The first try reused TensorListMetadata, which caused illegal memory access issues when there were too many tensors in the list. We just launch multiple kernels with a simpler version of the struct (to minimize kernels launched).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/119927
Approved by: https://github.com/albanD
Meta registration wrongly assumes 4D inputs, while the underlying op allows 3D inputs for the `mha_varlen_fwd()` case.
Testing: I added `detach()`es so the NJT test `test_sdpa_compile()` won't fail for a view-related reason. It should pass now with this fix.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/119812
Approved by: https://github.com/drisspg
Fixes https://github.com/pytorch/pytorch/issues/118129
Suppressions automatically added with
```
import re
with open("error_file.txt", "r") as f:
errors = f.readlines()
error_lines = {}
for error in errors:
match = re.match(r"(.*):(\d+):\d+: error:.*\[(.*)\]", error)
if match:
file_path, line_number, error_type = match.groups()
if file_path not in error_lines:
error_lines[file_path] = {}
error_lines[file_path][int(line_number)] = error_type
for file_path, lines in error_lines.items():
with open(file_path, "r") as f:
code = f.readlines()
for line_number, error_type in sorted(lines.items(), key=lambda x: x[0], reverse=True):
code[line_number - 1] = code[line_number - 1].rstrip() + f" # type: ignore[{error_type}]\n"
with open(file_path, "w") as f:
f.writelines(code)
```
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Co-authored-by: Catherine Lee <csl@fb.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/118533
Approved by: https://github.com/Skylion007, https://github.com/zou3519
Fixes https://github.com/pytorch/pytorch/issues/118129
Suppressions automatically added with
```
import re
with open("error_file.txt", "r") as f:
errors = f.readlines()
error_lines = {}
for error in errors:
match = re.match(r"(.*):(\d+):\d+: error:.*\[(.*)\]", error)
if match:
file_path, line_number, error_type = match.groups()
if file_path not in error_lines:
error_lines[file_path] = {}
error_lines[file_path][int(line_number)] = error_type
for file_path, lines in error_lines.items():
with open(file_path, "r") as f:
code = f.readlines()
for line_number, error_type in sorted(lines.items(), key=lambda x: x[0], reverse=True):
code[line_number - 1] = code[line_number - 1].rstrip() + f" # type: ignore[{error_type}]\n"
with open(file_path, "w") as f:
f.writelines(code)
```
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/118533
Approved by: https://github.com/Skylion007, https://github.com/zou3519
This should fix remaining errors with Resize op in torchvision: https://github.com/pytorch/vision/actions/runs/7298953575?pr=8127
```
/opt/conda/envs/ci/lib/python3.8/site-packages/torch/nn/functional.py:4072: in interpolate
return torch._C._nn._upsample_bicubic2d_aa(input, output_size, align_corners, scale_factors)
E torch._dynamo.exc.TorchRuntimeError: Failed running call_function <function interpolate at 0x7f4443fe00d0>(*(FakeTensor(..., size=(1, s0, s1, s2)),), **{'size': [s4, floor(s3*s4/floor(s1*s3/s2))], 'mode': 'bicubic', 'align_corners': False, 'antialias': True}):
E aten/src/ATen/RegisterCompositeImplicitAutograd.cpp:5567: SymIntArrayRef expected to contain only concrete integers
E
E from user code:
E File "/pytorch/vision/torchvision/transforms/v2/functional/_geometry.py", line 260, in resize_image
E image = interpolate(
E
E Set TORCH_LOGS="+dynamo" and TORCHDYNAMO_VERBOSE=1 for more information
E
E
E You can suppress this exception and fall back to eager by setting:
E import torch._dynamo
E torch._dynamo.config.suppress_errors = True
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/117347
Approved by: https://github.com/peterbell10
Summary:
This PR adds in support for passing in a alpha Tensor, which represents
a tensor of alpha values to fuse into the matmul.
```
cusparselt_sparse_mm = alpha A @ B + bias
```
This operation is necessary for quantization, where we would like to
fuse one of the dequant matmuls into the sparse op.
Test Plan:
```
python test/test_sparse_semi_structured -k alpha
```
Reviewers:
Subscribers:
Tasks:
Tags:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/112056
Approved by: https://github.com/cpuhrsch
_cslt_sparse_mm + additional stride checking in test.
Summary:
This PR adds in meta registrations for _cslt_sparse_mm.
Based on the work @drisspg did
in #114370.
Additionally, it updates the tests by checking that the strides of the
spare result and the result returned by sparse+compile are the same, to
avoid errors like those found in
https://github.com/pytorch/pytorch/pull/114477.
Test Plan:
```
python test/test_sparse_semi_structred -k compile_cusparselt
python test/test_sparse_semi_structred -k compile_cutlass
```
Reviewers:
Subscribers:
Tasks:
Tags:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/114685
Approved by: https://github.com/alexsamardzic, https://github.com/drisspg
# Summary
Improved Fix for Attention Mask Alignment Issue (#112577)
This PR addresses Issue #112577 by refining the previously implemented fix, which was found to be incorrect and causes un-needed memory regressions. The update simplifies the approach to handling the alignment of the attention mask for mem eff attention.
## Changes
Alignment Check and Padding: Initially, the alignment of the attention mask is checked. If misalignment is detected, padding is applied, followed by slicing. During this process, a warning is raised to alert users.
Should this be warn_once?
We only call expand, once on the aligned mask.
Reference
https://github.com/facebookresearch/xformers/blob/main/xformers/ops/fmha/cutlass.py#L115
@albanD, @mruberry, @jbschlosser, @walterddr, and @mikaylagawarecki.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/114173
Approved by: https://github.com/danthe3rd
# Summary
Improved Fix for Attention Mask Alignment Issue (#112577)
This PR addresses Issue #112577 by refining the previously implemented fix, which was found to be incorrect and causes un-needed memory regressions. The update simplifies the approach to handling the alignment of the attention mask for mem eff attention.
## Changes
Alignment Check and Padding: Initially, the alignment of the attention mask is checked. If misalignment is detected, padding is applied, followed by slicing. During this process, a warning is raised to alert users.
Should this be warn_once?
We only call expand, once on the aligned mask.
Reference
https://github.com/facebookresearch/xformers/blob/main/xformers/ops/fmha/cutlass.py#L115
@albanD, @mruberry, @jbschlosser, @walterddr, and @mikaylagawarecki.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/114173
Approved by: https://github.com/danthe3rd
masked_scatter_backward was previously implemented as a
CompositeExplicitAutograd, which involved a decomp that calls
masked_select, and masked_select in general produces data-dependent
shapes that inductor doesn't support. But masked_scatter_backward
reshapes the return value of masked_select such that the end result has
a static shape again.
I have converted masked_scatter_backward into an aten op to avoid this
issue.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/109642
Approved by: https://github.com/ezyang
ghstack dependencies: #108170
Testing out some new rules that are in beta, I think I will apply this one codebase wide once it's out of preview. Replaces the hack of using `[:]` to do copies of list with the proper copy method. More efficient and more readable.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/112990
Approved by: https://github.com/ezyang
- Extend `test_torch_dispatch_meta_outplace` to test torch ops that do not have an out parameter but have aten op overloads that have out parameters. Additionally, Python decompositions may register `OpOverloadPacket`'s so decompositions need to be tested to ensure all `OpOverloads` still function for the `Meta` key (e.g. if a python decomposition is registered for an aten op `aten.foo` with overloads `[default, out]`, the python function needs to support receiving out arguments)
- Add out parameter wrappers to python decomps for aten ops that have out overloads
CC. @ezyang @albanD @lezcano
Fixes#107713
Pull Request resolved: https://github.com/pytorch/pytorch/pull/107707
Approved by: https://github.com/lezcano
aten.softmax will generate a different decomposition for fp16/bf16 and fp32 because when invoked in lower precision it will upcast the inputs to fp32 and then downcast after. This has been causing us to miss bf16 patterns. For example, Camembert improves 20% with this PR (as do I'm sure many other models).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/109142
Approved by: https://github.com/yanboliang
ghstack dependencies: #109663, #108894, #108917
aten.softmax will generate a different decomposition for fp16/bf16 and fp32 because when invoked in lower precision it will upcast the inputs to fp32 and then downcast after. This has been causing us to miss bf16 patterns. For example, Camembert improves 20% with this PR (as do I'm sure many other models).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/109142
Approved by: https://github.com/yanboliang
ghstack dependencies: #108894, #108917
This fixes numerous tests which were xfailing. For instance, the
`_segment_reduce.lengths` OpInfo test, which was previously relying on
the fallback kernel to determine the shape of the meta tensor. The
fallback kernel would fail with
segment_reduce(): Expected all rows of lengths along axis to sum to data.size(lengths.dim()-1) when !unsafe.
as it was trying to read the values of a meta tensor.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/109359
Approved by: https://github.com/ezyang
The sample inputs is a bit involved because there are a lot of
shenanigans in the derivative formula. Check comments.
This is exercised in vdd, internal test `buck2 run '@fbcode//mode/opt' fbcode//pytorch/benchmark/fb/test_gpu:run_test_gpu -- 'pytorch.benchmark.fb.test_gpu.test_gpu.TestBenchmarkFbGpu.test_train_blue_reels_vdd_v3_inductor_speedup'`
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/109211
Approved by: https://github.com/albanD, https://github.com/zou3519
# Summary
## PR Dependencies
I don't use ghstack :( this is a PR where it would have been helpful. That beings said I am going to peel off some PRs to make reviewing this easier:
- [x] Separate build flags for Flash and MemEff: #107985
### Description
This pull request updates the version of _scaled_dot_product_flash_attention from version 1 to version 2. The changes are based on the flash attention code originally authored by @tridao
### Changes Made
The majority of the changes in this pull request involve:
- Copying over the flash_attention sources.
- Updating header files.
- Removing padding and slicing code from within the flash_attention kernel and relocating it to the composite implicit region of the SDPA. This was need to make the kernel functional and appease autograd.
- Introducing a simple kernel generator to generate different instantiations of the forward and backward flash templates.
- Adding conditional compilation (ifdef) to prevent building when nvcc is invoked with gencode < sm80.
- Introducing a separate dependent option for mem_eff_attention, as flash_attention v2 lacks support for Windows and cannot be built for sm50 generation codes.
- Modifying build.sh to reduce parallelization on sm86 runners and to lower the maximum parallelization on the manywheel builds. This adjustment was made to address out-of-memory issues during the compilation of FlashAttentionV2 sources.
- Adding/Updating tests.
### Notes for Reviewers
This is not a fun review, and I apologize in advance.
Most of the files-changed are in the flash_attn/ folder. The only files of interest here IMO:
- aten/src/ATen/native/transformers/cuda/flash_attn/flash_api.cpp
- aten/src/ATen/native/transformers/cuda/flash_attn/kernels/generate_kernels.py ( this has been incorporated upstream to flash-attention github)
There are a number of files all related to avoiding OOMs in CI/CD. These are typically shell scripts.
### Follow up items
- Include the updates from e07aa036db and 9e5e8bc91e | https://github.com/pytorch/pytorch/issues/108108
### Work Items
- [x] I don't think Windows will be supported for 3.1.0 - Need to update cmakee
- [x] Let multi_query/attention pass through and test | UPDATE: I have the fast path implemented here: https://github.com/pytorch/pytorch/pull/106730 but since this will require changes to semantics of math to call repeat_interleave, I think this should be done as a followup.
- [x] Had to drop cutlass back to 3.0.0 to get it to compile. Need to figure out how to upgrade to 3.1.0 and later. Spoke with Tri and he is going to be taking a look. Note: compiling with clang currently errors for the cute headers.
- [x] Update test exercise above codepath
- [x] Still need to disable on seq_len % 128 != 0 for backward( Tri beat me to it a4f148b6ab)
- [x] Add determinism warning to BWD, Tri got to this one as well: 1c41d2b
- [x] Update dispatcher to universally prefer FlashV2
- [x] Update tests to exercise new head_dims
- [x] Move the head_dim padding from kernel to top level composite implicit function in order to make it purely functional
- [x] Create template generator script
- [x] Initial cmake support for building kernels/ folder
- [x] Replay CudaGraph changes
### Results
#### Forward only
The TFlops are reported here are on a100 that is underclocked.
#### Forward+Backward
Ran a sweep and for large compute bound sizes we do see a ~2x performance increase for forw+back.
<img width="1684" alt="Screenshot 2023-07-20 at 3 47 47 PM" src="https://github.com/pytorch/pytorch/assets/32754868/fdd26e07-0077-4878-a417-f3a418b6fb3b">
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105602
Approved by: https://github.com/huydhn, https://github.com/cpuhrsch
# Summary
## PR Dependencies
I don't use ghstack :( this is a PR where it would have been helpful. That beings said I am going to peel off some PRs to make reviewing this easier:
- [x] Separate build flags for Flash and MemEff: #107985
### Description
This pull request updates the version of _scaled_dot_product_flash_attention from version 1 to version 2. The changes are based on the flash attention code originally authored by @tridao
### Changes Made
The majority of the changes in this pull request involve:
- Copying over the flash_attention sources.
- Updating header files.
- Removing padding and slicing code from within the flash_attention kernel and relocating it to the composite implicit region of the SDPA. This was need to make the kernel functional and appease autograd.
- Introducing a simple kernel generator to generate different instantiations of the forward and backward flash templates.
- Adding conditional compilation (ifdef) to prevent building when nvcc is invoked with gencode < sm80.
- Introducing a separate dependent option for mem_eff_attention, as flash_attention v2 lacks support for Windows and cannot be built for sm50 generation codes.
- Modifying build.sh to reduce parallelization on sm86 runners and to lower the maximum parallelization on the manywheel builds. This adjustment was made to address out-of-memory issues during the compilation of FlashAttentionV2 sources.
- Adding/Updating tests.
### Notes for Reviewers
This is not a fun review, and I apologize in advance.
Most of the files-changed are in the flash_attn/ folder. The only files of interest here IMO:
- aten/src/ATen/native/transformers/cuda/flash_attn/flash_api.cpp
- aten/src/ATen/native/transformers/cuda/flash_attn/kernels/generate_kernels.py ( this has been incorporated upstream to flash-attention github)
There are a number of files all related to avoiding OOMs in CI/CD. These are typically shell scripts.
### Follow up items
- Include the updates from e07aa036db and 9e5e8bc91e | https://github.com/pytorch/pytorch/issues/108108
### Work Items
- [x] I don't think Windows will be supported for 3.1.0 - Need to update cmakee
- [x] Let multi_query/attention pass through and test | UPDATE: I have the fast path implemented here: https://github.com/pytorch/pytorch/pull/106730 but since this will require changes to semantics of math to call repeat_interleave, I think this should be done as a followup.
- [x] Had to drop cutlass back to 3.0.0 to get it to compile. Need to figure out how to upgrade to 3.1.0 and later. Spoke with Tri and he is going to be taking a look. Note: compiling with clang currently errors for the cute headers.
- [x] Update test exercise above codepath
- [x] Still need to disable on seq_len % 128 != 0 for backward( Tri beat me to it a4f148b6ab)
- [x] Add determinism warning to BWD, Tri got to this one as well: 1c41d2b
- [x] Update dispatcher to universally prefer FlashV2
- [x] Update tests to exercise new head_dims
- [x] Move the head_dim padding from kernel to top level composite implicit function in order to make it purely functional
- [x] Create template generator script
- [x] Initial cmake support for building kernels/ folder
- [x] Replay CudaGraph changes
### Results
#### Forward only
The TFlops are reported here are on a100 that is underclocked.
#### Forward+Backward
Ran a sweep and for large compute bound sizes we do see a ~2x performance increase for forw+back.
<img width="1684" alt="Screenshot 2023-07-20 at 3 47 47 PM" src="https://github.com/pytorch/pytorch/assets/32754868/fdd26e07-0077-4878-a417-f3a418b6fb3b">
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105602
Approved by: https://github.com/huydhn, https://github.com/cpuhrsch
# Summary
## PR Dependencies
I don't use ghstack :( this is a PR where it would have been helpful. That beings said I am going to peel off some PRs to make reviewing this easier:
- [x] Separate build flags for Flash and MemEff: #107985
### Description
This pull request updates the version of _scaled_dot_product_flash_attention from version 1 to version 2. The changes are based on the flash attention code originally authored by @tridao
### Changes Made
The majority of the changes in this pull request involve:
- Copying over the flash_attention sources.
- Updating header files.
- Removing padding and slicing code from within the flash_attention kernel and relocating it to the composite implicit region of the SDPA. This was need to make the kernel functional and appease autograd.
- Introducing a simple kernel generator to generate different instantiations of the forward and backward flash templates.
- Adding conditional compilation (ifdef) to prevent building when nvcc is invoked with gencode < sm80.
- Introducing a separate dependent option for mem_eff_attention, as flash_attention v2 lacks support for Windows and cannot be built for sm50 generation codes.
- Modifying build.sh to reduce parallelization on sm86 runners and to lower the maximum parallelization on the manywheel builds. This adjustment was made to address out-of-memory issues during the compilation of FlashAttentionV2 sources.
- Adding/Updating tests.
### Notes for Reviewers
This is not a fun review, and I apologize in advance.
Most of the files-changed are in the flash_attn/ folder. The only files of interest here IMO:
- aten/src/ATen/native/transformers/cuda/flash_attn/flash_api.cpp
- aten/src/ATen/native/transformers/cuda/flash_attn/kernels/generate_kernels.py ( this has been incorporated upstream to flash-attention github)
There are a number of files all related to avoiding OOMs in CI/CD. These are typically shell scripts.
### Follow up items
- Include the updates from e07aa036db and 9e5e8bc91e | https://github.com/pytorch/pytorch/issues/108108
### Work Items
- [x] I don't think Windows will be supported for 3.1.0 - Need to update cmakee
- [x] Let multi_query/attention pass through and test | UPDATE: I have the fast path implemented here: https://github.com/pytorch/pytorch/pull/106730 but since this will require changes to semantics of math to call repeat_interleave, I think this should be done as a followup.
- [x] Had to drop cutlass back to 3.0.0 to get it to compile. Need to figure out how to upgrade to 3.1.0 and later. Spoke with Tri and he is going to be taking a look. Note: compiling with clang currently errors for the cute headers.
- [x] Update test exercise above codepath
- [x] Still need to disable on seq_len % 128 != 0 for backward( Tri beat me to it a4f148b6ab)
- [x] Add determinism warning to BWD, Tri got to this one as well: 1c41d2b
- [x] Update dispatcher to universally prefer FlashV2
- [x] Update tests to exercise new head_dims
- [x] Move the head_dim padding from kernel to top level composite implicit function in order to make it purely functional
- [x] Create template generator script
- [x] Initial cmake support for building kernels/ folder
- [x] Replay CudaGraph changes
### Results
#### Forward only
The TFlops are reported here are on a100 that is underclocked.
#### Forward+Backward
Ran a sweep and for large compute bound sizes we do see a ~2x performance increase for forw+back.
<img width="1684" alt="Screenshot 2023-07-20 at 3 47 47 PM" src="https://github.com/pytorch/pytorch/assets/32754868/fdd26e07-0077-4878-a417-f3a418b6fb3b">
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105602
Approved by: https://github.com/huydhn, https://github.com/cpuhrsch
We allow registering decomps for HigherOrderOp via the existing decomp
mechanisms:
- I refactored those APIs to accept torch._ops.OperatorBase, which is the base
class for torch.ops.HigherOrderOperator and torch.ops.OpOverload
- HigherOrderOps must directly call maybe_handle_decomp in their
ProxyTorchDispatchMode handling in order to resolve decompositions. We
can change this in the future so that they do not need to do this.
Next, we add an inductor decomp for out_dtype. This decomp shouldn't be
generally available because we want to preserve out_dtype to the backend
for other use cases (i.e. executorch).
Test Plan:
- new tests
Pull Request resolved: https://github.com/pytorch/pytorch/pull/108080
Approved by: https://github.com/HDCharles
Some notable changes:
1. `constrain_as_size` allows min value to be less than 2 as it will unconditionally assume min >= 2 for compiler purposes. Instead, we add additional check to make sure max value is always greater than 2.
2. Previously, we used to runtime assert on the unbacked symint's val range which would be always between [2, max]. I modified this logic to assert on [0, max] unless user explicitly specifies the min range.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/106591
Approved by: https://github.com/gmagogsfm, https://github.com/ezyang
Some notable changes:
1. `constrain_as_size` allows min value to be less than 2 as it will unconditionally assume min >= 2 for compiler purposes. Instead, we add additional check to make sure max value is always greater than 2.
2. Previously, we used to runtime assert on the unbacked symint's val range which would be always between [2, max]. I modified this logic to assert on [0, max] unless user explicitly specifies the min range.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/106591
Approved by: https://github.com/gmagogsfm, https://github.com/ezyang
1. add a python meta registration, to fix an issue with the forward pass. The problem was that previously, the C++ meta registration calls [numel()](7b14a14e27/aten/src/ATen/native/TensorAdvancedIndexing.cpp (L329)) which fails (LMK if it's better to fix the C++ implementation to not do this check)
2. Modify the backward to fix an issue in the backward. The backward is not a custom op - it's a custom manual backward implementation. In particular, there's some situations that don't support double backward; the check for whether double backward is allowed requires a .item() call. To fix the meta/fake tensor case, this PR will avoid setting the double backward error only if `GradMode::is_enabled()` - which shouldn't be turned on in PT2.
3. Update skips.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/106429
Approved by: https://github.com/zou3519
# Summary
### Review Points
- Automatically pad tensors to create aligned masks when seqlen_kv is not multiple of 16. This will cause memory spike ~ 2 * attn_mask size which could in theory be big. At appears though that doing this + mem_eff is faster than no_pad + math. SO seems to be worth it
- Using expand to view the attn_mask in 4d. This is a little different to how we enforce q,k,v to be viewed in 4d prior to calling. Also not supprint b*n_heads, seq_lenq, seq_lenkv case.
- Should enable, #96099
### Profiling
I ran a bunch of comparisons between sdpa.MATH and sdp.MemEffAttention. I added a attn_bias of shape (1, 1, seqlen_q, seqln_k). For these experiments seqlen_q == seqlen_k. These were all ran on an a100 80gb gpu.
Configs:
```
# Run a bunch of experiments
batch_sizes = [8, 16, 32]
num_heads = [16, 32]
max_seq_lens = [15, 64, 128, 512, 555, 1024]
embed_dims = [32, 64, 128]
dtypes = [torch.float16, torch.bfloat16, torch.float32]
pad_percentages = [None]
backends = [SDPBackend.EFFICIENT_ATTENTION, SDPBackend.MATH]
run_backward = True
attn_mask = True
```
The function calls `sdpa(input**).sum().backward()`.
I calculated the geomean speedup of the efficient attention path of the math path for all these configs:
`Geomean Speedup: 1.977`
An example comparision with batchsize = 8, num_heads = 32, embed_dim = 64, and dtype = torch.float16:
This was done using the current state of the branch where we force alignment of mask when the last dim is not divisible by 16, which shows up in seq_len = 15 and 555 case.
The full data can be found here:
[attn_mask_sweep.csv](https://github.com/pytorch/pytorch/files/11962399/attn_mask_sweep.csv)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/104310
Approved by: https://github.com/cpuhrsch
# Summary
### Review Points
- Automatically pad tensors to create aligned masks when seqlen_kv is not multiple of 16. This will cause memory spike ~ 2 * attn_mask size which could in theory be big. At appears though that doing this + mem_eff is faster than no_pad + math. SO seems to be worth it
- Using expand to view the attn_mask in 4d. This is a little different to how we enforce q,k,v to be viewed in 4d prior to calling. Also not supprint b*n_heads, seq_lenq, seq_lenkv case.
- Should enable, #96099
### Profiling
I ran a bunch of comparisons between sdpa.MATH and sdp.MemEffAttention. I added a attn_bias of shape (1, 1, seqlen_q, seqln_k). For these experiments seqlen_q == seqlen_k. These were all ran on an a100 80gb gpu.
Configs:
```
# Run a bunch of experiments
batch_sizes = [8, 16, 32]
num_heads = [16, 32]
max_seq_lens = [15, 64, 128, 512, 555, 1024]
embed_dims = [32, 64, 128]
dtypes = [torch.float16, torch.bfloat16, torch.float32]
pad_percentages = [None]
backends = [SDPBackend.EFFICIENT_ATTENTION, SDPBackend.MATH]
run_backward = True
attn_mask = True
```
The function calls `sdpa(input**).sum().backward()`.
I calculated the geomean speedup of the efficient attention path of the math path for all these configs:
`Geomean Speedup: 1.977`
An example comparision with batchsize = 8, num_heads = 32, embed_dim = 64, and dtype = torch.float16:
This was done using the current state of the branch where we force alignment of mask when the last dim is not divisible by 16, which shows up in seq_len = 15 and 555 case.
The full data can be found here:
[attn_mask_sweep.csv](https://github.com/pytorch/pytorch/files/11962399/attn_mask_sweep.csv)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/104310
Approved by: https://github.com/cpuhrsch
This PR re-lands
- [Typing] Fix PEP 484 Violation (#105022)
- Update mypy to 1.4.1 (#91983)
That were reverted due to the conflict with internal source repo.
Mostly fixes for PEP-484 violation (i.e. when default arg is set to None, but type is not annotated as optional)
Plus few real fixes:
- Add missing `_get_upgraders_entry_map` to `torch/_C/__init__.pyi`
- Add missing return statement to `torch._export. deserialize_graph`
- Fix error message in `torch.ao.ns.fx.weight_utils.get_lstm_mod_weights`
- Add assert it `torch/optim/optimizer.py` that Optional list is not None
TODO (in followup PR):
- Fix erroneous `isinstance` check in `torch/ao/quantization/_pt2e/qat_utils.py`
Unrelated, to bypass CI failures due to the gcc9 dependency update in Ubuntu-18.04:
- Add hack to squash older libstdc++ from conda environment in favor one from OS to `.ci/docker/install_conda.sh`
- Update bazel cuda builds to focal, as with libstdc++-6.0.32 bazel builds loose the ability to catch exceptions (probably because they link with cupti statically, but I could not found where it is done)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105227
Approved by: https://github.com/atalman, https://github.com/albanD, https://github.com/Skylion007
This PR re-lands
- [Typing] Fix PEP 484 Violation (#105022)
- Update mypy to 1.4.1 (#91983)
That were reverted due to the conflict with internal source repo.
Mostly fixes for PEP-484 violation (i.e. when default arg is set to None, but type is not annotated as optional)
Plus few real fixes:
- Add missing `_get_upgraders_entry_map` to `torch/_C/__init__.pyi`
- Add missing return statement to `torch._export. deserialize_graph`
- Fix error message in `torch.ao.ns.fx.weight_utils.get_lstm_mod_weights`
- Add assert it `torch/optim/optimizer.py` that Optional list is not None
TODO (in followup PR):
- Fix erroneous `isinstance` check in `torch/ao/quantization/_pt2e/qat_utils.py`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105227
Approved by: https://github.com/atalman, https://github.com/albanD, https://github.com/Skylion007
Not sure, how it worked before, but if arguments must be annotated is optional if they are defaulted to None
Towards enabling mypy-1.4.1 in lintrunner
<!--
copilot:poem
-->
### <samp>🤖 Generated by Copilot at 5e1b9f4</samp>
> _We annotate the arguments of doom_
> _To show the `None` values of gloom_
> _We improve the type checking and readability_
> _With `Optional` annotations of metal-ity_
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105022
Approved by: https://github.com/izaitsevfb, https://github.com/huydhn, https://github.com/Skylion007
The idea here is to create do a graph mutation to:
* Create an initial dependency token at the beginning of the program.
* Replace non-functional version of assertion statements to functional version.
* The functional version of assertion statement will:
* Accept a dependency token from output of previous functional assertion statement (or the initial dependency token if there isn't any).
* Generate a dependency token as the output of assertion statement.
* Augment the output to include the dependency token generated by last assertion statement.
The goal here is to:
* Form an explicit dependency chain and avoid potential reordering during other passes of compiling.
* Make the assertions a part of overall execution graph will affect the final output (or it could potentially be DCEed).
**NOTE:**
* Currently only cover `contrain_range` and WIP to support other assertions. Send out this PR to collect feedback first.
* Here it only focus on implementation itself. Will integrate it with current export in future PR.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103757
Approved by: https://github.com/avikchaudhuri
At high current implementation of constrains functions (constrain_as_**) will raise exception for the following code snippets:
```
def f(x):
a = x.item()
constrain_as_size(a, 4, 7)
return torch.empty((a, 4))
inp = torch.tensor([5])
ep = torch._export.export(f, (inp,))
```
The reason is because current constrain logic is:
1) Purely python so it won't survive AOT export (the full node is gone after AOT export since AOT export only maintains aten level op).
2) Utilize side effect to add range constraints for traced symbol's shape env ([code](9591e52880/torch/fx/experimental/symbolic_shapes.py (L370-L372))).
3) If runtime assertion is turned on (by default). [`_AddRuntimeAssertionsForConstraintsPass`](9591e52880/torch/_export/passes/add_runtime_assertions_for_constraints_pass.py (L98-L100)) will try to append assertion node based on range constrains extracted from shape env of symbol during another interpretation round.
4). However, since 1), in the round of AOT export, range constraints logic won't run for symbols generated during this round. And later there is no range constrains information available for assertion round and caused issue.
5) As a result of above, it will failure at `torch.empty((a, 4))` (there is no constrains for `a` that it must be positive).
The fix here is just to implement range constrain logic as a native aten op (CPU implementation as no-op) to make it be able to survive AOT export.
**NOTE:**
[Logic](2d745b95d7/torch/fx/experimental/symbolic_shapes.py (L350-L365C15)) within [`constrain_range`](2d745b95d7/torch/fx/experimental/symbolic_shapes.py (LL313C74-L313C74)) is split out as `constrain_range_int` to capture case when non `SymInt` is passed in and reused in the new `_constrain_range`. The reason is when non `SymInt` is provided:
* If it directly calls `sym_constrain_range`, the C++ version will be called which will be no-op.
* So in this case it calls `constrain_range_int` instead to be able to capture issue like user provides a input whose tensor's shape could be out of range during exporting, like the following for above code example:
```
...
inp = torch.tensor([10])
ep = torch._export.export(f, (inp,)) # immediately raise error
```
Differential Revision: [D46734204](https://our.internmc.facebook.com/intern/diff/D46734204)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103346
Approved by: https://github.com/tugsbayasgalan
Fixes#103606
Was using this script to exercise new code, cause I can never remember which test it is.
```
import torch
@torch.compile(fullgraph=True, dynamic=True)
def shift_right(tensor: torch.Tensor) -> torch.Tensor:
return (tensor >> 2).to(torch.long)
def main():
sample_input = torch.tensor([4, 4, 16, 32], dtype=torch.uint8)
print(shift_right(sample_input))
if __name__ == "__main__":
main()
```
And iterated through the error messages
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103637
Approved by: https://github.com/ezyang
Per title.
there's an off chance that query_reshaped etc was actually discontiguous after reshape, but even in that case I'm pretty sure the computed gradients would still be contiguous, and we are properly transposing output gradients to produce correct strides.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/101128
Approved by: https://github.com/drisspg
This PR introduces a new operator called aten._assert_async.msg, which allows passing a tensor value and assertion message as inputs. As part of TorchDynamo, we're replacing the use of torch._assert with this new operator so that make_fx also knows how to handle assertions. This is subset of https://github.com/pytorch/pytorch/pull/98878, refer there for historic reviews.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/100101
Approved by: https://github.com/jansel
Summary:
Original commit changeset: ba36f8751adc
Original Phabricator Diff: D44788697
Test Plan: model loading is fine after reverting the diff
Reviewed By: zyan0, sayitmemory
Differential Revision: D44921259
---
Fixes #ISSUE_NUMBER
Pull Request resolved: https://github.com/pytorch/pytorch/pull/99168
Approved by: https://github.com/izaitsevfb
