[autograd.Function] add nice error message for incorrect usage of vmap #92023
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This PR: - adds a nice error message if the user doesn't follow the API of the vmap staticmethod correctly. That is, the user must return two arguments from the vmap staticmethod API: (outputs, out_dims), and out_dims must be a PyTree with either the same structure as `outputs` our be broadcastable to the same structure as `outputs`. - Fixes an edge case for out_dims=None. out_dims is allowed to be None, but wrap_outputs_maintaining_identity was treating "None" as "This is not the vmap case" Test Plan: - new tests [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/92023
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 6b7f0fa: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Jan 11, 2023
zou3519
added
triaged
…sage of vmap" This PR: - adds a nice error message if the user doesn't follow the API of the vmap staticmethod correctly. That is, the user must return two arguments from the vmap staticmethod API: (outputs, out_dims), and out_dims must be a PyTree with either the same structure as `outputs` our be broadcastable to the same structure as `outputs`. - Fixes an edge case for out_dims=None. out_dims is allowed to be None, but wrap_outputs_maintaining_identity was treating "None" as "This is not the vmap case" Test Plan: - new tests [ghstack-poisoned]
…sage of vmap" This PR: - adds a nice error message if the user doesn't follow the API of the vmap staticmethod correctly. That is, the user must return two arguments from the vmap staticmethod API: (outputs, out_dims), and out_dims must be a PyTree with either the same structure as `outputs` our be broadcastable to the same structure as `outputs`. - Fixes an edge case for out_dims=None. out_dims is allowed to be None, but wrap_outputs_maintaining_identity was treating "None" as "This is not the vmap case" Test Plan: - new tests [ghstack-poisoned]
soulitzer
reviewed
Jan 13, 2023
| @@ -267,16 +291,14 @@ def custom_function_call_vmap(interpreter, autograd_function, *operands): | |||
| return custom_function_call(autograd_function, *operands) | |||
|
|
|||
| with interpreter.lower(): | |||
| unwrapped_output, out_dims = autograd_function.vmap(info, in_dims, *unwrapped_operands) | |||
| result = autograd_function.vmap(info, in_dims, *unwrapped_operands) | |||
| validate_vmap_returns_two(result) | |||
There was a problem hiding this comment.
nit: I think we might as well fully describe it here, then a reader won't even have to read the function to figure out what it does - "validate_vmap_returns_tuple_of_two_elements".
| @staticmethod | ||
| def vmap(info, in_dims, input): | ||
| assert in_dims == (0,) | ||
| return torch.zeros(input.shape[1:], device=input.device), None |
There was a problem hiding this comment.
Why does this work? Does _broadcast_to_and_flatten do something special if out_dims is None? Otherwise it seems that the wrap_fn we define should just not wrap the output in a batched tensor.
There was a problem hiding this comment.
Your reading is correct -- _broadcast_to_and_flatten doesn't have any special logic for None. It is intended that vmap does not construct a BatchedTensor in this case.
The expected behavior of this is that the output (call it zeros) is not wrapped in a BatchedTensor. That's OK. Inside of vmap, a Tensor is either a BatchedTensor or a regular Tensor. This is the latter case.
There are two cases:
- Eventually the regular Tensor interacts with a BatchedTensor. E.g.
regular_tensor * batched_tensor. This produces a BatchedTensor. - The regular Tensor gets returned from the function being vmapped over. vmap has special handling to take this Tensor and expand out the batch dimension.
soulitzer
approved these changes
Jan 13, 2023
| @@ -145,11 +145,14 @@ def jvp(ctx, *tangents): | |||
| ) | |||
| return Generated | |||
|
|
|||
| NO_OUT_DIMS = "not specified" | |||
There was a problem hiding this comment.
This is to avoid confusing None with "not specified" right? Maybe we could have a comment about what happens in the None case?
…sage of vmap" This PR: - adds a nice error message if the user doesn't follow the API of the vmap staticmethod correctly. That is, the user must return two arguments from the vmap staticmethod API: (outputs, out_dims), and out_dims must be a PyTree with either the same structure as `outputs` our be broadcastable to the same structure as `outputs`. - Fixes an edge case for out_dims=None. out_dims is allowed to be None, but wrap_outputs_maintaining_identity was treating "None" as "This is not the vmap case" Test Plan: - new tests [ghstack-poisoned]
github-actions
bot
added
the
module: mkldnn
|
Hi @zou3519 , do you need to make some changes in |
…sage of vmap" This PR: - adds a nice error message if the user doesn't follow the API of the vmap staticmethod correctly. That is, the user must return two arguments from the vmap staticmethod API: (outputs, out_dims), and out_dims must be a PyTree with either the same structure as `outputs` our be broadcastable to the same structure as `outputs`. - Fixes an edge case for out_dims=None. out_dims is allowed to be None, but wrap_outputs_maintaining_identity was treating "None" as "This is not the vmap case" Test Plan: - new tests cc gujinghui @PenghuiCheng @XiaobingSuper jianyuh jgong5 mingfeima sanchitintel ashokei jingxu10 min-jean-cho yanbing-j @Guobing-Chen @Xia-Weiwen [ghstack-poisoned]
yanbing-j
removed
the
module: mkldnn
…sage of vmap" This PR: - adds a nice error message if the user doesn't follow the API of the vmap staticmethod correctly. That is, the user must return two arguments from the vmap staticmethod API: (outputs, out_dims), and out_dims must be a PyTree with either the same structure as `outputs` our be broadcastable to the same structure as `outputs`. - Fixes an edge case for out_dims=None. out_dims is allowed to be None, but wrap_outputs_maintaining_identity was treating "None" as "This is not the vmap case" Test Plan: - new tests cc gujinghui @PenghuiCheng @XiaobingSuper jianyuh jgong5 mingfeima sanchitintel ashokei jingxu10 min-jean-cho yanbing-j @Guobing-Chen @Xia-Weiwen [ghstack-poisoned]
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Stack from ghstack:
This PR:
vmap staticmethod correctly. That is, the user must return two
arguments from the vmap staticmethod API: (outputs, out_dims), and
out_dims must be a PyTree with either the same structure as
outputsour be broadcastable to the same structure as
outputs.but wrap_outputs_maintaining_identity was treating "None" as "This is
not the vmap case"
Test Plan:
cc @gujinghui @PenghuiCheng @XiaobingSuper @jianyuh @jgong5 @mingfeima @sanchitintel @ashokei @jingxu10 @min-jean-cho @yanbing-j @Guobing-Chen @Xia-Weiwen