Closed
Description
🐛 Describe the bug
If I torch.compile torch.amp.GradScaler, it works. But if I copy paste grad_scaler.py and import GradScaler from there, I receive an error.
To reproduce (testcase taken from here):
import torch
N, D_in, D_out = 64, 1024, 16
x = torch.randn(N, D_in, device='cuda')
y = torch.randn(N, D_out, device='cuda')
model = torch.nn.Linear(D_in, D_out).cuda()
optimizer = torch.optim.SGD(model.parameters(), lr=1e-2)
loss_fn = torch.nn.MSELoss()
from torch.amp import GradScaler
# from gradscaler2 import GradScaler
scaler = GradScaler()
@torch.compile
def run_fwd_bwd():
with torch.amp.autocast('cuda'):
y_pred = model(x)
loss = loss_fn(y_pred, y)
scaler.scale(loss).backward()
scaler.step(optimizer)
optimizer.zero_grad(set_to_none=True)
scaler.update()
for t in range(20):
run_fwd_bwd()
The above code will run fine.
If you swap the GradScaler import to:
# from torch.amp import GradScaler
from gradscaler2 import GradScaler
and copypaste https://raw.githubusercontent.com/pytorch/pytorch/refs/heads/main/torch/amp/grad_scaler.py into the local file gradscaler2.py, then it will fail, with the following error:
Error logs
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] Graph break from `Tensor.item()`, consider setting:
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] torch._dynamo.config.capture_scalar_outputs = True
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] or:
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] env TORCHDYNAMO_CAPTURE_SCALAR_OUTPUTS=1
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] to include these operations in the captured graph.
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0]
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] Graph break: from user code at:
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] File "/mnt/clusterstorage/workspace/kevin/basic_training.py", line 22, in torch_dynamo_resume_in_run_fwd_bwd_at_21
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] scaler.step(optimizer)
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] File "/mnt/clusterstorage/workspace/kevin/gradscaler2.py", line 457, in step
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] retval = self._maybe_opt_step(optimizer, optimizer_state, *args, **kwargs)
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] File "/mnt/clusterstorage/workspace/kevin/gradscaler2.py", line 351, in _maybe_opt_step
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] if not sum(v.item() for v in optimizer_state["found_inf_per_device"].values()):
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] File "/mnt/clusterstorage/workspace/kevin/gradscaler2.py", line 351, in <genexpr>
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0] if not sum(v.item() for v in optimizer_state["found_inf_per_device"].values()):
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0]
W1020 03:27:52.390000 188995 torch/_dynamo/variables/tensor.py:776] [1/0]
Traceback (most recent call last):
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/output_graph.py", line 1446, in _call_user_compiler
compiled_fn = compiler_fn(gm, self.example_inputs())
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/repro/after_dynamo.py", line 129, in __call__
compiled_gm = compiler_fn(gm, example_inputs)
File "/usr/local/lib/python3.10/dist-packages/torch/__init__.py", line 2235, in __call__
return compile_fx(model_, inputs_, config_patches=self.config)
File "/usr/local/lib/python3.10/dist-packages/torch/_inductor/compile_fx.py", line 1521, in compile_fx
return aot_autograd(
File "/usr/local/lib/python3.10/dist-packages/torch/_dynamo/backends/common.py", line 72, in __call__
cg = aot_module_simplified(gm, example_inputs, **self.kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_functorch/aot_autograd.py", line 1071, in aot_module_simplified
compiled_fn = dispatch_and_compile()
File "/usr/local/lib/python3.10/dist-packages/torch/_functorch/aot_autograd.py", line 1056, in dispatch_and_compile
compiled_fn, _ = create_aot_dispatcher_function(
File "/usr/local/lib/python3.10/dist-packages/torch/_functorch/aot_autograd.py", line 522, in create_aot_dispatcher_function
return _create_aot_dispatcher_function(
File "/usr/local/lib/python3.10/dist-packages/torch/_functorch/aot_autograd.py", line 623, in _create_aot_dispatcher_function
fw_metadata = run_functionalized_fw_and_collect_metadata(
File "/usr/local/lib/python3.10/dist-packages/torch/_functorch/_aot_autograd/collect_metadata_analysis.py", line 173, in inner
flat_f_outs = f(*flat_f_args)
File "/usr/local/lib/python3.10/dist-packages/torch/_functorch/_aot_autograd/traced_function_transforms.py", line 859, in functional_call
out = PropagateUnbackedSymInts(mod).run(
File "/usr/loc
B533
al/lib/python3.10/dist-packages/torch/fx/interpreter.py", line 146, in run
self.env[node] = self.run_node(node)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/experimental/symbolic_shapes.py", line 5498, in run_node
result = super().run_node(n)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/interpreter.py", line 203, in run_node
return getattr(self, n.op)(n.target, args, kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/fx/interpreter.py", line 275, in call_function
return target(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/functional_tensor.py", line 534, in __torch_dispatch__
outs_unwrapped = func._op_dk(
File "/usr/local/lib/python3.10/dist-packages/torch/utils/_stats.py", line 21, in wrapper
return fn(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 1238, in __torch_dispatch__
return self.dispatch(func, types, args, kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 1692, in dispatch
return self._cached_dispatch_impl(func, types, args, kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 1339, in _cached_dispatch_impl
output = self._dispatch_impl(func, types, args, kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 1983, in _dispatch_impl
op_impl_out = op_impl(self, func, *args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_impls.py", line 551, in foreach_run_and_map_input_device
fake_mode.fake_tensor_converter.from_meta_and_device(
File "/usr/local/lib/python3.10/dist-packages/torch/_subclasses/fake_tensor.py", line 465, in from_meta_and_device
t.device.type == "meta"
AttributeError: 'list' object has no attribute 'device'
While executing %_amp_foreach_non_finite_check_and_unscale_ : [num_users=0] = call_function[target=torch._amp_foreach_non_finite_check_and_unscale_](args = ([%l_optimizer_param_groups_0_params_0_grad, %l_optimizer_param_groups_0_params_1_grad], %retval, %retval_1), kwargs = {})
Original traceback:
File "/mnt/clusterstorage/workspace/kevin/gradscaler2.py", line 451, in step
self.unscale_(optimizer)
File "/mnt/clusterstorage/workspace/kevin/gradscaler2.py", line 338, in unscale_
optimizer_state["found_inf_per_device"] = self._unscale_grads_(
File "/mnt/clusterstorage/workspace/kevin/gradscaler2.py", line 279, in _unscale_grads_
torch._amp_foreach_non_finite_check_and_unscale_(
Minified repro
from math import inf
import torch
from torch import tensor, device
import torch.fx as fx
import torch._dynamo
from torch._dynamo.testing import rand_strided
from torch._dynamo.debug_utils import run_fwd_maybe_bwd
import torch._dynamo.config
import torch._inductor.config
import torch._functorch.config
import torch.fx.experimental._config
from torch.nn import *
class Repro(torch.nn.Module):
def __init__(self) -> None:
super().__init__()
def forward(self, L_L_optimizer_param_groups_0_params_0_grad_ : torch.Tensor, L_L_optimizer_param_groups_0_params_1_grad_ : torch.Tensor, retval, retval_1):
l_l_optimizer_param_groups_0_params_0_grad_ = L_L_optimizer_param_groups_0_params_0_grad_
l_l_optimizer_param_groups_0_params_1_grad_ = L_L_optimizer_param_groups_0_params_1_grad_
_set_grad_enabled = torch._C._set_grad_enabled(False); _set_grad_enabled = None
_amp_foreach_non_finite_check_and_unscale_ = torch._amp_foreach_non_finite_check_and_unscale_([l_l_optimizer_param_groups_0_params_0_grad_, l_l_optimizer_param_groups_0_params_1_grad_], retval, retval_1); l_l_optimizer_param_groups_0_params_0_grad_ = l_l_optimizer_param_groups_0_params_1_grad_ = retval = retval_1 = None
return (_amp_foreach_non_finite_check_and_unscale_,)
mod = Repro()
def load_args(reader):
buf0 = reader.storage('db1318cb970abdd196e5b690171477cca3ad8647', 65536, device=device(type='cuda', index=0))
reader.tensor(buf0, (16, 1024), is_leaf=True) # L_L_optimizer_param_groups_0_params_0_grad_
buf1 = reader.storage('7c518087601bc171d0842474bb14ee7425812ab7', 64, device=device(type='cuda', index=0))
reader.tensor(buf1, (16,), is_leaf=True) # L_L_optimizer_param_groups_0_params_1_grad_
buf2 = reader.storage('9069ca78e7450a285173431b3e52c5c25299e473', 4, device=device(type='cuda', index=0))
reader.tensor(buf2, (), is_leaf=True) # retval
buf3 = reader.storage('042d080d32daa72198e939a275e3d89a10eb9ec1', 4, device=device(type='cuda', index=0))
reader.tensor(buf3, (), is_leaf=True) # retval_1
load_args._version = 0
if __name__ == '__main__':
from torch._dynamo.repro.after_dynamo import run_repro
run_repro(mod, load_args, accuracy=False, command='run',
save_dir='/mnt/clusterstorage/workspace/kevin/checkpoints', autocast=False, backend='inductor')
Versions
PyTorch version: 2.5.0-rc10
Is debug build: False
CUDA used to build PyTorch: 12.6
ROCM used to build PyTorch: N/A
OS: Ubuntu 22.04.5 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: version 3.30.4
Libc version: glibc-2.35
Python version: 3.10.12 (main, Sep 11 2024, 15:47:36) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-6.5.13-650-3434-22042-coreweave-1-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.6.77
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: GPU 0: NVIDIA H100 80GB HBM3
Nvidia driver version: 535.183.06
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.3.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.3.0
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 46 bits physical, 57 bits virtual
Byte Order: Little Endian
CPU(s): 128
On-line CPU(s) list: 0-127
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) Platinum 8462Y+
CPU family: 6
Model: 143
Thread(s) per core: 2
Core(s) per socket: 32
Socket(s): 2
Stepping: 8
CPU max MHz: 4100.0000
CPU min MHz: 800.0000
BogoMIPS: 5600.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cat_l2 cdp_l3 invpcid_single cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hfi vnmi avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr ibt amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
Virtualization: VT-x
L1d cache: 3 MiB (64 instances)
L1i cache: 2 MiB (64 instances)
L2 cache: 128 MiB (64 instances)
L3 cache: 120 MiB (2 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38,40,42,44,46,48,50,52,54,56,58,60,62,64,66,68,70,72,74,76,78,80,82,84,86,88,90,92,94,96,98,100,102,104,106,108,110,112,114,116,118,120,122,124,126
NUMA node1 CPU(s): 1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39,41,43,45,47,49,51,53,55,57,59,61,63,65,67,69,71,73,75,77,79,81,83,85,87,89,91,93,95,97,99,101,103,105,107,109,111,113,115,117,119,121,123,125,127
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; RSB filling; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Versions of relevant libraries:
[pip3] clip-anytorch==2.6.0
[pip3] dctorch==0.1.2
[pip3] DISTS-pytorch==0.1
[pip3] gpytorch==1.13
[pip3] lovely-numpy==0.2.13
[pip3] mypy-extensions==1.0.0
[pip3] numpy==1.24.3
[pip3] torch==2.5.0rc10
[pip3] torchaudio==2.5.0rc4
[pip3] torchdiffeq==0.2.4
[pip3] torchsde==0.2.6
[pip3] torchvision==0.20.0rc6
[pip3] triton==3.1.0
[pip3] welford-torch==0.2.4
[conda] Could not collect
My final goal is to run _amp_foreach_non_finite_check_and_unscale_ inside my own torch.compiled code.
cc @ezyang @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @amjames @rec