[Gemini] chunk init using runtime visited param order (#2115)

colossalai/gemini/chunk/search_utils.py

@@ -1,10 +1,10 @@
 import math
-from typing import Dict, List, Tuple
+from typing import Dict, List, Optional, Tuple
 
 import numpy as np
 import torch.nn as nn
 
-from colossalai.gemini.memory_tracer import OrderedParamGenerator
+from colossalai.gemini.memory_tracer import MemStats, OrderedParamGenerator
 from colossalai.tensor import ColoParameter
 
 
@@ -73,7 +73,8 @@ def search_chunk_configuration(
         search_range_mb: float,
         search_interval_byte: int,    # hidden size is the best value for the interval
         min_chunk_size_mb: float = 32,
-        filter_exlarge_params: bool = True) -> Tuple[Dict, int]:
+        filter_exlarge_params: bool = True,
+        memstas: Optional[MemStats] = None) -> Tuple[Dict, int]:
     """search_chunk_configuration
 
     Args:
@@ -86,9 +87,13 @@ def search_chunk_configuration(
         Tuple[Dict, int]: chunk config (a dict of dp_degree -> chunk init args) and its memory chunk waste in byte.
     """
 
-    param_order = OrderedParamGenerator()
+    if memstas is not None:
-    for p in model.parameters():
+        param_order = memstas.param_order()
-        param_order.append(p)
+    else:
+        # build the param visited order right now
+        param_order = OrderedParamGenerator()
+        for p in model.parameters():
+            param_order.append(p)
 
     search_range_byte = round(search_range_mb * 1024**2)
     min_chunk_size_byte = round(min_chunk_size_mb * 1024**2)

colossalai/gemini/chunk/utils.py

@@ -7,6 +7,7 @@ import torch.nn as nn
 
 from colossalai.gemini.chunk import ChunkManager
 from colossalai.gemini.chunk.search_utils import in_ddp, search_chunk_configuration
+from colossalai.gemini.memory_tracer import MemStats
 
 
 def init_chunk_manager(model: nn.Module,
@@ -37,13 +38,13 @@ def init_chunk_manager(model: nn.Module,
     total_size = sum(params_sizes) / 1024**2
 
     dist.barrier()
-    begine = time()
+    begin = time()
 
     config_dict, wasted_size = search_chunk_configuration(model, **kwargs_dict)
 
     dist.barrier()
     end = time()
-    span_s = end - begine
+    span_s = end - begin
     wasted_size /= 1024**2
 
     if dist.get_rank() == 0:

colossalai/gemini/gemini_mgr.py

@@ -25,6 +25,7 @@ class GeminiManager:
             If it's 'auto', they are moving dynamically based on CPU and CUDA memory usage. It will utilize heterogeneous memory space evenly and well.
             Note that 'auto' policy can only work well when no other processes use CUDA during your training.
         chunk_manager (ChunkManager): A ``ChunkManager`` instance.
+        memstats (MemStats, optional): a mem stats collected by a runtime mem tracer. if None then GeminiManager will collect it during a warmup iteration.
     """
 
     def __init__(self, placement_policy: str, chunk_manager: ChunkManager, memstats: Optional[MemStats] = None) -> None:
@@ -33,8 +34,11 @@ class GeminiManager:
         self.policy_name = placement_policy
         policy_cls = PlacementPolicyFactory.create(placement_policy)
         self._chunk_manager = chunk_manager
+
+        self._premade_memstats_ = memstats is not None
+        self._memstats = memstats
         self._mem_stats_collector = ChunkMemStatsCollector(chunk_manager,
-                                                           memstats) if policy_cls.need_mem_stats else None
+                                                           self._memstats) if policy_cls.need_mem_stats else None
         self._placement_policy = policy_cls(chunk_manager, self._mem_stats_collector)
         self._compute_list: List[Tuple[Chunk, ...]] = []
         self._compute_idx: int = -1
@@ -46,6 +50,19 @@ class GeminiManager:
         self._warmup = True
         self._comp_cuda_demand_time = 0
 
+    def memstats(self):
+        """memstats
+
+        get the memory statistics during training.
+        The stats could be collected by a runtime memory tracer, or collected by the GeminiManager.
+        Note, for the latter， you can not access the memstats before warmup iteration finishes.
+        """
+        if self._premade_memstats_:
+            return self._memstats
+        else:
+            assert not self._warmup, "Gemini Manager has memstats after warm up! Now is during warmup."
+            return self._mem_stats_collector._memstats
+
     def pre_iter(self, *args):
         if self._mem_stats_collector and self._warmup:
             self._mem_stats_collector.start_collection()

colossalai/gemini/memory_tracer/memory_stats.py

@@ -23,6 +23,12 @@ class MemStats(object):
 
         self._param_runtime_order = OrderedParamGenerator()
 
+    def param_order(self):
+        if self._param_runtime_order.is_empty():
+            raise RuntimeError
+        else:
+            return self._param_runtime_order
+
     def append_overall_data(self, device_type: str, val: float):
         if device_type == 'cuda':
             self._overall_cuda_list.append(val)

colossalai/gemini/memory_tracer/memstats_collector.py

@@ -37,7 +37,7 @@ class MemStatsCollector:
             self._memstats = MemStats()
 
     def next_period_non_model_data_usage(self, device_type: str) -> int:
-        """Get max non model data memory usage of current sampling period
+        """Maximum non model data memory usage during the next Op run
 
         Args:
             device_type (str): device type, can be 'cpu' or 'cuda'.
@@ -47,6 +47,9 @@ class MemStatsCollector:
         """
         assert not self._start_flag, 'Cannot get mem stats info during collection phase.'
         assert self._step_total > 0, 'Cannot get mem stats info before collection phase.'
+        assert len(self._memstats.non_model_data_list(device_type)) > self._step_idx, \
+            f"{len(self._memstats.non_model_data_list(device_type))} should be > than step idx {self._step_idx}, "\
+                f"step total {self._step_total}"
         next_non_model_data = self._memstats.non_model_data_list(device_type)[self._step_idx]
         self._step_idx = (self._step_idx + 1) % self._step_total
         return next_non_model_data
@@ -61,7 +64,8 @@ class MemStatsCollector:
 
     def finish_collection(self):
         self.sample_overall_data()
-        self._step_total = len(self._sampling_time)
+        # self._step_total = len(self._sampling_time)
+        self._step_total = len(self._memstats.non_model_data_list('cuda'))
         self._start_flag = False
         self._mem_monitor.finish()
 

colossalai/gemini/memory_tracer/param_runtime_order.py

@@ -35,5 +35,8 @@ class OrderedParamGenerator(ParamGenerator):
             visited_set.add(p)
         del visited_set
 
+    def is_empty(self):
+        return len(self.param_visited_order) > 0
+
     def clear(self):
         self.param_visited_order = []

colossalai/nn/parallel/data_parallel.py

@@ -8,6 +8,7 @@ import torch.distributed as dist
 
 from colossalai.gemini.chunk import Chunk, ChunkManager, TensorState
 from colossalai.gemini.gemini_mgr import GeminiManager
+from colossalai.gemini.memory_tracer import OrderedParamGenerator
 from colossalai.logging import get_dist_logger
 from colossalai.nn.parallel.utils import get_temp_total_chunk_on_cuda
 from colossalai.tensor import ProcessGroup as ColoProcessGroup
@@ -216,8 +217,18 @@ class ZeroDDP(ColoDDP):
         self.grads_device: Dict[torch.Tensor, torch.device] = {}
 
         cpu_offload = self.gemini_manager.policy_name != 'cuda'
-        # TODO: get param order and filter unused params
+
-        for p in module.parameters():
+        if self.gemini_manager._premade_memstats_:
+            # build chunk in param runtime visited order.
+            param_order = self.gemini_manager.memstats()._param_runtime_order
+        else:
+            # build chunk in param initialized order.
+            # Note: in this way, it can not get filter unused params during runtime.
+            param_order = OrderedParamGenerator()
+            for p in module.parameters():
+                param_order.append(p)
+
+        for p in param_order.generate():
             assert isinstance(p, ColoParameter)
 
             if getattr(p, '_ddp_to_ignore', False):
@@ -243,7 +254,7 @@ class ZeroDDP(ColoDDP):
         self.chunk_manager.close_all_groups()
         self._cast_buffers()
 
-        params_list = [p for p in module.parameters() if not getattr(p, '_ddp_to_ignore', False)]
+        params_list = [p for p in param_order.generate() if not getattr(p, '_ddp_to_ignore', False)]
         for p, fp32_p in zip(params_list, self.fp32_params):
             chunk_16 = self.chunk_manager.get_chunk(p)
             chunk_32 = self.chunk_manager.get_chunk(fp32_p)

colossalai/nn/parallel/gemini_parallel.py

@@ -4,6 +4,7 @@ import torch
 
 from colossalai.gemini.chunk import init_chunk_manager
 from colossalai.gemini.gemini_mgr import GeminiManager
+from colossalai.gemini.memory_tracer import MemStats
 
 from .data_parallel import ZeroDDP
 
@@ -18,7 +19,8 @@ class GeminiDDP(ZeroDDP):
                  force_outputs_fp32: bool = False,
                  search_range_mb: int = 32,
                  hidden_dim: Optional[int] = None,
-                 min_chunk_size_mb: Optional[float] = None) -> None:
+                 min_chunk_size_mb: Optional[float] = None,
+                 memstats: Optional[MemStats] = None) -> None:
         """
         A torch.Module warpper using ZeRO-DP and Genimi.
         ZeRO is for parallel. Gemini is for memory management.
@@ -44,11 +46,12 @@ class GeminiDDP(ZeroDDP):
             min_chunk_size_mb (float, optional): the minimum chunk size in MegaByte.
                 If the aggregate size of parameters is still samller than the minimum chunk size,
                 all parameters will be compacted into one small chunk.
+            memstats (MemStats, optional) the memory statistics collector by a runtime memory tracer.
         """
         chunk_manager = init_chunk_manager(model=module,
                                            init_device=device,
                                            hidden_dim=hidden_dim,
                                            search_range_mb=search_range_mb,
                                            min_chunk_size_mb=min_chunk_size_mb)
-        gemini_manager = GeminiManager(placement_policy, chunk_manager)
+        gemini_manager = GeminiManager(placement_policy, chunk_manager, memstats)
         super().__init__(module, gemini_manager, pin_memory, force_outputs_fp32)

tests/test_gemini/update/test_gemini_use_rmt.py

@@ -8,7 +8,8 @@ import colossalai
 from colossalai.gemini.chunk import ChunkManager, search_chunk_configuration
 from colossalai.gemini.gemini_mgr import GeminiManager
 from colossalai.gemini.memory_tracer.runtime_mem_tracer import RuntimeMemTracer
-from colossalai.nn.parallel import ZeroDDP
+from colossalai.nn.optimizer.gemini_optimizer import GeminiAdamOptimizer
+from colossalai.nn.parallel import GeminiDDP, ZeroDDP
 from colossalai.tensor import ProcessGroup
 from colossalai.testing import parameterize, rerun_if_address_is_in_use
 from colossalai.utils import free_port
@@ -44,29 +45,27 @@ def run_gemini_use_rmt(placement_policy, keep_gather, model_name: str, use_grad_
             run_fwd_bwd(runtime_mem_tracer, input_ids, label, criterion, runtime_mem_tracer)
     memstats = runtime_mem_tracer.memstats()
     runtime_tracer_non_model_data = runtime_mem_tracer._memstats._non_model_data_cuda_list
-    print('runtime tracer: ', runtime_tracer_non_model_data)
+    print('runtime tracer non model data points: ', len(runtime_tracer_non_model_data))
 
-    world_size = torch.distributed.get_world_size()
+    model = GeminiDDP(model, device='cuda', placement_policy=placement_policy, search_range_mb=1, memstats=memstats)
-    config_dict, _ = search_chunk_configuration(model, search_range_mb=1, search_interval_byte=100)
+    zero_optim = GeminiAdamOptimizer(model, lr=1e-3, initial_scale=1)
-    config_dict[world_size]['chunk_size'] = 5000
-    config_dict[world_size]['keep_gathered'] = keep_gather
-    chunk_manager = ChunkManager(config_dict)
-    gemini_manager = GeminiManager(placement_policy, chunk_manager, memstats)
-    model = ZeroDDP(model, gemini_manager, pin_memory=True)
 
     pg = ProcessGroup()
     set_seed(pg.dp_local_rank())
     for i, (input_ids, label) in enumerate(train_dataloader):
         # you can only test a single fwd + bwd.
         # after bwd param is grad for Gemini, due to the chunk reuse optimization.
-        if i > 1:
+        # print(f'iteration {i}')
+        if i > 4:
             break
         input_ids, label = input_ids.cuda(), label.cuda()
 
+        zero_optim.zero_grad()
         set_seed(42)
-        loss = run_fwd_bwd(model, input_ids, label, criterion, model)
+        loss = run_fwd_bwd(model, input_ids, label, criterion, zero_optim)
+        zero_optim.step()
 
-    gemini_non_model_data = gemini_manager._mem_stats_collector._memstats.non_model_data_list('cuda')
+    gemini_non_model_data = model.gemini_manager._mem_stats_collector._memstats.non_model_data_list('cuda')
 
     # print('gemini non model data:', gemini_non_model_data)
 

tests/test_gemini/update/test_optim.py

@@ -1,5 +1,4 @@
 from functools import partial
-from time import time
 
 import pytest
 import torch