[example] add benchmark (#2276)

* add benchmark

* merge common func

* add total and avg tflops

Co-authored-by: Ziyue Jiang <ziyue.jiang@gmail.com>

colossalai/pipeline/rpc/_pipeline_base.py

@@ -240,6 +240,10 @@ class WorkerBase(ABC):
             output = [output[i] for i in offsets]
         return output
 
+    def get_numels(self) -> int:
+        numel = sum(param.numel() for param in self.module_partition.parameters())
+        return numel
+
     def get_parameters(self) -> List[torch.Tensor]:
         return [p for p in self.module_partition.parameters()]
 
@@ -1115,6 +1119,15 @@ class PipelineEngineBase(ABC, nn.Module):
         for fut in sync_futs:
             fut.wait()
 
+    def remote_numels(self) -> Dict[int, int]:
+        numels = {}
+        actual_stage_num = self._get_actual_stage_num()
+        for stage_id in range(actual_stage_num):
+            worker_rref = self.pp_rank_to_worker_rref[stage_id]
+            numel = worker_rref.rpc_sync().get_numels()
+            numels[stage_id] = numel
+        return numels
+
     def remote_parameters(self) -> Dict[int, List[torch.Tensor]]:
         parameters = {}
         actual_stage_num = self._get_actual_stage_num()

examples/language/gpt/train_gpt_demo.py

@@ -8,6 +8,7 @@ import torch.nn as nn
 from model_zoo import model_builder
 from packaging import version
 from torch.nn.parallel import DistributedDataParallel as DDP
+from utils import get_data, get_tflops
 
 import colossalai
 from colossalai.logging import disable_existing_loggers, get_dist_logger
@@ -95,13 +96,6 @@ 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)
-    return input_ids, attention_mask
-
-
 def get_cpu_mem():
     return psutil.Process().memory_info().rss / 1024**2
 
@@ -114,10 +108,6 @@ def get_mem_info(prefix=''):
     return f'{prefix}GPU memory usage: {get_gpu_mem():.2f} MB, CPU memory usage: {get_cpu_mem():.2f} MB'
 
 
-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 get_model_size(model: nn.Module):
     total_numel = 0
     for module in model.modules():

examples/language/gpt/train_gpt_pp_demo.py

@@ -6,6 +6,7 @@ import torch
 from model_zoo import model_builder
 from torch import nn
 from tqdm import tqdm
+from utils import get_data, get_tflops
 
 from colossalai.fx import ColoTracer
 from colossalai.fx.passes.adding_split_node_pass import avgnode_split_pass, split_with_split_nodes_pass
@@ -26,7 +27,7 @@ def parse_args():
     parser.add_argument('--num_microbatches', type=int, default=2)
     parser.add_argument('--device', type=str, choices=['cpu', 'cuda'], default='cuda')
     parser.add_argument('--master_addr', type=str, default='localhost')
-    parser.add_argument('--master_port', type=str, default='29020')
+    parser.add_argument('--master_port', type=str, default='29011')
     parser.add_argument('--num_worker_threads', type=int, default=128)
     return parser.parse_args()
 
@@ -66,12 +67,10 @@ def create_partition_module(pp_rank: int, stage_num: int, model, data_kwargs):
     return split_submodules[pp_rank + 1]
 
 
-def partition(logger, model_type, data_kwargs, pp_rank: int, chunk: int, stage_num: int):
+def partition(model_type, data_kwargs, pp_rank: int, chunk: int, stage_num: int):
     # build model
     model = model_builder(model_type)(checkpoint=False)
     module = create_partition_module(pp_rank, stage_num, model, data_kwargs)
-    num_params = sum(param.numel() for param in module.parameters())
-    logger.info(f'{pp_rank=} number of args in this partition:{num_params}')
     return module
 
 
@@ -86,6 +85,7 @@ def run_master(args):
     SEQ_LEN = 1024
     VOCAB_SIZE = 50257
     NUM_STEPS = 10
+    WARMUP_STEPS = 1
 
     disable_existing_loggers()
     logger = get_dist_logger()
@@ -102,7 +102,7 @@ def run_master(args):
     warmup_data_kwargs = {'input_ids': input_ids, 'attention_mask': attn_mask}
 
     # set 1f1b pipeline engine
-    pp_engine = OneFOneBPipelineEngine(partition_fn=partial(partition, logger, model_type, warmup_data_kwargs),
+    pp_engine = OneFOneBPipelineEngine(partition_fn=partial(partition, model_type, warmup_data_kwargs),
                                        stage_num=stage_num,
                                        num_microbatches=num_microbatches,
                                        device=device,
@@ -111,21 +111,45 @@ def run_master(args):
                                        metric=None,
                                        checkpoint=False)
 
+    partition_numels = pp_engine.remote_numels()
+    for rank, numel in partition_numels.items():
+        logger.info(f'{rank=} numel in the partition:{numel}')
+
     # build optim
     pp_engine.initialize_optimizer(HybridAdam, lr=1e-3)
 
-    times = []
-    for n in tqdm(range(NUM_STEPS)):
+    ranks_tflops = {}
+    for n in range(NUM_STEPS):
         # we just use randomly generated data here
         input_ids, attn_mask = get_data(batch_size, SEQ_LEN, VOCAB_SIZE)
         batch = {'input_ids': input_ids, 'attention_mask': attn_mask}
 
         start = time.time()
         outputs = pp_engine.forward_backward(batch=batch, labels=input_ids, forward_only=False)
-        cost_time = time.time() - start
-        times.append(cost_time)
+        step_time = time.time() - start
 
-    logger.info("avg cost time : {}s".format(sum(times) / len(times)))
+        for rank, numel in partition_numels.items():
+            if rank not in ranks_tflops:
+                ranks_tflops[rank] = []
+            step_tflops = get_tflops(numel, batch_size, SEQ_LEN, step_time)
+
+            logger.info(
+                f"Rank{rank} , [{n + 1}/{NUM_STEPS}] , Step time: {step_time:.3f}s, TFLOPS: {get_tflops(numel, batch_size, SEQ_LEN, step_time):.3f}",
+                ranks=[0],
+            )
+
+            if n >= WARMUP_STEPS:
+                ranks_tflops[rank].append(step_tflops)
+
+    median_index = ((NUM_STEPS - WARMUP_STEPS) >> 1) + WARMUP_STEPS
+    gpu_tflops = []
+    for rank, tflops_list in ranks_tflops.items():
+        tflops_list.sort()
+        gpu_tflops.append(tflops_list[median_index])
+        logger.info(f"GPU{rank} Median TFLOPS is {tflops_list[median_index]:.3f}")
+
+    logger.info(f"Total TFLOPS is {sum(gpu_tflops):.3f}")
+    logger.info(f"Avg TFLOPS per GPU is {sum(gpu_tflops) / world_size:.3f}")
 
 
 if __name__ == '__main__':

examples/language/gpt/utils.py

@@ -0,0 +1,12 @@
+import torch
+
+
+# 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)
+    return input_ids, attention_mask
+
+
+def get_tflops(model_numel, batch_size, seq_len, step_time):
+    return model_numel * batch_size * seq_len * 8 / 1e12 / (step_time + 1e-12)