clip_grad_norm}/"

benchmark/gpt2/data.py

+import json
+import os
+
+import torch
+from colossalai.registry import DATASETS
+from torch.utils.data import Dataset
+from transformers import GPT2Tokenizer
+
+
+@DATASETS.register_module
+class WebtextDataset(Dataset):
+    def __init__(self, path, seq_len=1024) -> None:
+        super().__init__()
+        root = os.path.dirname(path)
+        encoded_data_cache_path = os.path.join(root, f'gpt_webtext_{seq_len}.pt')
+        if os.path.isfile(encoded_data_cache_path):
+            seq_len_, data, attention_mask = torch.load(encoded_data_cache_path)
+            if seq_len_ == seq_len:
+                self.data = data
+                self.attention_mask = attention_mask
+                return
+        raw_data = []
+        with open(path) as f:
+            for line in f.readlines():
+                raw_data.append(json.loads(line)['text'])
+        tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
+        tokenizer.pad_token = tokenizer.unk_token
+        encoded_data = tokenizer(raw_data, padding=True, truncation=True, max_length=seq_len, return_tensors='pt')
+        self.data = encoded_data['input_ids']
+        self.attention_mask = encoded_data['attention_mask']
+        torch.save((seq_len, self.data, self.attention_mask), encoded_data_cache_path)
+
+    def __len__(self):
+        return len(self.data)
+
+    def __getitem__(self, index):
+        return (self.data[index], self.attention_mask[index]), self.data[index]

benchmark/gpt2/train.py

+import contextlib
+import os
+
+import colossalai
+import torch
+from colossalai.core import global_context as gpc
+from colossalai.engine.schedule import (InterleavedPipelineSchedule, PipelineSchedule)
+from colossalai.logging import get_dist_logger
+from colossalai.nn import CosineAnnealingWarmupLR
+from colossalai.trainer import Trainer, hooks
+from colossalai.utils import MultiTimer, get_dataloader
+from colossalai.zero import zero3_model_context
+from model_zoo.gpt import GPTLMLoss, gpt2_small, gpt2_medium, gpt2_large, gpt2_xl
+
+from data import WebtextDataset
+
+
+def train_gpt():
+    args = colossalai.get_default_parser().parse_args()
+    # standard launch
+    # colossalai.launch(config=args.config,
+    #                   rank=args.rank,
+    #                   world_size=args.world_size,
+    #                   local_rank=args.local_rank,
+    #                   host=args.host,
+    #                   port=args.port)
+
+    # launch from torchrun
+    colossalai.launch_from_torch(config=args.config)
+
+    logger = get_dist_logger()
+    if hasattr(gpc.config, 'LOG_PATH'):
+        if gpc.get_global_rank() == 0:
+            log_path = gpc.config.LOG_PATH
+            if not os.path.exists(log_path):
+                os.mkdir(log_path)
+            logger.log_to_file(log_path)
+
+    train_dataset = WebtextDataset(os.environ['DATA'], seq_len=gpc.config.SEQ_LENGTH)
+    train_dataloader = get_dataloader(train_dataset,
+                                      seed=42,
+                                      batch_size=gpc.config.BATCH_SIZE // gpc.data_parallel_size,
+                                      pin_memory=True,
+                                      shuffle=True,
+                                      drop_last=True)
+    logger.info(f'Loaded {len(train_dataset)}/{len(train_dataloader)} samples/batches', ranks=[0])
+
+    # zero3 under test
+    # use_zero3 = hasattr(gpc.config, 'zero') and gpc.config.zero.level == 3
+    # cm = zero3_model_context() if use_zero3 else contextlib.nullcontext()
+    # with cm:
+    #     model = gpc.config.model.pop('type')(**gpc.config.model)
+
+    model = gpt2_medium(vocab_size=gpc.config.VOCAB_SIZE,
+                        max_position_embeddings=gpc.config.SEQ_LENGTH,
+                        checkpoint=True)
+
+    criterion = GPTLMLoss()
+
+    optimizer = torch.optim.Adam(model.parameters(), lr=0.00015, weight_decay=1e-2)
+
+    steps_per_epoch = len(train_dataloader) // gpc.config.gradient_accumulation
+
+    lr_scheduler = CosineAnnealingWarmupLR(optimizer=optimizer,
+                                           total_steps=gpc.config.NUM_EPOCHS * steps_per_epoch,
+                                           warmup_steps=gpc.config.WARMUP_EPOCHS * steps_per_epoch,
+                                           eta_min=1e-5)
+
+    engine, train_dataloader, _, lr_scheduler = colossalai.initialize(model=model,
+                                                                      optimizer=optimizer,
+                                                                      criterion=criterion,
+                                                                      train_dataloader=train_dataloader,
+                                                                      lr_scheduler=lr_scheduler)
+
+    # pipeline under test
+    # num_model_chunks = getattr(gpc.config.model, 'num_chunks', 1)
+    # if num_model_chunks > 1:
+    #     logger.info('Build InterleavedPipelineSchedule', ranks=[0])
+    #     schedule = InterleavedPipelineSchedule(gpc.config.NUM_MICRO_BATCHES, num_model_chunks)
+    # else:
+    #     logger.info('Build PipelineSchedule', ranks=[0])
+    #     schedule = PipelineSchedule(gpc.config.NUM_MICRO_BATCHES)
+
+    timer = MultiTimer()
+
+    trainer = Trainer(engine=engine, logger=logger, timer=timer)
+
+    hook_list = [
+        hooks.LogMetricByEpochHook(logger=logger),
+        hooks.LogMetricByStepHook(),
+        hooks.LossHook(),
+        hooks.ThroughputHook(),
+        hooks.LRSchedulerHook(lr_scheduler=lr_scheduler, by_epoch=False),
+        # hooks.TensorboardHook(log_dir='./tb_logs', ranks=[0]),
+        # hooks.LogMemoryByEpochHook(logger),
+        # hooks.LogTimingByEpochHook(timer, logger),
+        # hooks.SaveCheckpointHook(checkpoint_dir='./ckpt')
+    ]
+
+    logger.info("Training start", ranks=[0])
+    trainer.fit(train_dataloader=train_dataloader, epochs=gpc.config.NUM_EPOCHS, hooks=hook_list, display_progress=True)
+
+
+if __name__ == '__main__':
+    train_gpt()

benchmark/imagenet100/train.py

@@ -14,9 +14,7 @@ from colossalai.core import global_context as gpc
 from colossalai.logging import get_dist_logger
 from colossalai.nn import Accuracy, CrossEntropyLoss
 from colossalai.nn.lr_scheduler import CosineAnnealingWarmupLR
-from colossalai.trainer import Trainer
-from colossalai.trainer.hooks import (AccuracyHook, LogMemoryByEpochHook, LogMetricByEpochHook, LogMetricByStepHook,
-                                      LogTimingByEpochHook, LossHook, LRSchedulerHook, ThroughputHook)
+from colossalai.trainer import Trainer, hooks
 from colossalai.utils import MultiTimer
 from model_zoo.vit import vit_small_patch16_224
 from nvidia.dali import types
@@ -185,22 +183,22 @@ def train_imagenet():
     trainer = Trainer(engine=engine, logger=logger, timer=timer)
     logger.info("Trainer is built", ranks=[0])
 
-    hooks = [
-        LogMetricByEpochHook(logger=logger),
-        LogMetricByStepHook(),
-        # LogTimingByEpochHook(timer=timer, logger=logger),
-        # LogMemoryByEpochHook(logger=logger),
-        AccuracyHook(accuracy_func=Accuracy()),
-        LossHook(),
-        ThroughputHook(),
-        LRSchedulerHook(lr_scheduler=lr_scheduler, by_epoch=True)
+    hook_list = [
+        hooks.LogMetricByEpochHook(logger=logger),
+        hooks.LogMetricByStepHook(),
+        # hooks.LogTimingByEpochHook(timer=timer, logger=logger),
+        # hooks.LogMemoryByEpochHook(logger=logger),
+        hooks.AccuracyHook(accuracy_func=Accuracy()),
+        hooks.LossHook(),
+        hooks.ThroughputHook(),
+        hooks.LRSchedulerHook(lr_scheduler=lr_scheduler, by_epoch=True)
     ]
 
     logger.info("Train start", ranks=[0])
     trainer.fit(train_dataloader=train_dataloader,
                 test_dataloader=test_dataloader,
                 epochs=gpc.config.NUM_EPOCHS,
-                hooks=hooks,
+                hooks=hook_list,
                 display_progress=True,
                 test_interval=1)
 

benchmark/imagenet1k/train.py

@@ -14,9 +14,7 @@ from colossalai.core import global_context as gpc
 from colossalai.logging import get_dist_logger
 from colossalai.nn import Accuracy, CrossEntropyLoss
 from colossalai.nn.lr_scheduler import CosineAnnealingWarmupLR
-from colossalai.trainer import Trainer
-from colossalai.trainer.hooks import (AccuracyHook, LogMemoryByEpochHook, LogMetricByEpochHook, LogMetricByStepHook,
-                                      LogTimingByEpochHook, LossHook, LRSchedulerHook, ThroughputHook)
+from colossalai.trainer import Trainer, hooks
 from colossalai.utils import MultiTimer
 from model_zoo.vit import vit_small_patch16_224
 from nvidia.dali import types
@@ -185,22 +183,22 @@ def train_imagenet():
     trainer = Trainer(engine=engine, logger=logger, timer=timer)
     logger.info("Trainer is built", ranks=[0])
 
-    hooks = [
-        LogMetricByEpochHook(logger=logger),
-        LogMetricByStepHook(),
-        # LogTimingByEpochHook(timer=timer, logger=logger),
-        # LogMemoryByEpochHook(logger=logger),
-        AccuracyHook(accuracy_func=Accuracy()),
-        LossHook(),
-        ThroughputHook(),
-        LRSchedulerHook(lr_scheduler=lr_scheduler, by_epoch=True)
+    hook_list = [
+        hooks.LogMetricByEpochHook(logger=logger),
+        hooks.LogMetricByStepHook(),
+        # hooks.LogTimingByEpochHook(timer=timer, logger=logger),
+        # hooks.LogMemoryByEpochHook(logger=logger),
+        hooks.AccuracyHook(accuracy_func=Accuracy()),
+        hooks.LossHook(),
+        hooks.ThroughputHook(),
+        hooks.LRSchedulerHook(lr_scheduler=lr_scheduler, by_epoch=True)
     ]
 
     logger.info("Train start", ranks=[0])
     trainer.fit(train_dataloader=train_dataloader,
                 test_dataloader=test_dataloader,
                 epochs=gpc.config.NUM_EPOCHS,
-                hooks=hooks,
+                hooks=hook_list,
                 display_progress=True,
                 test_interval=1)
 

model_zoo/gpt/__init__.py

+from .gpt import *
\ No newline at end of file

model_zoo/gpt/gpt.py

+import math
+from typing import Callable
+
+import torch
+from colossalai import nn as col_nn
+from colossalai.nn.layer.utils import CheckpointModule
+from colossalai.registry import LAYERS, MODELS, LOSSES
+from colossalai.utils import get_current_device
+from torch import dtype, nn
+
+__all__ = ['GPT', 'GPTLMLoss', 'gpt2_small', 'gpt2_medium', 'gpt2_large', 'gpt2_xl', 'gpt3']
+
+
+@LAYERS.register_module
+class GPTEmbedding(nn.Module):
+    def __init__(self,
+                 embedding_dim: int,
+                 vocab_size: int,
+                 max_position_embeddings: int,
+                 num_tokentypes: int = 0,
+                 padding_idx: int = 0,
+                 dropout: float = 0.,
+                 dtype: dtype = None) -> None:
+        super().__init__()
+        self.word_embeddings = col_nn.Embedding(vocab_size, embedding_dim, padding_idx=padding_idx, dtype=dtype)
+        self.position_embeddings = col_nn.Embedding(max_position_embeddings, embedding_dim, dtype=dtype)
+        if num_tokentypes > 0:
+            self.tokentype_embeddings = col_nn.Embedding(num_tokentypes, embedding_dim, dtype=dtype)
+        else:
+            self.tokentype_embeddings = None
+        self.dropout = col_nn.Dropout(dropout)
+
+    @property
+    def word_embedding_weight(self):
+        return self.word_embeddings.weight
+
+    def forward(self, input_ids, position_ids=None, tokentype_ids=None):
+        seq_length = input_ids.size(1)
+        if position_ids is None:
+            position_ids = torch.arange(seq_length, dtype=torch.long, device=get_current_device()).unsqueeze(0)
+        x = self.word_embeddings(input_ids) + self.position_embeddings(position_ids)
+        if self.tokentype_embeddings is not None and tokentype_ids is not None:
+            x = x + self.tokentype_embeddings(tokentype_ids)
+        x = self.dropout(x)
+        return x
+
+
+@LAYERS.register_module
+class GPTSelfAttention(nn.Module):
+    def __init__(self,
+                 dim: int,
+                 num_heads: int,
+                 attention_dropout: float,
+                 dropout: float,
+                 bias: bool = True,
+                 dtype: dtype = None) -> None:
+        super().__init__()
+
+        self.attention_head_size = dim // num_heads
+        self.query_key_value = col_nn.Linear(dim, 3 * dim, dtype=dtype, bias=bias)
+        self.attention_dropout = col_nn.Dropout(attention_dropout)
+        self.dense = col_nn.Linear(dim, dim, dtype=dtype, bias=True)
+        self.dropout = col_nn.Dropout(dropout)
+        self.softmax = nn.Softmax(dim=-1)
+
+    def forward(self, x, attention_mask=None):
+        qkv = self.query_key_value(x)
+        all_head_size = qkv.shape[-1] // 3
+        num_attention_heads = all_head_size // self.attention_head_size
+        new_qkv_shape = qkv.shape[:-1] + \
+            (num_attention_heads, 3 * self.attention_head_size)
+        qkv = qkv.view(new_qkv_shape)
+        qkv = qkv.permute((0, 2, 1, 3))
+        q, k, v = torch.chunk(qkv, 3, dim=-1)
+
+        x = torch.matmul(q, k.transpose(-1, -2))
+        x = x / math.sqrt(self.attention_head_size)
+
+        # causal mask
+        q_len, k_len = q.size(-2), k.size(-2)
+        causal_mask = torch.tril(torch.ones((q_len, k_len), dtype=torch.uint8,
+                                            device=get_current_device())).view(1, 1, q_len, k_len).bool()
+        x = torch.where(causal_mask, x, torch.tensor(-1e4, dtype=x.dtype, device=get_current_device()))
+
+        if attention_mask is not None:
+            x = x + attention_mask
+        x = self.softmax(x)
+        x = self.attention_dropout(x)
+
+        x = torch.matmul(x, v)
+        x = x.transpose(1, 2)
+        new_context_layer_shape = x.size()[:-2] + (all_head_size, )
+        x = x.reshape(new_context_layer_shape)
+
+        x = self.dense(x)
+        x = self.dropout(x)
+
+        return x
+
+
+@LAYERS.register_module
+class GPTMLP(nn.Module):
+    def __init__(self,
+                 dim: int,
+                 mlp_ratio: int,
+                 activation: Callable,
+                 dropout: float,
+                 dtype: dtype = None,
+                 bias: bool = True):
+        super().__init__()
+        self.dense_1 = col_nn.Linear(dim, mlp_ratio * dim, dtype=dtype, bias=bias)
+        self.activation = activation
+        self.dense_2 = col_nn.Linear(mlp_ratio * dim, dim, dtype=dtype, bias=bias)
+        self.dropout = col_nn.Dropout(dropout)
+
+    def forward(self, x):
+        x = self.dense_1(x)
+        x = self.activation(x)
+        x = self.dense_2(x)
+        x = self.dropout(x)
+        return x
+
+
+@LAYERS.register_module
+class GPTBlock(CheckpointModule):
+    def __init__(self,
+                 dim: int,
+                 num_heads: int,
+                 mlp_ratio: int,
+                 activation: Callable,
+                 attention_dropout: float = 0.,
+                 dropout: float = 0.,
+                 dtype: dtype = None,
+                 bias: bool = True,
+                 checkpoint: bool = False):
+        super().__init__()
+        self.norm1 = col_nn.LayerNorm(normalized_shape=dim, eps=1e-6, dtype=dtype)
+        self.attn = GPTSelfAttention(dim=dim,
+                                     num_heads=num_heads,
+                                     attention_dropout=attention_dropout,
+                                     dropout=dropout,
+                                     bias=bias,
+                                     dtype=dtype)
+        self.norm2 = col_nn.LayerNorm(normalized_shape=dim, eps=1e-6, dtype=dtype)
+        self.mlp = GPTMLP(dim=dim, mlp_ratio=mlp_ratio, activation=activation, dropout=dropout, dtype=dtype, bias=bias)
+
+    def _forward(self, x, attention_mask=None):
+        x = x + self.attn(self.norm1(x), attention_mask)
+        x = x + self.mlp(self.norm2(x))
+        return x, attention_mask
+
+
+@LAYERS.register_module
+class GPTLMHead(nn.Module):
+    def __init__(self,
+                 dim: int,
+                 vocab_size: int,
+                 word_embeeding_weight: nn.Parameter = None,
+                 bias: bool = False,
+                 dtype: dtype = None) -> None:
+        super().__init__()
+        self.dense = col_nn.Classifier(dim, vocab_size, word_embeeding_weight, bias=bias, dtype=dtype)
+
+    def forward(self, x):
+        x = self.dense(x)
+        return x
+
+
+@LOSSES.register_module
+class GPTLMLoss(nn.Module):
+    def __init__(self):
+        super().__init__()
+        self.loss = col_nn.CrossEntropyLoss()
+
+    def forward(self, logits, labels):
+        shift_logits = logits[..., :-1, :].contiguous()
+        shift_labels = labels[..., 1:].contiguous()
+        # Flatten the tokens
+        return self.loss(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
+
+
+@MODELS.register_module
+class GPT(nn.Module):
+    def __init__(self,
+                 vocab_size: int = 50304,
+                 max_position_embeddings: int = 1024,
+                 dim: int = 768,
+                 num_heads: int = 12,
+                 depth: int = 12,
+                 mlp_ratio: int = 4,
+                 dropout: float = 0.1,
+                 embedding_dropout: float = 0.1,
+                 attention_dropout: float = 0.1,
+                 layernorm_epsilon: float = 1e-5,
+                 activation: Callable = nn.functional.gelu,
+                 checkpoint: bool = False,
+                 dtype: dtype = None,
+                 bias: bool = True,
+                 padding_idx: int = 0) -> None:
+        super().__init__()
+        self.dtype = dtype
+        self.embed = GPTEmbedding(embedding_dim=dim,
+                                  vocab_size=vocab_size,
+                                  max_position_embeddings=max_position_embeddings,
+                                  padding_idx=padding_idx,
+                                  dropout=embedding_dropout,
+                                  dtype=dtype)
+        self.blocks = nn.ModuleList([
+            GPTBlock(
+                dim=dim,
+                num_heads=num_heads,
+                mlp_ratio=mlp_ratio,
+                activation=activation,
+                attention_dropout=attention_dropout,
+                dropout=dropout,
+                dtype=dtype,
+                bias=bias,
+                checkpoint=checkpoint,
+            ) for _ in range(depth)
+        ])
+
+        self.norm = col_nn.LayerNorm(normalized_shape=dim, eps=layernorm_epsilon, dtype=dtype)
+
+        self.head = GPTLMHead(dim=dim,
+                              vocab_size=vocab_size,
+                              word_embeeding_weight=self.embed.word_embedding_weight,
+                              bias=bias,
+                              dtype=dtype)
+
+    def forward(self, input_ids, attention_mask=None):
+        # We create a 3D attention mask from a 2D tensor mask.
+        # Sizes are [batch_size, 1, 1, to_seq_length]
+        # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
+        # Adapted from huggingface
+        if attention_mask is not None:
+            batch_size = input_ids.shape[0]
+            attention_mask = attention_mask.view(batch_size, -1)
+            attention_mask = attention_mask.unsqueeze(1).unsqueeze(2)
+            attention_mask = attention_mask.to(dtype=self.dtype)  # fp16 compatibility
+            attention_mask = (1.0 - attention_mask) * -10000.0
+
+        x = self.embed(input_ids)
+
+        for block in self.blocks:
+            x, attention_mask = block(x, attention_mask)
+
+        x = self.head(self.norm(x))
+
+        return x
+
+
+def _create_gpt_model(**model_kwargs):
+    model = GPT(**model_kwargs)
+    return model
+
+
+@MODELS.register_module
+def gpt2_small(**kwargs):
+    model_kwargs = dict(dim=768, depth=12, num_heads=12, **kwargs)
+    return _create_gpt_model(**model_kwargs)
+
+
+@MODELS.register_module
+def gpt2_medium(**kwargs):
+    model_kwargs = dict(dim=1024, depth=24, num_heads=16, **kwargs)
+    return _create_gpt_model(**model_kwargs)
+
+
+@MODELS.register_module
+def gpt2_large(**kwargs):
+    model_kwargs = dict(dim=1280, depth=36, num_heads=20, **kwargs)
+    return _create_gpt_model(**model_kwargs)
+
+
+@MODELS.register_module
+def gpt2_xl(**kwargs):
+    model_kwargs = dict(dim=1600, depth=48, num_heads=25, **kwargs)
+    return _create_gpt_model(**model_kwargs)
+
+
+@MODELS.register_module
+def gpt3(**kwargs):
+    model_kwargs = dict(dim=12288, max_position_embeddings=2048, depth=96, num_heads=96, **kwargs)
+    return _create_gpt_model(**model_kwargs)

model_zoo/vit/vit.py

@@ -89,7 +89,7 @@ class ViTEmbedding(nn.Module):
 
 
 @LAYERS.register_module
-class ViTSelfAttention(CheckpointModule):
+class ViTSelfAttention(nn.Module):
     def __init__(self,
                  dim: int,
                  num_heads: int,
@@ -97,9 +97,8 @@ class ViTSelfAttention(CheckpointModule):
                  dropout: float,
                  bias: bool = True,
                  dtype: dtype = None,
-                 checkpoint: bool = False,
                  init_method: str = 'torch'):
-        super().__init__(checkpoint)
+        super().__init__()
         self.attention_head_size = dim // num_heads
         self.query_key_value = col_nn.Linear(dim,
                                              3 * dim,
@@ -111,7 +110,7 @@ class ViTSelfAttention(CheckpointModule):
         self.dropout = col_nn.Dropout(dropout)
         self.softmax = nn.Softmax(dim=-1)
 
-    def _forward(self, x):
+    def forward(self, x):
         qkv = self.query_key_value(x)
         all_head_size = qkv.shape[-1] // 3
         num_attention_heads = all_head_size // self.attention_head_size
@@ -138,7 +137,7 @@ class ViTSelfAttention(CheckpointModule):
 
 
 @LAYERS.register_module
-class ViTMLP(CheckpointModule):
+class ViTMLP(nn.Module):
     def __init__(self,
                  dim: int,
                  mlp_ratio: int,
@@ -146,9 +145,8 @@ class ViTMLP(CheckpointModule):
                  dropout: float,
                  dtype: dtype = None,
                  bias: bool = True,
-                 checkpoint: bool = False,
                  init_method: str = 'torch'):
-        super().__init__(checkpoint)
+        super().__init__()
         self.dense_1 = col_nn.Linear(dim,
                                      mlp_ratio * dim,
                                      dtype=dtype,
@@ -163,7 +161,7 @@ class ViTMLP(CheckpointModule):
                                      **_init_rules[init_method]['transformer'])
         self.dropout_2 = col_nn.Dropout(dropout)
 
-    def _forward(self, x):
+    def forward(self, x):
         x = self.dense_1(x)
         x = self.activation(x)
         x = self.dropout_1(x)
@@ -192,7 +190,7 @@ class ViTHead(nn.Module):
             self.representation = None
             representation_size = dim
 
-        self.linear = col_nn.Classifier(representation_size,
+        self.dense = col_nn.Classifier(representation_size,
                                        num_classes,
                                        dtype=dtype,
                                        bias=bias,
@@ -202,12 +200,12 @@ class ViTHead(nn.Module):
         x = x[:, 0]
         if self.representation is not None:
             x = self.representation(x)
-        x = self.linear(x)
+        x = self.dense(x)
         return x
 
 
 @LAYERS.register_module
-class ViTBlock(nn.Module):
+class ViTBlock(CheckpointModule):
     def __init__(self,
                  dim: int,
                  num_heads: int,
@@ -216,32 +214,31 @@ class ViTBlock(nn.Module):
                  attention_dropout: float = 0.,
                  dropout: float = 0.,
                  drop_path: float = 0.,
+                 layernorm_epsilon: float = 1e-6,
                  dtype: dtype = None,
                  bias: bool = True,
                  checkpoint: bool = False,
                  init_method: str = 'torch'):
-        super().__init__()
-        self.norm1 = col_nn.LayerNorm(normalized_shape=dim, eps=1e-6, dtype=dtype)
+        super().__init__(checkpoint)
+        self.norm1 = col_nn.LayerNorm(normalized_shape=dim, eps=layernorm_epsilon, dtype=dtype)
         self.attn = ViTSelfAttention(dim=dim,
                                      num_heads=num_heads,
                                      attention_dropout=attention_dropout,
                                      dropout=dropout,
                                      bias=bias,
                                      dtype=dtype,
-                                     checkpoint=checkpoint,
                                      init_method=init_method)
         self.drop_path = col_nn.DropPath(drop_path) if drop_path > 0. else nn.Identity()
-        self.norm2 = col_nn.LayerNorm(normalized_shape=dim, eps=1e-6, dtype=dtype)
+        self.norm2 = col_nn.LayerNorm(normalized_shape=dim, eps=layernorm_epsilon, dtype=dtype)
         self.mlp = ViTMLP(dim=dim,
                           mlp_ratio=mlp_ratio,
                           activation=activation,
                           dropout=dropout,
                           dtype=dtype,
                           bias=bias,
-                          checkpoint=checkpoint,
                           init_method=init_method)
 
-    def forward(self, x):
+    def _forward(self, x):
         x = x + self.drop_path(self.attn(self.norm1(x)))
         x = x + self.drop_path(self.mlp(self.norm2(x)))
         return x
@@ -261,6 +258,7 @@ class VisionTransformer(nn.Module):
                  attention_dropout: float = 0.,
                  dropout: float = 0.1,
                  drop_path: float = 0.,
+                 layernorm_epsilon: float = 1e-6,
                  activation: Callable = nn.functional.gelu,
                  representation_size: int = None,
                  dtype: dtype = None,
@@ -295,7 +293,7 @@ class VisionTransformer(nn.Module):
             ) for i in range(depth)
         ]
 
-        norm = col_nn.LayerNorm(normalized_shape=dim, eps=1e-6, dtype=dtype)
+        norm = col_nn.LayerNorm(normalized_shape=dim, eps=layernorm_epsilon, dtype=dtype)
 
         head = ViTHead(dim=dim,
                        num_classes=num_classes,