[Dynamo] Symbolic shape guards #87570
Conversation
…h into voz/sym_shape_guards
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/87570
Note: Links to docs will display an error until the docs builds have been completed. ✅ No Failures, 2 PendingAs of commit ad0f43f: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
| ), | ||
| **VariableTracker.propagate(self, [a, b]), | ||
| ) | ||
|
|
There was a problem hiding this comment.
This fixes "TypeError: maximum(): argument 'other' (position 2) must be Tensor, not int)" on vision_maskrcnn in symbolic-shapes branch, we're going to patch this into the branch.
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
|
@pytorchbot merge -f "https://hud.pytorch.org/pr/87570 says its done, but github UI not updating" |
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
agh caching bug, I guess. I needn't have forced. |
|
Hey @voznesenskym. |
|
|
||
| def _record_ref(e, index, symbol, kind): | ||
| sym_expr = _extract(symbol) | ||
| if sym_expr: |
There was a problem hiding this comment.
if sym_expr is None otherwise you'll drop even when sym_expr is 0
**Introduces symbolic shape guards into dynamo.** In this PR, we take the existing fake tensor infra and plumbing in dynamo and we start passing a shape_env around. This shape_env does not get plumbed down to middle layers / backend yet - it only collects expressions from frontend invocations at the moment. We then translate these expressions into guards at the point where we take other guards installed throughout dynamo - and add them to check_fn. Part 1 of https://docs.google.com/document/d/1QJ-M4zfMkD-fjHIqW089RptjLl9EgozZGCceUbvmgfY/edit# cc @jansel @lezcano @fdrocha @mlazos @soumith @yanboliang @penguinwu @anijain2305 Pull Request resolved: #87570 Approved by: https://github.com/ezyang
| from . import UserDefinedObjectVariable | ||
|
|
||
| return UserDefinedObjectVariable(example_value) | ||
| elif isinstance(example_value, torch.SymIntNode): |
There was a problem hiding this comment.
should this also check for SymFloatNode?
**Introduces symbolic shape guards into dynamo.** In this PR, we take the existing fake tensor infra and plumbing in dynamo and we start passing a shape_env around. This shape_env does not get plumbed down to middle layers / backend yet - it only collects expressions from frontend invocations at the moment. We then translate these expressions into guards at the point where we take other guards installed throughout dynamo - and add them to check_fn. Part 1 of https://docs.google.com/document/d/1QJ-M4zfMkD-fjHIqW089RptjLl9EgozZGCceUbvmgfY/edit# cc @jansel @lezcano @fdrocha @mlazos @soumith @yanboliang @penguinwu @anijain2305 Pull Request resolved: pytorch#87570 Approved by: https://github.com/ezyang
**Introduces symbolic shape guards into dynamo.** In this PR, we take the existing fake tensor infra and plumbing in dynamo and we start passing a shape_env around. This shape_env does not get plumbed down to middle layers / backend yet - it only collects expressions from frontend invocations at the moment. We then translate these expressions into guards at the point where we take other guards installed throughout dynamo - and add them to check_fn. Part 1 of https://docs.google.com/document/d/1QJ-M4zfMkD-fjHIqW089RptjLl9EgozZGCceUbvmgfY/edit# cc @jansel @lezcano @fdrocha @mlazos @soumith @yanboliang @penguinwu @anijain2305 Pull Request resolved: pytorch#87570 Approved by: https://github.com/ezyang
Introduces symbolic shape guards into dynamo.
In this PR, we take the existing fake tensor infra and plumbing in dynamo and we start passing a shape_env around. This shape_env does not get plumbed down to middle layers / backend yet - it only collects expressions from frontend invocations at the moment. We then translate these expressions into guards at the point where we take other guards installed throughout dynamo - and add them to check_fn.
Part 1 of https://docs.google.com/document/d/1QJ-M4zfMkD-fjHIqW089RptjLl9EgozZGCceUbvmgfY/edit#
cc @jansel @lezcano @fdrocha @mlazos @soumith @yanboliang @penguinwu @anijain2305