This PR removes `ProcessGroupCudaP2P` and changes async-TP to use `SymmetricMemory`. The async-TP implementation is still workspace-based, but it now doesn't require a buffer size to be specified upfront.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/128762
Approved by: https://github.com/wanchaol
The `usort` config in `pyproject.toml` has no effect due to a typo. Fixing the typo make `usort` do more and generate the changes in the PR. Except `pyproject.toml`, all changes are generated by `lintrunner -a --take UFMT --all-files`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/127126
Approved by: https://github.com/kit1980
The `usort` config in `pyproject.toml` has no effect due to a typo. Fixing the typo make `usort` do more and generate the changes in the PR. Except `pyproject.toml`, all changes are generated by `lintrunner -a --take UFMT --all-files`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/127126
Approved by: https://github.com/kit1980
ghstack dependencies: #127122, #127123, #127124, #127125
## Context
This stack prototypes automatic micro-pipelining of `all-gather -> matmul` and `matmul -> reduce-scatter` via Inductor. The idea originates from the paper [Overlap Communication with Dependent Computation via
Decomposition in Large Deep Learning Models](https://dl.acm.org/doi/pdf/10.1145/3567955.3567959). The implementation and some key optimizations are heavily influenced by @lw's implementation in xformers.
The stack contains several components:
- `ProcessGroupCudaP2P` - a thin wrapper around `ProcessGroupNCCL`. It in addition maintains a P2P workspace that enables SM-free, one-sided P2P communication which is needed for optimal micro-pipelining.
- `fused_all_gather_matmul` and `fused_matmul_reduce_scatter` dispatcher ops.
- Post-grad fx pass that detects `all-gather -> matmul` and `matmul -> reduce-scatter` and replaces them with the fused dispatcher ops.
To enable the prototype feature:
- Set the distributed backend to `cuda_p2p`.
- Set `torch._inductor.config._micro_pipeline_tp` to `True`.
*NOTE: the prototype sets nothing in stone w.r.t to each component's design. The purpose is to have a performant baseline with reasonable design on which each component can be further improved.*
## Benchmark
Setup:
- 8 x H100 (500W) + 3rd gen NVSwitch.
- Llama3 8B training w/ torchtitan.
- 8-way TP. Reduced the number of layers from 32 to 8 for benchmarking purpose.
Trace (baseline): https://interncache-all.fbcdn.net/manifold/perfetto-artifacts/tree/ui/index.html#!/?url=https://interncache-all.fbcdn.net/manifold/perfetto_internal_traces/tree/shared_trace/yifu_tmpjaz8zgx0
<img width="832" alt="image" src="https://github.com/pytorch/pytorch/assets/4156752/4addba77-5abc-4d2e-93ea-f68078587fe1">
Trace (w/ micro pipelining): https://interncache-all.fbcdn.net/manifold/perfetto-artifacts/tree/ui/index.html#!/?url=https://interncache-all.fbcdn.net/manifold/perfetto_internal_traces/tree/shared_trace/yifu_tmpn073b4wn
<img width="963" alt="image" src="https://github.com/pytorch/pytorch/assets/4156752/4f44e78d-8196-43ab-a1ea-27390f07e9d2">
## This PR
`ProcessGroupCudaP2P` is a thin wrapper around `ProcessGroupNCCL`. By default, it routes all collectives to the underlying `ProcessGroupNCCL`. In addition, `ProcessGroupCudaP2P` initializes a P2P workspace that allows direct GPU memory access among the members. The workspace can be used in Python to optimize intra-node communication patterns or to create custom intra-node collectives in CUDA.
`ProcessGroupCudaP2P` aims to bridge the gap where certain important patterns can be better optimized via fine-grained P2P memory access than with collectives in the latest version of NCCL. It is meant to complement NCCL rather than replacing it.
Usage:
```
# Using ProcessGroupCudaP2P
dist.init_process_group(backend="cuda_p2p", ...)
# Using ProcessGroupCudaP2P while specifying ProcessGroupCudaP2P.Options
pg_options = ProcessGroupCudaP2P.Options()
dist.init_process_group(backend="cuda_p2p", pg_options=pg_options, ...)
# Using ProcessGroupCudaP2P while specifying ProcessGroupNCCL.Options
pg_options = ProcessGroupNCCL.Options()
dist.init_process_group(backend="cuda_p2p", pg_options=pg_options, ...)
# Using ProcessGroupCudaP2P while specifying both
# ProcessGroupCudaP2P.Options and ProcessGroupNCCL.Options
pg_options = ProcessGroupCudaP2P.Options()
pg_options.nccl_options = ProcessGroupNCCL.Options()
dist.init_process_group(backend="cuda_p2p", pg_options=pg_options, ...)
# Down-casting the backend to access p2p buffers for cuda_p2p specific
# optimizations
if is_cuda_p2p_group(group):
backend = get_cuda_p2p_backend(group)
if required_p2p_buffer_size > backend.get_buffer_size():
# fallback
p2p_buffer = backend.get_p2p_buffer(...)
else:
# fallback
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/122163
Approved by: https://github.com/wanchaol
## Context
This stack prototypes automatic micro-pipelining of `all-gather -> matmul` and `matmul -> reduce-scatter` via Inductor. The idea originates from the paper [Overlap Communication with Dependent Computation via
Decomposition in Large Deep Learning Models](https://dl.acm.org/doi/pdf/10.1145/3567955.3567959). The implementation and some key optimizations are heavily influenced by @lw's implementation in xformers.
The stack contains several components:
- `ProcessGroupCudaP2P` - a thin wrapper around `ProcessGroupNCCL`. It in addition maintains a P2P workspace that enables SM-free, one-sided P2P communication which is needed for optimal micro-pipelining.
- `fused_all_gather_matmul` and `fused_matmul_reduce_scatter` dispatcher ops.
- Post-grad fx pass that detects `all-gather -> matmul` and `matmul -> reduce-scatter` and replaces them with the fused dispatcher ops.
To enable the prototype feature:
- Set the distributed backend to `cuda_p2p`.
- Set `torch._inductor.config._micro_pipeline_tp` to `True`.
*NOTE: the prototype sets nothing in stone w.r.t to each component's design. The purpose is to have a performant baseline with reasonable design on which each component can be further improved.*
## Benchmark
Setup:
- 8 x H100 (500W) + 3rd gen NVSwitch.
- Llama3 8B training w/ torchtitan.
- 8-way TP. Reduced the number of layers from 32 to 8 for benchmarking purpose.
Trace (baseline): https://interncache-all.fbcdn.net/manifold/perfetto-artifacts/tree/ui/index.html#!/?url=https://interncache-all.fbcdn.net/manifold/perfetto_internal_traces/tree/shared_trace/yifu_tmpjaz8zgx0
<img width="832" alt="image" src="https://github.com/pytorch/pytorch/assets/4156752/4addba77-5abc-4d2e-93ea-f68078587fe1">
Trace (w/ micro pipelining): https://interncache-all.fbcdn.net/manifold/perfetto-artifacts/tree/ui/index.html#!/?url=https://interncache-all.fbcdn.net/manifold/perfetto_internal_traces/tree/shared_trace/yifu_tmpn073b4wn
<img width="963" alt="image" src="https://github.com/pytorch/pytorch/assets/4156752/4f44e78d-8196-43ab-a1ea-27390f07e9d2">
## This PR
`ProcessGroupCudaP2P` is a thin wrapper around `ProcessGroupNCCL`. By default, it routes all collectives to the underlying `ProcessGroupNCCL`. In addition, `ProcessGroupCudaP2P` initializes a P2P workspace that allows direct GPU memory access among the members. The workspace can be used in Python to optimize intra-node communication patterns or to create custom intra-node collectives in CUDA.
`ProcessGroupCudaP2P` aims to bridge the gap where certain important patterns can be better optimized via fine-grained P2P memory access than with collectives in the latest version of NCCL. It is meant to complement NCCL rather than replacing it.
Usage:
```
# Using ProcessGroupCudaP2P
dist.init_process_group(backend="cuda_p2p", ...)
# Using ProcessGroupCudaP2P while specifying ProcessGroupCudaP2P.Options
pg_options = ProcessGroupCudaP2P.Options()
dist.init_process_group(backend="cuda_p2p", pg_options=pg_options, ...)
# Using ProcessGroupCudaP2P while specifying ProcessGroupNCCL.Options
pg_options = ProcessGroupNCCL.Options()
dist.init_process_group(backend="cuda_p2p", pg_options=pg_options, ...)
# Using ProcessGroupCudaP2P while specifying both
# ProcessGroupCudaP2P.Options and ProcessGroupNCCL.Options
pg_options = ProcessGroupCudaP2P.Options()
pg_options.nccl_options = ProcessGroupNCCL.Options()
dist.init_process_group(backend="cuda_p2p", pg_options=pg_options, ...)
# Down-casting the backend to access p2p buffers for cuda_p2p specific
# optimizations
if is_cuda_p2p_group(group):
backend = get_cuda_p2p_backend(group)
if required_p2p_buffer_size > backend.get_buffer_size():
# fallback
p2p_buffer = backend.get_p2p_buffer(...)
else:
# fallback
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/122163
Approved by: https://github.com/wanchaol
Update ruff to 0.4.1 .
This version fixes a lot false negatives/false positives, is 20-40% faster, and has various other bug fixes.
Below is a before and after table showing the execution time of ruff lint and ruff format in milliseconds courtesy of https://astral.sh/blog/ruff-v0.4.0
| Repository | Linter (v0.3) | Linter (v0.4) | Formatter (v0.3) | Formatter (v0.4) |
|----------------------------------------------------|---------------|---------------|------------------|------------------|
| [pytorch/pytorch](https://github.com/pytorch/pytorch) | 328.7 | 251.8 | 351.1 | 274.9 |
Pull Request resolved: https://github.com/pytorch/pytorch/pull/124549
Approved by: https://github.com/ezyang
This replaces a bunch of unnecessary lambdas with the operator package. This is semantically equivalent, but the operator package is faster, and arguably more readable. When the FURB rules are taken out of preview, I will enable it as a ruff check.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/116027
Approved by: https://github.com/malfet
Applies the remaining flake8-comprehension fixes and checks. This changes replace all remaining unnecessary generator expressions with list/dict/set comprehensions which are more succinct, performant, and better supported by our torch.jit compiler. It also removes useless generators such as 'set(a for a in b)`, resolving it into just the set call.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94676
Approved by: https://github.com/ezyang
Preferring dash over underscore in command-line options. Add `--command-arg-name` to the argument parser. The old arguments with underscores `--command_arg_name` are kept for backward compatibility.
Both dashes and underscores are used in the PyTorch codebase. Some argument parsers only have dashes or only have underscores in arguments. For example, the `torchrun` utility for distributed training only accepts underscore arguments (e.g., `--master_port`). The dashes are more common in other command-line tools. And it looks to be the default choice in the Python standard library:
`argparse.BooleanOptionalAction`: 4a9dff0e5a/Lib/argparse.py (L893-L895)
```python
class BooleanOptionalAction(Action):
def __init__(...):
if option_string.startswith('--'):
option_string = '--no-' + option_string[2:]
_option_strings.append(option_string)
```
It adds `--no-argname`, not `--no_argname`. Also typing `_` need to press the shift or the caps-lock key than `-`.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94505
Approved by: https://github.com/ezyang, https://github.com/seemethere
Fix use-dict-literal pylint suggestions by changing `dict()` to `{}`. This PR should do the change for every Python file except test/jit/test_list_dict.py, where I think the intent is to test the constructor.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/83718
Approved by: https://github.com/albanD
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/62662
Replaced the methods set_tensor(.) and get_tensor() in the python exposed API from the C++ logic with buffer() and set_buffer(.) to be a cleaner interface.
Reviewed By: SciPioneer
Differential Revision: D30012869
fbshipit-source-id: bd8efab583dd89c96f9aeb3dd48a12073f0b1482
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/61507
Benchmark Python-only DDP vs production C++ based DistributedDataParallel.
- Implemented a pure python DDP: PythonDDP with support of SYNC and ASYNC reduction
- Added compare_ddp to measure the difference in forward and backward step
Kudos on Shen and Yi for the great idea.
Test Plan:
Test on DevGPUS with 2 CUDA devices.
$python compare_ddp.py
Python only DDP has slightly better (-1%) forward performance and slightly slower (2%-20%) backward performance.
This suggested that we need to keep C++ Core since the maximum latency increase can be 20%. See README.md for details.
Imported from OSS
Differential Revision:
D29685364
D29685364
Reviewed By: mrshenli
Pulled By: bowangbj
fbshipit-source-id: 429e4473fac0ec4c70d6db12d946d2636dd6477a
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/60631
Per #48360, speed up `Transformer.generate_square_subsequent_mask`. New impl is informally ~5x faster, though absolute difference is probably small.
PR includes Python and C++ versions as well as a couple of places where the previous impl had been copied around.
Test Plan: Imported from OSS
Reviewed By: jbschlosser, albanD
Differential Revision: D29356673
Pulled By: bhosmer
fbshipit-source-id: 4c062ba0ead61a445aeef451c78777bf0b3a631e
Summary:
* Open json config file safely using a context manager (using a with block).
* This will make sure that the file closed even if an exception is raised.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/58077
Reviewed By: anjali411
Differential Revision: D28711177
Pulled By: H-Huang
fbshipit-source-id: 597ba578311b1f1d6706e487872db4e784c78c3c
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/57925
1. adds test_scripts.py that will run added scripts and verify that there are no errors
2. adds local ddp_nccl_allreduce experiment script
test with command `pytest test_scripts.py`
Test Plan: Imported from OSS
Reviewed By: agolynski
Differential Revision: D28382452
Pulled By: gcramer23
fbshipit-source-id: 21028a990ebfedf1aad6b007a723c02403e8bea8
Summary:
Context: https://github.com/pytorch/pytorch/pull/53299#discussion_r587882857
These are the only hand-written parts of this diff:
- the addition to `.github/workflows/lint.yml`
- the file endings changed in these four files (to appease FB-internal land-blocking lints):
- `GLOSSARY.md`
- `aten/src/ATen/core/op_registration/README.md`
- `scripts/README.md`
- `torch/csrc/jit/codegen/fuser/README.md`
The rest was generated by running this command (on macOS):
```
git grep -I -l ' $' -- . ':(exclude)**/contrib/**' ':(exclude)third_party' | xargs gsed -i 's/ *$//'
```
I looked over the auto-generated changes and didn't see anything that looked problematic.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/53406
Test Plan:
This run (after adding the lint but before removing existing trailing spaces) failed:
- https://github.com/pytorch/pytorch/runs/2043032377
This run (on the tip of this PR) succeeded:
- https://github.com/pytorch/pytorch/runs/2043296348
Reviewed By: walterddr, seemethere
Differential Revision: D26856620
Pulled By: samestep
fbshipit-source-id: 3f0de7f7c2e4b0f1c089eac9b5085a58dd7e0d97
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/50860
Since fairscale.nn.Pipe still uses 'balance' and 'devices' parameters,
other frameworks like fairseq still use these parameters. As a result, the
`convert_to_balance` method is a nice utility to use for migrating to PyTorch
Pipe without changing a lot of code in other frameworks.
In addition to this I've renamed the method to be more illustrative of what it
does and also allowed an optional devices parameter.
ghstack-source-id: 120430775
Test Plan:
1) waitforbuildbot
2) Tested with fairseq
Reviewed By: SciPioneer
Differential Revision: D25987273
fbshipit-source-id: dccd42cf1a74b08c876090d3a10a94911cc46dd8
Summary:
Quiet errors from flake8. Only a couple of code changes for deprecated Python syntax from before 2.4. The rest is just adding noqa markers.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/48453
Reviewed By: mruberry
Differential Revision: D25181871
Pulled By: ngimel
fbshipit-source-id: f8d7298aae783b1bce2a46827b088fc390970641
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/35198
The need for this tool was motivated by #28883. In the past, we have
done ad-hoc benchmarking, but it's time for something more structured.
It would be nice to add more model architectures so that we can get a
full picture of the performance impact of a code change simply by
running this suite a few times.
Test Plan: Imported from OSS
Differential Revision: D20591296
Pulled By: mrshenli
fbshipit-source-id: ee66ce0ebca02086453b02df0a94fde27ab4be49
