[utils] Impl clip_grad_norm for ColoTensor and ZeroOptimizer (#1442)

colossalai/utils/common.py

@@ -1,11 +1,13 @@
 #!/usr/bin/env python
 # -*- encoding: utf-8 -*-
 import os
+from pprint import pp
 import random
 import socket
 from pathlib import Path
-from typing import Callable, List, Union
+from typing import Callable, List, Union, Dict, Optional
 import functools
+
 import torch
 from torch._six import inf
 from torch.nn.parameter import Parameter
@@ -22,9 +24,11 @@ from colossalai.constants import (IS_TENSOR_PARALLEL, NUM_PARTITIONS, TENSOR_PAR
 from colossalai.context.parallel_mode import ParallelMode
 from colossalai.core import global_context as gpc
 from colossalai.global_variables import tensor_parallel_env as env
-
 from .multi_tensor_apply import multi_tensor_applier
 
+from colossalai.tensor import ColoParameter, ProcessGroup
+from collections import defaultdict
+
 
 def print_rank_0(msg: str, logger=None):
     """Print messages and save logs(optional). This is executed only if you are the rank-0 gpu.
@@ -162,6 +166,121 @@ def _get_tensor_norm(norm: Union[float, torch.Tensor], move_to_cuda) -> torch.Te
 # ======== Gradient Clipping =========
 
 
+def _compute_local_lp(params: List[ColoParameter], norm_type: float) -> float:
+    if len(params) == 0:
+        return 0.0
+    grads = [p.grad for p in params]
+    use_cuda_kernel = grads[0].device.type == 'cuda'
+    if norm_type == inf:
+        local_lp = max([g.abs().max() for g in grads])
+    elif norm_type == 2.0 and use_cuda_kernel:
+        local_lp = _calc_l2_norm(grads)**norm_type
+    else:
+        local_lp = _calc_lp(grads, norm_type)
+    if isinstance(local_lp, torch.Tensor):
+        return local_lp.item()
+    return local_lp
+
+
+def _compute_buckets_lp(params: List[ColoParameter], norm_type: float) -> float:
+    if len(params) == 0:
+        return 0.0
+    buckets: Dict[Optional[ProcessGroup], List[ColoParameter]] = defaultdict(list)
+    for p in params:
+        if p.is_replicate():
+            buckets[None].append(p)
+        else:
+            buckets[p.get_process_group().tp_process_group()].append(p)
+    total_lp = 0.0
+    for group, bucket in buckets.items():
+        local_lp = _compute_local_lp(bucket, norm_type)
+        if group is not None:
+            local_lp_tensor = torch.tensor([local_lp], device=torch.cuda.current_device())
+            if norm_type == inf:
+                dist.all_reduce(local_lp_tensor, op=dist.ReduceOp.MAX, group=group)
+            else:
+                dist.all_reduce(local_lp_tensor, group=group)
+            local_lp = local_lp_tensor.item()
+        if norm_type == inf:
+            total_lp = max(total_lp, local_lp)
+        else:
+            total_lp += local_lp
+    return total_lp
+
+
+def _compute_pp_grad_lp(total_lp: float, norm_type: float) -> float:
+    if gpc.is_initialized(ParallelMode.PIPELINE) and gpc.get_world_size(ParallelMode.PIPELINE) > 1:
+        total_lp_tensor = torch.tensor([total_lp], device=torch.cuda.current_device())
+        if norm_type == inf:
+            dist.all_reduce(total_lp_tensor, op=dist.ReduceOp.MAX, group=gpc.get_group(ParallelMode.PIPELINE))
+        else:
+            dist.all_reduce(total_lp_tensor, group=gpc.get_group(ParallelMode.PIPELINE))
+        total_lp = total_lp_tensor.item()
+    return total_lp
+
+
+def _compute_grad_lp(parameters, norm_type: float = 2.0) -> float:
+    if isinstance(parameters, torch.Tensor):
+        parameters = [parameters]
+    grad_dtype = None
+    cpu_grad_params: List[ColoParameter] = []
+    cuda_grad_params: List[ColoParameter] = []
+    for p in parameters:
+        if p.grad is None:
+            continue
+        assert isinstance(p, ColoParameter)
+        if grad_dtype is None:
+            grad_dtype = p.grad.dtype
+        assert p.grad.dtype == grad_dtype, f'Expected all grads are {grad_dtype}, got {p.grad.dtype}'
+        if p.grad.device.type == 'cuda':
+            cuda_grad_params.append(p)
+        else:
+            cpu_grad_params.append(p)
+    norm_type = float(norm_type)
+    cpu_lp = _compute_buckets_lp(cpu_grad_params, norm_type)
+    cuda_lp = _compute_buckets_lp(cuda_grad_params, norm_type)
+    if norm_type == inf:
+        total_lp = max(cpu_lp, cuda_lp)
+    else:
+        total_lp = cpu_lp + cuda_lp
+    return _compute_pp_grad_lp(total_lp, norm_type)
+
+
+def compute_grad_norm(parameters, norm_type: float = 2.0) -> float:
+    norm_type = float(norm_type)
+    total_norm = _compute_grad_lp(parameters, norm_type)
+    if norm_type != inf:
+        total_norm = total_norm**(1 / norm_type)
+    return total_norm
+
+
+def _clip_grad_norm(parameters, max_norm: float, total_norm: float) -> None:
+    clip_coef = max_norm / (total_norm + 1e-6)
+    if clip_coef < 1.0:
+        cuda_grads: List[torch.Tensor] = []
+        cpu_grads: List[torch.Tensor] = []
+        if isinstance(parameters, torch.Tensor):
+            parameters = [parameters]
+        for p in parameters:
+            if p.grad is None:
+                continue
+            if p.grad.device.type == 'cuda':
+                cuda_grads.append(p.grad.detach())
+            else:
+                cpu_grads.append(p.grad.detach())
+        if len(cuda_grads) > 0:
+            dummy_overflow_buf = torch.cuda.IntTensor([0])
+            multi_tensor_applier(colossal_C.multi_tensor_scale, dummy_overflow_buf, [cuda_grads, cuda_grads], clip_coef)
+        for g in cpu_grads:
+            g.mul_(clip_coef)
+
+
+def clip_grad_norm(parameters, max_norm: float, norm_type: float = 2.0) -> float:
+    total_norm = compute_grad_norm(parameters, norm_type)
+    _clip_grad_norm(parameters, max_norm, total_norm)
+    return total_norm
+
+
 def clip_grad_norm_fp32(parameters, max_norm, norm_type=2):
     """Clips gradient norm of an iterable of parameters whose gradients are in fp32.
 

colossalai/zero/zero_optimizer.py

@@ -8,9 +8,11 @@ from colossalai.amp.naive_amp.grad_scaler import DynamicGradScaler
 from colossalai.logging import get_dist_logger
 from colossalai.nn.optimizer import ColossalaiOptimizer
 from colossalai.utils import get_current_device, disposable
+from colossalai.utils.common import _compute_grad_lp, compute_grad_norm, _clip_grad_norm
 from collections import defaultdict, abc as container_abcs
 from copy import deepcopy
 from itertools import chain
+from torch._six import inf
 
 
 class OptimState(Enum):
@@ -143,11 +145,38 @@ class ZeroOptimizer(ColossalaiOptimizer):
         self._update_fp16_params()
         return ret
 
-    def clip_grad_norm(self, model: torch.nn.Module, max_norm: float):
+    def compute_grad_norm(self, norm_type: float = 2.0) -> float:
+        norm_type = float(norm_type)
+        if not self.chunk_manager.enable_distributed_storage:
+            return compute_grad_norm(self.module.parameters(), norm_type)
+
+        non_distributed_params = []
+        distributed_params = []
+        for p in self.module.parameters():
+            if getattr(p, '_ddp_to_ignore', False):
+                non_distributed_params.append(p)
+            else:
+                distributed_params.append(p)
+        non_distributed_norm = _compute_grad_lp(non_distributed_params, norm_type)
+        distributed_norm_tensor = torch.tensor([_compute_grad_lp(distributed_params, norm_type)],
+                                               device=get_current_device())
+        if norm_type == inf:
+            dist.all_reduce(distributed_norm_tensor,
+                            op=dist.ReduceOp.MAX,
+                            group=self.chunk_manager.process_group.dp_process_group())
+            total_norm = max(non_distributed_norm, distributed_norm_tensor.item())
+        else:
+            dist.all_reduce(distributed_norm_tensor, group=self.chunk_manager.process_group.dp_process_group())
+            total_norm = non_distributed_norm + distributed_norm_tensor.item()
+            total_norm = total_norm**(1 / norm_type)
+        return total_norm
+
+    def clip_grad_norm(self, model: torch.nn.Module, max_norm: float, norm_type: float = 2.0):
         if self.optim_state == OptimState.SCALED:
             self._unscale_grads()
-        # TODO(ver217): fix zero clip grad norm
-        return super().clip_grad_norm(model, max_norm)
+        total_norm = self.compute_grad_norm(norm_type)
+        _clip_grad_norm(self.module.parameters(), max_norm, total_norm)
+        return total_norm
 
     def backward(self, loss: torch.Tensor):
         loss = self.loss_scale * loss

tests/test_utils/test_norm_gradient_clipping.py

@@ -0,0 +1,79 @@
+from colossalai.tensor import distspec, ColoTensorSpec, ProcessGroup
+from colossalai.tensor.colo_parameter import ColoParameter
+import colossalai
+import pytest
+import torch
+import torch.multiprocessing as mp
+from colossalai.logging import disable_existing_loggers
+from colossalai.utils import free_port, get_current_device
+from torch.nn.utils import clip_grad_norm_
+from functools import partial
+from colossalai.testing import parameterize, rerun_if_address_is_in_use
+from colossalai.utils.common import clip_grad_norm
+from torch.nn.parameter import Parameter
+
+
+def close(num: float, other: float, rtol: float = 1e-5, atol: float = 1e-8):
+    return abs(num - other) <= atol + rtol * other
+
+
+def shard_param(p: ColoParameter) -> None:
+    pg = p.get_process_group()
+    p._redistribute(distspec.shard([0], [pg.tp_world_size()]))
+    p.grad = p.grad.chunk(pg.tp_world_size(), 0)[pg.tp_local_rank()].clone().detach()
+
+
+def check_grad_equal(p: Parameter, colo_p: ColoParameter) -> None:
+    pg = colo_p.get_process_group()
+    if p.shape != colo_p.shape:
+        grad = p.grad.chunk(pg.tp_world_size(), 0)[pg.tp_local_rank()]
+    else:
+        grad = p.grad
+    assert torch.allclose(grad, colo_p.grad), f'diff: {torch.abs(grad - colo_p.grad)}'
+
+
+@parameterize('dtype', [torch.float])
+@parameterize('device', ['mixed', 'cuda', 'cpu'])
+@parameterize('norm_type', [2.0, 3.0, float('inf')])
+def run_grad_clip_norm(world_size: int, dtype: torch.dtype, device: str, norm_type: float):
+    print(f'{world_size}, {dtype}, {device}, {norm_type}')
+    cuda_device = get_current_device()
+    devices = [cuda_device] * 4
+    if device == 'cpu':
+        devices = [torch.device('cpu')] * 4
+    elif device == 'mixed':
+        devices = [cuda_device] * 2 + [torch.device('cpu')] * 2
+    pg = ProcessGroup(tp_degree=world_size)
+    params = [Parameter(torch.empty(4, 4, dtype=dtype, device=devices[i])) for i in range(4)]
+    colo_params = [
+        ColoParameter(torch.empty(4, 4, dtype=dtype, device=devices[i]), spec=ColoTensorSpec(pg)) for i in range(4)
+    ]
+    for p, colo_p in zip(params, colo_params):
+        grad = torch.rand_like(p)
+        p.grad = grad
+        colo_p.grad = grad.clone().detach()
+    shard_param(colo_params[0])
+    shard_param(colo_params[2])
+    torch_norm = clip_grad_norm_(params, 1.0, norm_type=norm_type)
+    colo_norm = clip_grad_norm(colo_params, 1.0, norm_type=norm_type)
+    assert close(torch_norm, colo_norm), f'diff: {abs(torch_norm-colo_norm)}'
+    for p, colo_p in zip(params, colo_params):
+        check_grad_equal(p, colo_p)
+
+
+def run_dist(rank, world_size, port):
+    disable_existing_loggers()
+    colossalai.launch(config={}, rank=rank, world_size=world_size, host='localhost', port=port, backend='nccl')
+    run_grad_clip_norm(world_size=world_size)
+
+
+@pytest.mark.dist
+@pytest.mark.parametrize('world_size', [1, 2])
+@rerun_if_address_is_in_use()
+def test_zero_clip_grad(world_size: int):
+    run_func = partial(run_dist, world_size=world_size, port=free_port())
+    mp.spawn(run_func, nprocs=world_size)
+
+
+if __name__ == '__main__':
+    test_zero_clip_grad(2)