This should fix https://x.com/wightmanr/status/1953147089518772254?t=ng_R4t0-tRhO_qQE8NqOhw&s=19. Still working on adding a reasonable test.
You can see more of a description of the problem in the code comments. But the TLDR is that:
* When using DDPOptimizer, we partition the graph and compile several subgraphs. So 1 dynamo graphs becomes N AOT/inductor artifacts
* We have some existing logic to stash graph metadata (`fw_metadata`) in dynamo's TracingContext. When using DDPOptimizer, we generate one `fw_metadata` per **AOT** graph, and we stash it on the 1 TracingContext from dynamo. So we end up clobbering the `fw_metadata` for graph i-1 when AOT and inductor start compiling graph i
* This is normally ok, but it becomes a problem if inductor ever wants to read from this `fw_metadata` during **backward compilation**. Why? We (by default) compile the backwards lazily. So when using DDPOptimizer, we will compile backward graph N, then bw graph N-1, etc. But... at the time that we have stated compiling bw graph N-1, its corresponding fw_metadata has already been clobbered! So we end up reusing graph N's metadata for all of our backward graph compilations. With donated buffer metadata, that means we end up donated and writing into incorrect input buffers
The fix that I added was to add more dedicated DDPOptimizer metadata into the TracingContext, so we can properly switch between these N different `fw_metadata` objects in the backward.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/160745
Approved by: https://github.com/ezyang, https://github.com/zou3519
* Automatically applies ruff rule 401. Turns loops into equivalent list comprehensions which are faster and do not leak the scope of the loop variables.
* list comprehensions not only often have better typing, but are 50+% faster than for loops on overhead. They also preserve length information etc and are better for the interpreter to optimize.
* Manually went back and made mypy happy after the change.
* Also fixed style lints in files covered by flake8 but not by pyfmt
Pull Request resolved: https://github.com/pytorch/pytorch/pull/140980
Approved by: https://github.com/justinchuby, https://github.com/malfet
This PR fixes Issue #111279.
While #111279 reported the issue with `MultiheadAttention`, a minimal reproduction would be:
```python
class ToyModel(nn.Module):
def __init__(self,):
super().__init__()
self.linear = nn.Linear(128, 10)
def forward(self, x: torch.Tensor) -> torch.Tensor:
return self.linear.forward(x) # Error
# return self.linear(x) # OK
```
Dynamo treats `self.linear(x)` as `call_module` while treating `self.linear.forward(x)` as a [`get_attr` and a `call_method`](https://github.com/pytorch/pytorch/blob/main/torch/_dynamo/variables/nn_module.py#L358-L378). However, existing DDPOptimizer assumes, for a `get_attr` node, `getattr(gm, node.target)` gives a tensor with the `requires_grad` attribute. Existing DDPOptimizer also does not support `call_method` nodes.
This PR adds support for `call_method` and check on `get_attr`. It also checks if a module's parameters have been added to a bucket to support multiple method calls from the same module.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/121771
Approved by: https://github.com/yf225
# Note: Returning Fake Tensors on First AOT Autograd Call
#
# Inductor will optimize strides of outputs when it deems it profitable.
# For instance, converting to channels last. When we split the graph here
# into multiple inductor compilations, we need to make sure that the
# output strides of one compilation is appropriately passed to the subsequent
# compilations. However, the mapping from inductor output to dynamo output
# is non-trivial due to aot_autograd's deduping, de-aliasing, mutation, re-writing,
# subclass handling, etc. In order to replay all this logic we set a flag such that
# the first invocation of inductor in aot_autograd will return Fake Tensors with
# appropriate strides. Then, all of aot autograd's runtime logic is replayed.
# This gives us the appropriately strided outputs here which will reflect runtime strides.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120523
Approved by: https://github.com/yf225, https://github.com/bdhirsh
Overall design: https://docs.google.com/document/d/1CX_hJ0PNy9f3R1y8TJrfkSeLkvGjjjLU84BSXgS2AZ8/edit
How to read the diff:
* Most files are me augmenting pre-existing logging with structured variants. For the most part it's simple (esp FX graphs, which have a canonical string representation); it gets more complicated when I decided to JSON-ify some data structure instead of keeping the ad hoc printing (notably, guards and dynamo output graph sizes)
* torch/_functorch/_aot_autograd/collect_metadata_analysis.py is some unrelated fixes I noticed while auditing artifact logs
* torch/_logging/_internal.py has the actual trace log implementation. The trace logger is implement as a logger named torch.__trace which is disconnected from the logging hierarchy. It gets its own handler and formatter (TorchLogsFormatter with _is_trace True). `trace_structured` is the main way to emit a trace log. Unusually, there's a separate "metadata" and "payload" field. The metadata field should not be too long (as it is serialized as a single line) and is always JSON (we put contextual things like compile id in it); the payload field can be long and is emitted after the metadata log line and can span multiple lines.
* torch/_logging/structured.py contains some helpers for converting Python data structures into JSON form. Notably, we have a string interning implementation here, which helps reduce the cost of serializing filenames into the log.
* test/dynamo/test_structured_trace.py the tests are cribbed from test_logging.py, but all rewritten to use expect tests on munged versions of what we'd actually output. Payloads are never tested, since they tend not be very stable.
https://github.com/ezyang/tlparse is a POC Rust program that can interpret these logs.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120289
Approved by: https://github.com/Skylion007
ghstack dependencies: #120712
Overall design: https://docs.google.com/document/d/1CX_hJ0PNy9f3R1y8TJrfkSeLkvGjjjLU84BSXgS2AZ8/edit
How to read the diff:
* Most files are me augmenting pre-existing logging with structured variants. For the most part it's simple (esp FX graphs, which have a canonical string representation); it gets more complicated when I decided to JSON-ify some data structure instead of keeping the ad hoc printing (notably, guards and dynamo output graph sizes)
* torch/_functorch/_aot_autograd/collect_metadata_analysis.py is some unrelated fixes I noticed while auditing artifact logs
* torch/_logging/_internal.py has the actual trace log implementation. The trace logger is implement as a logger named torch.__trace which is disconnected from the logging hierarchy. It gets its own handler and formatter (TorchLogsFormatter with _is_trace True). There's a teensy bit of FB specific code to automatically enable trace logging if a /logs directory exists. `trace_structured` is the main way to emit a trace log. Unusually, there's a separate "metadata" and "payload" field. The metadata field should not be too long (as it is serialized as a single line) and is always JSON (we put contextual things like compile id in it); the payload field can be long and is emitted after the metadata log line and can span multiple lines.
* torch/_logging/structured.py contains some helpers for converting Python data structures into JSON form. Notably, we have a string interning implementation here, which helps reduce the cost of serializing filenames into the log.
* test/dynamo/test_structured_trace.py the tests are cribbed from test_logging.py, but all rewritten to use expect tests on munged versions of what we'd actually output. Payloads are never tested, since they tend not be very stable.
https://github.com/ezyang/tlparse is a POC Rust program that can interpret these logs.
Testing that the fbcode detection works at https://www.internalfb.com/mlhub/pipelines/runs/fblearner/534553450 (Meta-only)
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/120289
Approved by: https://github.com/Skylion007
The original motivation for MYPYINDUCTOR was a faster type checking configuration that only checked a subset of files. With the removal of `follow_imports = ignore`, we are now able to use dmypy to do fast incremental typechecking, eliminating the need for this.
Perhaps erroneously, when I tee'ed up this PR I elected to delete the `follow_imports = skip` designations in the mypy-inductor.ini. This lead to a number of extra type error suppressions that I manually edited. You will need to review.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/118432
Approved by: https://github.com/Skylion007
ghstack dependencies: #118414, #118418
There are now 3 ways to see logs from ddpoptimzer.
1) TORCH_LOGS="distributed"
2) TORCH_LOGS="dynamo"
3) TORCH_LOGS="torch._dynamo.backends.distributed"
(1 and 2 are different supersets of 3 that also include other content)
Note: ddp_graphs is still a separate 'artifact' logger, which just
includes graph dumps from the graph-splitting process.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/114376
Approved by: https://github.com/wanchaol
Summary:
The primary problem we are setting out to solve here is fake tensor freshness. Before this PR, fake tensors after dynamo represented fake tensors *at the end* of trace, so subsequent retraces like aot_autograd would start off with fake tensors in the wrong (end result) state, rather than their expected fresh state. The solution here is to start a fresh fake mode, and re-fakify the tensors. The nuance comes from ensuring that symbols are uniformly created for the symbolic sizes and strides of the tensor.
This PR is the result of *a lot* of back and forth with ezyang and eellison. Initially, the first pass at this was not super different from what we have in the PR - the broad strokes were the same:
1) We cache source->symbol in shape_env
2) We pass policy objects around, stored at dynamo fakificaiton time, and reused for later fakification
3) We create a new fake mode for backends
(from https://github.com/pytorch/pytorch/pull/113605/files)
This is ugly, and has some layering violations. We detoured our decision making through a few other alternatives. Immutable/mutable fake tensor mode was the most interesting alternative, https://github.com/pytorch/pytorch/pull/113653, and was struck down on concerns of complexity in fake mode combined with it not covering all edge cases. We also detoured on what to do about tensor memoization returning back potentially different tensors than requested, and if that was an anti pattern (it is) we want to hack in with the symbol cache (we don't).
We went back to the drawing board here, but with a few concessions:
1) the cache for source->symbol must live outside of shape_env, for both lifecycle, and layering reasons
2) A good amount of work needs to be done to pipe policy around fake_mode and meta_utils correctly, to cover all the cases (ezyang did this)
cc penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng
imported-using-ghimport
Test Plan: Imported from OSS
Reviewed By: huydhn, Chillee
Differential Revision: D51566250
Pulled By: voznesenskym
Pull Request resolved: https://github.com/pytorch/pytorch/pull/114526
Approved by: https://github.com/Chillee, https://github.com/huydhn
The primary problem we are setting out to solve here is fake tensor freshness. Before this PR, fake tensors after dynamo represented fake tensors *at the end* of trace, so subsequent retraces like aot_autograd would start off with fake tensors in the wrong (end result) state, rather than their expected fresh state. The solution here is to start a fresh fake mode, and re-fakify the tensors. The nuance comes from ensuring that symbols are uniformly created for the symbolic sizes and strides of the tensor.
This PR is the result of *a lot* of back and forth with @ezyang and @eellison. Initially, the first pass at this was not super different from what we have in the PR - the broad strokes were the same:
1) We cache source->symbol in shape_env
2) We pass policy objects around, stored at dynamo fakificaiton time, and reused for later fakification
3) We create a new fake mode for backends
(from https://github.com/pytorch/pytorch/pull/113605/files)
This is ugly, and has some layering violations. We detoured our decision making through a few other alternatives. Immutable/mutable fake tensor mode was the most interesting alternative, https://github.com/pytorch/pytorch/pull/113653, and was struck down on concerns of complexity in fake mode combined with it not covering all edge cases. We also detoured on what to do about tensor memoization returning back potentially different tensors than requested, and if that was an anti pattern (it is) we want to hack in with the symbol cache (we don't).
We went back to the drawing board here, but with a few concessions:
1) the cache for source->symbol must live outside of shape_env, for both lifecycle, and layering reasons
2) A good amount of work needs to be done to pipe policy around fake_mode and meta_utils correctly, to cover all the cases (@ezyang did this)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/113926
Approved by: https://github.com/ezyang, https://github.com/eellison
From talking to @wconstab, we agreed that because of the way DDPOptimizer is written, it is (sort of) incompatible with inductor's `keep_output_stride=False` optimizations (and will cause silent correctness problems if you use them ogether). Added an assertion.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/108235
Approved by: https://github.com/wconstab
ghstack dependencies: #108081
Subgraphs are partitions cut out of a whole graph. Outputs of a subgraph are either global outputs of the original graph, or can be outputs of a partition that feed inputs of the subsequent partition. Subgraphs are created using the fx utility 'passes.split_module', which requires that each partition
have at least one output node.
In cases where DDPOptimizer asked the partitioner to cut the graph around a set of nodes which only
performed inplace mutation, the partitioner could be left trying to create a subgraph with no output nodes, violating its assumptions.
To circumvent this, DDPOptimizer can expand the set of nodes marked for inclusion in a subgraph that has no outputs until it includes a node that is an output for that subgraph. It still traverses nodes of the original graph in reverse order and only considers widening a subgraph by iterating further in reverse order than it would have ordinarily done (past the cut point dictated by paramter count). It may still be possible the subgraph reaches the input node of the graph without satisfying the subgraph-output condition, in which case an error would still be raised by the partitioner.
Fixes#103385
Pull Request resolved: https://github.com/pytorch/pytorch/pull/103488
Approved by: https://github.com/anijain2305
This replaces fake_mode_from_tensors but it preferentially looks for
fake_mode in TracingContext and also if there is an active fake mode
on the dispatch stack, before groveling in tensors to find it.
This advances PegasusForCausalLM, which was previously failing because
we generated a graph that had a parameter (non-fake) and a SymInt,
and thus previously we failed to detect the correct fake mode.
Signed-off-by: Edward Z. Yang <ezyang@meta.com>
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98321
Approved by: https://github.com/voznesenskym
GraphModules that were created during DDPOptimizer graph breaking
lacked `compile_subgraph_reason`, which caused an exception when
running .explain().
Now the reason is provided and users can use .explain() to find out
that DDPOptimizer is causing graph breaks.
Fixes#94579
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94749
Approved by: https://github.com/voznesenskym
