b00b641ff1
Fixes #155013 Pull Request resolved: https://github.com/pytorch/pytorch/pull/155762 Approved by: https://github.com/svekars
14 KiB
.. role:: hidden
:class: hidden-section
Automatic Mixed Precision package - torch.amp
% Both modules below are missing doc entry. Adding them here for now.
% This does not add anything to the rendered page
.. py:module:: torch.cpu.amp
.. py:module:: torch.cuda.amp
.. automodule:: torch.amp
.. currentmodule:: torch.amp
{class}torch.amp provides convenience methods for mixed precision,
where some operations use the torch.float32 (float) datatype and other operations
use lower precision floating point datatype (lower_precision_fp): torch.float16 (half) or torch.bfloat16. Some ops, like linear layers and convolutions,
are much faster in lower_precision_fp. Other ops, like reductions, often require the dynamic
range of float32. Mixed precision tries to match each op to its appropriate datatype.
Ordinarily, "automatic mixed precision training" with datatype of torch.float16 uses {class}torch.autocast and
{class}torch.amp.GradScaler together, as shown in the {ref}Automatic Mixed Precision examples<amp-examples>
and Automatic Mixed Precision recipe.
However, {class}torch.autocast and {class}torch.GradScaler are modular, and may be used separately if desired.
As shown in the CPU example section of {class}torch.autocast, "automatic mixed precision training/inference" on CPU with
datatype of torch.bfloat16 only uses {class}torch.autocast.
:::{warning}
torch.cuda.amp.autocast(args...) and torch.cpu.amp.autocast(args...) is deprecated. Please use torch.amp.autocast("cuda", args...) or torch.amp.autocast("cpu", args...) instead.
torch.cuda.amp.GradScaler(args...) and torch.cpu.amp.GradScaler(args...) is deprecated. Please use torch.amp.GradScaler("cuda", args...) or torch.amp.GradScaler("cpu", args...) instead.
:::
{class}torch.autocast and {class}torch.cpu.amp.autocast are new in version 1.10.
:local: true
(autocasting)=
Autocasting
.. currentmodule:: torch.amp.autocast_mode
.. autofunction:: is_autocast_available
.. currentmodule:: torch
.. autoclass:: autocast
:members:
.. currentmodule:: torch.amp
.. autofunction:: custom_fwd
.. autofunction:: custom_bwd
.. currentmodule:: torch.cuda.amp
.. autoclass:: autocast
:members:
.. autofunction:: custom_fwd
.. autofunction:: custom_bwd
.. currentmodule:: torch.cpu.amp
.. autoclass:: autocast
:members:
(gradient-scaling)=
Gradient Scaling
If the forward pass for a particular op has float16 inputs, the backward pass for
that op will produce float16 gradients.
Gradient values with small magnitudes may not be representable in float16.
These values will flush to zero ("underflow"), so the update for the corresponding parameters will be lost.
To prevent underflow, "gradient scaling" multiplies the network's loss(es) by a scale factor and invokes a backward pass on the scaled loss(es). Gradients flowing backward through the network are then scaled by the same factor. In other words, gradient values have a larger magnitude, so they don't flush to zero.
Each parameter's gradient (.grad attribute) should be unscaled before the optimizer
updates the parameters, so the scale factor does not interfere with the learning rate.
:::{note} AMP/fp16 may not work for every model! For example, most bf16-pretrained models cannot operate in the fp16 numerical range of max 65504 and will cause gradients to overflow instead of underflow. In this case, the scale factor may decrease under 1 as an attempt to bring gradients to a number representable in the fp16 dynamic range. While one may expect the scale to always be above 1, our GradScaler does NOT make this guarantee to maintain performance. If you encounter NaNs in your loss or gradients when running with AMP/fp16, verify your model is compatible. :::
.. currentmodule:: torch.cuda.amp
.. autoclass:: GradScaler
:members:
.. currentmodule:: torch.cpu.amp
.. autoclass:: GradScaler
:members:
(autocast-op-reference)=
Autocast Op Reference
(autocast-eligibility)=
Op Eligibility
Ops that run in float64 or non-floating-point dtypes are not eligible, and will
run in these types whether or not autocast is enabled.
Only out-of-place ops and Tensor methods are eligible.
In-place variants and calls that explicitly supply an out=... Tensor
are allowed in autocast-enabled regions, but won't go through autocasting.
For example, in an autocast-enabled region a.addmm(b, c) can autocast,
but a.addmm_(b, c) and a.addmm(b, c, out=d) cannot.
For best performance and stability, prefer out-of-place ops in autocast-enabled
regions.
Ops called with an explicit dtype=... argument are not eligible,
and will produce output that respects the dtype argument.
(autocast-cuda-op-reference)=
CUDA Op-Specific Behavior
The following lists describe the behavior of eligible ops in autocast-enabled regions.
These ops always go through autocasting whether they are invoked as part of a {class}torch.nn.Module,
as a function, or as a {class}torch.Tensor method. If functions are exposed in multiple namespaces,
they go through autocasting regardless of the namespace.
Ops not listed below do not go through autocasting. They run in the type defined by their inputs. However, autocasting may still change the type in which unlisted ops run if they're downstream from autocasted ops.
If an op is unlisted, we assume it's numerically stable in float16.
If you believe an unlisted op is numerically unstable in float16,
please file an issue.
CUDA Ops that can autocast to float16
__matmul__,
addbmm,
addmm,
addmv,
addr,
baddbmm,
bmm,
chain_matmul,
multi_dot,
conv1d,
conv2d,
conv3d,
conv_transpose1d,
conv_transpose2d,
conv_transpose3d,
GRUCell,
linear,
LSTMCell,
matmul,
mm,
mv,
prelu,
RNNCell
CUDA Ops that can autocast to float32
__pow__,
__rdiv__,
__rpow__,
__rtruediv__,
acos,
asin,
binary_cross_entropy_with_logits,
cosh,
cosine_embedding_loss,
cdist,
cosine_similarity,
cross_entropy,
cumprod,
cumsum,
dist,
erfinv,
exp,
expm1,
group_norm,
hinge_embedding_loss,
kl_div,
l1_loss,
layer_norm,
log,
log_softmax,
log10,
log1p,
log2,
margin_ranking_loss,
mse_loss,
multilabel_margin_loss,
multi_margin_loss,
nll_loss,
norm,
normalize,
pdist,
poisson_nll_loss,
pow,
prod,
reciprocal,
rsqrt,
sinh,
smooth_l1_loss,
soft_margin_loss,
softmax,
softmin,
softplus,
sum,
renorm,
tan,
triplet_margin_loss
CUDA Ops that promote to the widest input type
These ops don't require a particular dtype for stability, but take multiple inputs
and require that the inputs' dtypes match. If all of the inputs are
float16, the op runs in float16. If any of the inputs is float32,
autocast casts all inputs to float32 and runs the op in float32.
addcdiv,
addcmul,
atan2,
bilinear,
cross,
dot,
grid_sample,
index_put,
scatter_add,
tensordot
Some ops not listed here (e.g., binary ops like add) natively promote
inputs without autocasting's intervention. If inputs are a mixture of float16
and float32, these ops run in float32 and produce float32 output,
regardless of whether autocast is enabled.
Prefer binary_cross_entropy_with_logits over binary_cross_entropy
The backward passes of {func}torch.nn.functional.binary_cross_entropy (and {mod}torch.nn.BCELoss, which wraps it)
can produce gradients that aren't representable in float16. In autocast-enabled regions, the forward input
may be float16, which means the backward gradient must be representable in float16 (autocasting float16
forward inputs to float32 doesn't help, because that cast must be reversed in backward).
Therefore, binary_cross_entropy and BCELoss raise an error in autocast-enabled regions.
Many models use a sigmoid layer right before the binary cross entropy layer.
In this case, combine the two layers using {func}torch.nn.functional.binary_cross_entropy_with_logits
or {mod}torch.nn.BCEWithLogitsLoss. binary_cross_entropy_with_logits and BCEWithLogits
are safe to autocast.
(autocast-xpu-op-reference)=
XPU Op-Specific Behavior (Experimental)
The following lists describe the behavior of eligible ops in autocast-enabled regions.
These ops always go through autocasting whether they are invoked as part of a {class}torch.nn.Module,
as a function, or as a {class}torch.Tensor method. If functions are exposed in multiple namespaces,
they go through autocasting regardless of the namespace.
Ops not listed below do not go through autocasting. They run in the type defined by their inputs. However, autocasting may still change the type in which unlisted ops run if they're downstream from autocasted ops.
If an op is unlisted, we assume it's numerically stable in float16.
If you believe an unlisted op is numerically unstable in float16,
please file an issue.
XPU Ops that can autocast to float16
addbmm,
addmm,
addmv,
addr,
baddbmm,
bmm,
chain_matmul,
multi_dot,
conv1d,
conv2d,
conv3d,
conv_transpose1d,
conv_transpose2d,
conv_transpose3d,
GRUCell,
linear,
LSTMCell,
matmul,
mm,
mv,
RNNCell
XPU Ops that can autocast to float32
__pow__,
__rdiv__,
__rpow__,
__rtruediv__,
binary_cross_entropy_with_logits,
cosine_embedding_loss,
cosine_similarity,
cumsum,
dist,
exp,
group_norm,
hinge_embedding_loss,
kl_div,
l1_loss,
layer_norm,
log,
log_softmax,
margin_ranking_loss,
nll_loss,
normalize,
poisson_nll_loss,
pow,
reciprocal,
rsqrt,
soft_margin_loss,
softmax,
softmin,
sum,
triplet_margin_loss
XPU Ops that promote to the widest input type
These ops don't require a particular dtype for stability, but take multiple inputs
and require that the inputs' dtypes match. If all of the inputs are
float16, the op runs in float16. If any of the inputs is float32,
autocast casts all inputs to float32 and runs the op in float32.
bilinear,
cross,
grid_sample,
index_put,
scatter_add,
tensordot
Some ops not listed here (e.g., binary ops like add) natively promote
inputs without autocasting's intervention. If inputs are a mixture of float16
and float32, these ops run in float32 and produce float32 output,
regardless of whether autocast is enabled.
(autocast-cpu-op-reference)=
CPU Op-Specific Behavior
The following lists describe the behavior of eligible ops in autocast-enabled regions.
These ops always go through autocasting whether they are invoked as part of a {class}torch.nn.Module,
as a function, or as a {class}torch.Tensor method. If functions are exposed in multiple namespaces,
they go through autocasting regardless of the namespace.
Ops not listed below do not go through autocasting. They run in the type defined by their inputs. However, autocasting may still change the type in which unlisted ops run if they're downstream from autocasted ops.
If an op is unlisted, we assume it's numerically stable in bfloat16.
If you believe an unlisted op is numerically unstable in bfloat16,
please file an issue. float16 shares the lists of bfloat16.
CPU Ops that can autocast to bfloat16
conv1d,
conv2d,
conv3d,
bmm,
mm,
linalg_vecdot,
baddbmm,
addmm,
addbmm,
linear,
matmul,
_convolution,
conv_tbc,
mkldnn_rnn_layer,
conv_transpose1d,
conv_transpose2d,
conv_transpose3d,
prelu,
scaled_dot_product_attention,
_native_multi_head_attention
CPU Ops that can autocast to float32
avg_pool3d,
binary_cross_entropy,
grid_sampler,
grid_sampler_2d,
_grid_sampler_2d_cpu_fallback,
grid_sampler_3d,
polar,
prod,
quantile,
nanquantile,
stft,
cdist,
trace,
view_as_complex,
cholesky,
cholesky_inverse,
cholesky_solve,
inverse,
lu_solve,
orgqr,
inverse,
ormqr,
pinverse,
max_pool3d,
max_unpool2d,
max_unpool3d,
adaptive_avg_pool3d,
reflection_pad1d,
reflection_pad2d,
replication_pad1d,
replication_pad2d,
replication_pad3d,
mse_loss,
cosine_embedding_loss,
nll_loss,
nll_loss2d,
hinge_embedding_loss,
poisson_nll_loss,
cross_entropy_loss,
l1_loss,
huber_loss,
margin_ranking_loss,
soft_margin_loss,
triplet_margin_loss,
multi_margin_loss,
ctc_loss,
kl_div,
multilabel_margin_loss,
binary_cross_entropy_with_logits,
fft_fft,
fft_ifft,
fft_fft2,
fft_ifft2,
fft_fftn,
fft_ifftn,
fft_rfft,
fft_irfft,
fft_rfft2,
fft_irfft2,
fft_rfftn,
fft_irfftn,
fft_hfft,
fft_ihfft,
linalg_cond,
linalg_matrix_rank,
linalg_solve,
linalg_cholesky,
linalg_svdvals,
linalg_eigvals,
linalg_eigvalsh,
linalg_inv,
linalg_householder_product,
linalg_tensorinv,
linalg_tensorsolve,
fake_quantize_per_tensor_affine,
geqrf,
_lu_with_info,
qr,
svd,
triangular_solve,
fractional_max_pool2d,
fractional_max_pool3d,
adaptive_max_pool3d,
multilabel_margin_loss_forward,
linalg_qr,
linalg_cholesky_ex,
linalg_svd,
linalg_eig,
linalg_eigh,
linalg_lstsq,
linalg_inv_ex
CPU Ops that promote to the widest input type
These ops don't require a particular dtype for stability, but take multiple inputs
and require that the inputs' dtypes match. If all of the inputs are
bfloat16, the op runs in bfloat16. If any of the inputs is float32,
autocast casts all inputs to float32 and runs the op in float32.
cat,
stack,
index_copy
Some ops not listed here (e.g., binary ops like add) natively promote
inputs without autocasting's intervention. If inputs are a mixture of bfloat16
and float32, these ops run in float32 and produce float32 output,
regardless of whether autocast is enabled.
% This module needs to be documented. Adding here in the meantime
% for tracking purposes
.. py:module:: torch.amp.autocast_mode
.. py:module:: torch.cpu.amp.autocast_mode
.. py:module:: torch.cuda.amp.autocast_mode
.. py:module:: torch.cuda.amp.common
.. py:module:: torch.amp.grad_scaler
.. py:module:: torch.cpu.amp.grad_scaler
.. py:module:: torch.cuda.amp.grad_scaler