[example] add TP to GPT example (#1828)

examples/language/gpt/run.sh

@@ -1 +1 @@
-env OMP_NUM_THREADS=16 torchrun --standalone --nproc_per_node=2 train_gpt_demo.py 2>&1 | tee run.log
+env OMP_NUM_THREADS=16 torchrun --standalone --nproc_per_node=4 train_gpt_demo.py --tp_degree=2 --placement='cpu' 2>&1 | tee run.log

examples/language/gpt/train_gpt_demo.py

@@ -10,13 +10,48 @@ import colossalai
 from colossalai.logging import disable_existing_loggers, get_dist_logger
 from colossalai.nn.optimizer import HybridAdam
 from colossalai.nn.parallel import ZeroDDP
-from colossalai.tensor import ProcessGroup
+from colossalai.tensor import ColoParameter, ComputePattern, ComputeSpec, ProcessGroup, ShardSpec
 from colossalai.utils import get_current_device
 from colossalai.utils.model.colo_init_context import ColoInitContext
 from colossalai.zero import ZeroOptimizer
 from transformers import GPT2Config, GPT2LMHeadModel
 
 
+def parse_args():
+    parser = colossalai.get_default_parser()
+    parser.add_argument(
+        "--tp_degree",
+        type=int,
+        default=1,
+        help="Tensor Parallelism Degree.",
+    )
+    parser.add_argument(
+        "--placement",
+        type=str,
+        default='cpu',
+        help="Placement Policy for Gemini.",
+    )
+    args = parser.parse_args()
+    return args
+
+
+## Parameter Sharding Strategies for Tensor Parallelism
+def split_param_single_dim_tp1d(dim: int, param: ColoParameter, pg: ProcessGroup):
+    spec = (ShardSpec([dim], [pg.tp_world_size()]), ComputeSpec(ComputePattern.TP1D))
+    if param.process_group.tp_world_size() == 1:
+        param.set_process_group(pg)
+    param.set_tensor_spec(*spec)
+
+
+def split_param_row_tp1d(param: ColoParameter, pg: ProcessGroup):
+    split_param_single_dim_tp1d(0, param, pg)
+
+
+def split_param_col_tp1d(param: ColoParameter, pg: ProcessGroup):
+    split_param_single_dim_tp1d(-1, param, pg)
+
+
+## Define the Model and Loss Based on Huggingface transformers GPT2LMHeadModel
 class GPTLMModel(nn.Module):
 
     def __init__(self,
@@ -56,6 +91,7 @@ class GPTLMLoss(nn.Module):
         return self.loss_fn(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
 
 
+## Randomly Generated Data
 def get_data(batch_size, seq_len, vocab_size):
     input_ids = torch.randint(0, vocab_size, (batch_size, seq_len), device=torch.cuda.current_device())
     attention_mask = torch.ones_like(input_ids)
@@ -90,54 +126,96 @@ def get_tflops(model_numel, batch_size, seq_len, step_time):
     return model_numel * batch_size * seq_len * 8 / 1e12 / (step_time + 1e-12)
 
 
-def main():
+# Tensor Parallel
-    BATCH_SIZE = 8
+def tensor_parallelize(model: torch.nn.Module, pg: ProcessGroup):
-    SEQ_LEN = 1024
+    """tensor_parallelize
-    VOCAB_SIZE = 50257
+    Sharding the Model Parameters.
-    NUM_STEPS = 10
-    PLACEMENT_POLICY = 'auto'
-    disable_existing_loggers()
-    colossalai.launch_from_torch(config={})
-    pg = ProcessGroup()
-    logger = get_dist_logger()
 
-    logger.info(get_mem_info(), ranks=[0])
+    Args:
-    # build GPT model
+        model (torch.nn.Module): a torch module to be sharded
-    with ColoInitContext(device=get_current_device()):
+    """
-        model = gpt2_medium(checkpoint=True)
+    for mn, module in model.named_modules():
-    numel = sum([p.numel() for p in model.parameters()])
+        for pn, param in module.named_parameters(recurse=False):
-    logger.info(f'Model numel: {numel}', ranks=[0])
+            # set process group for all parameters
-    get_tflops_func = partial(get_tflops, numel, BATCH_SIZE, SEQ_LEN)
+            param.set_process_group(pg)
 
+            if 'mlp.c_fc' in mn:
+                if 'weight' in pn or 'bias' in pn:
+                    split_param_col_tp1d(param, pg)    # colmn slice
+                    # keep the shape of the output from c_fc
+                    param.compute_spec.set_output_replicate(False)
+            elif 'mlp.c_proj' in mn:
+                if 'weight' in pn:
+                    split_param_row_tp1d(param, pg)    # row slice
+            elif 'wte' in mn or 'wpe' in mn:
+                split_param_col_tp1d(param, pg)    # colmn slice
+            elif 'c_attn' in mn or 'c_proj' in mn:
+                split_param_col_tp1d(param, pg)    # colmn slice
+
+
+# Gemini + ZeRO DDP
+def gemini_zero_dpp(model: torch.nn.Module, pg: ProcessGroup, placememt_policy: str = "auto"):
     cai_version = colossalai.__version__
-    logger.info(f'using Colossal-AI version {cai_version}')
     if version.parse(cai_version) > version.parse("0.1.10"):
         from colossalai.nn.parallel import GeminiDDP
         model = GeminiDDP(model,
                           device=get_current_device(),
-                          placement_policy=PLACEMENT_POLICY,
+                          placement_policy=placememt_policy,
                           pin_memory=True,
                           search_range_mb=32)
     elif version.parse(cai_version) <= version.parse("0.1.10") and version.parse(cai_version) >= version.parse("0.1.9"):
         from colossalai.gemini import ChunkManager, GeminiManager
         chunk_size = ChunkManager.search_chunk_size(model, 64 * 1024**2, 32)
-        gemini_manager = GeminiManager(PLACEMENT_POLICY, chunk_manager)
+        gemini_manager = GeminiManager(placememt_policy, chunk_manager)
         chunk_manager = ChunkManager(chunk_size,
                                      pg,
                                      enable_distributed_storage=True,
-                                     init_device=GeminiManager.get_default_device(PLACEMENT_POLICY))
+                                     init_device=GeminiManager.get_default_device(placememt_policy))
         model = ZeroDDP(model, gemini_manager)
+    else:
+        raise NotImplemented(f"CAI version {cai_version} is not supported")
+    return model
 
+
+def main():
+    args = parse_args()
+
+    BATCH_SIZE = 8
+    SEQ_LEN = 1024
+    VOCAB_SIZE = 50257
+    NUM_STEPS = 10
+
+    disable_existing_loggers()
+    colossalai.launch_from_torch(config={})
+
+    pg = ProcessGroup(tp_degree=args.tp_degree)
+
+    logger = get_dist_logger()
+    logger.info(get_mem_info(), ranks=[0])
+
+    # build GPT model
+    with ColoInitContext(device=get_current_device()):
+        model = gpt2_medium(checkpoint=True)
+
+    numel = sum([p.numel() for p in model.parameters()])
+    logger.info(f'Model numel: {numel}', ranks=[0])
+    get_tflops_func = partial(get_tflops, numel, BATCH_SIZE, SEQ_LEN)
+
+    # Tensor Parallelism (TP)
+    tensor_parallelize(model, pg)
+    # Gemini + ZeRO DP, Note it must be used after TP
+    model = gemini_zero_dpp(model, pg, args.placement)
     logger.info(get_mem_info(prefix='After init model, '), ranks=[0])
 
     # build criterion
     criterion = GPTLMLoss()
 
-    # optimizer
+    # build optimizer
     optimizer = HybridAdam(model.parameters(), lr=1e-3)
     optimizer = ZeroOptimizer(optimizer, model, initial_scale=2**5)
     logger.info(get_mem_info(prefix='After init optim, '), ranks=[0])
 
+    torch.cuda.synchronize()
     model.train()
     for n in range(NUM_STEPS):
         # we just use randomly generated data here
@@ -156,6 +234,8 @@ def main():
             f'[{n+1}/{NUM_STEPS}] Loss:{loss.item():.3f}, Step time: {step_time:.3f}s, TFLOPS: {get_tflops_func(step_time):.3f}',
             ranks=[0])
 
+    torch.cuda.synchronize()
+
 
 if __name__ == '__main__':
     main()

examples/language/opt/run_clm.py

@@ -36,7 +36,6 @@ from datasets import load_dataset
 from packaging import version
 from torch.utils.data import DataLoader
 from tqdm.auto import tqdm
-from utils import colo_memory_cap
 
 import colossalai
 import transformers
@@ -47,7 +46,6 @@ from colossalai.nn.optimizer import HybridAdam
 from colossalai.nn.parallel import ZeroDDP
 from colossalai.tensor import ProcessGroup
 from colossalai.utils import get_current_device, get_dataloader
-from colossalai.utils.checkpoint import load_checkpoint, save_checkpoint
 from colossalai.utils.model.colo_init_context import ColoInitContext
 from colossalai.zero import ZeroOptimizer
 from transformers import (
@@ -249,12 +247,20 @@ def parse_args():
     return args
 
 
+def colo_memory_cap(size_in_GB):
+    from colossalai.utils import colo_device_memory_capacity, colo_set_process_memory_fraction, get_current_device
+    cuda_capacity = colo_device_memory_capacity(get_current_device())
+    if size_in_GB * (1024**3) < cuda_capacity:
+        colo_set_process_memory_fraction(size_in_GB * (1024**3) / cuda_capacity)
+        print("Using {} GB of GPU memory".format(size_in_GB))
+
+
 def main():
     args = parse_args()
     disable_existing_loggers()
     colossalai.launch_from_torch(config=dict())
     logger = get_dist_logger()
-    is_main_process = gpc.get_local_rank(ParallelMode.DATA) == 0
+    is_main_process = dist.get_rank() == 0
 
     if is_main_process:
         datasets.utils.logging.set_verbosity_warning()

examples/language/opt/utils.py

@@ -1,28 +0,0 @@
-import torch
-import torch.distributed as dist
-
-
-def memory_cap(size_in_GB):
-    print(f"use only {size_in_GB} GB of CUDA memory")
-    assert dist.is_initialized(), "memory_cap must be used after dist init"
-    local_rank = dist.get_rank()
-    cuda_capacity = torch.cuda.get_device_properties(local_rank).total_memory
-    size_in_B = (size_in_GB * 1024**3)
-    if size_in_B > cuda_capacity:
-        print(f'memory_cap is uselsess since {cuda_capacity / 1024**3} less than {size_in_GB}')
-        return
-    fraction = (size_in_GB * 1024**3) / cuda_capacity
-    print(f'mem faction is {fraction}')
-    torch.cuda.set_per_process_memory_fraction(fraction, local_rank)
-
-
-def colo_memory_cap(size_in_GB):
-    from colossalai.utils import colo_device_memory_capacity, colo_set_process_memory_fraction, get_current_device
-    cuda_capacity = colo_device_memory_capacity(get_current_device())
-    if size_in_GB * (1024**3) < cuda_capacity:
-        colo_set_process_memory_fraction(size_in_GB * (1024**3) / cuda_capacity)
-        print("Using {} GB of GPU memory".format(size_in_GB))
-
-
-if __name__ == '__main__':
-    memory_cap(40)