Implement narrow from a regular tensor to jagged tensor #112770

ani300 · 2023-11-02T18:53:38Z

Stack from ghstack (oldest at bottom):

-> Implement narrow from a regular tensor to jagged tensor #112770

cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @aakhundov @kadeng

[ghstack-poisoned]

pytorch-bot · 2023-11-02T18:53:41Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/112770

📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
❓ Need help or want to give feedback on the CI? Visit the bot commands wiki or our office hours

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 03a94ee with merge base d4c810c ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

ghstack-source-id: ec6d15238467d273092f99f675298e4aa8c42daa Pull Request resolved: #112770

cpuhrsch · 2023-11-02T20:01:20Z

torch/nested/__init__.py

+
+
+
+def narrow(tensor: Tensor, dim: int, start: Union[SymInt, Tensor], length: Union[SymInt, Tensor], layout=torch.jagged):


I don't think we should add a layout argument here. It doesn't follow the narrow syntax for regular tensors. If the user wants to change the layout they can use to. Since this is a view operation it shouldn't affect performance. Furthermore, layout change is likely not support when this is view only.

I don't think we should add a layout argument here. It doesn't follow the narrow syntax for regular tensors.

This is true, but we're also not limited to the original narrow() syntax since this is a new function in the torch.nested namespace.

So I sort of disagree. It's possible to create a view over a dense buffer with both the expanded jagged layout and the strided layout, and I think the layout arg is a natural extension for indicating which type of NT + metadata is to be constructed.

Also note that nested narrow for jagged is only supported on dim=1 due to representative restrictions, and it's potentially useful to take a ragged view on other dimensions, using the more general strided layout. This function would be the way to do it AFAICT.

If the user wants to change the layout they can use to.

Current semantics are such that this will copy, no? I don't quite see how we can guarantee torch.nested.narrow(...).to(layout=...) is non-copying if it's the difference between a C++ NT and a python subclass jagged layout NT.

torch.nested.narrow(...).to(layout=...) can be made non-copying via torch.compile and pattern matching.

But you're right that it's a new function. Maybe in the future we can extend torch.narrow with similar functionality.

Alright, I'm convinced.

soulitzer · 2023-11-03T03:14:06Z

torch/nested/_internal/nested_tensor.py

+        raise RuntimeError(
+            "When constructing a jagged nested tensor using narrow(), "
+            "your start and length must be a Tensor that broadcasts to input.shape[0] x 1"
+        ) from e


You can use is_expandable_to in torch/_prims_common/__init__.py instead

I need the expanded tensors, so I figured I could kill two birds with one stone, but I agree the explicit check is probably better

torch/nested/_internal/nested_tensor.py

soulitzer · 2023-11-03T03:32:21Z

torch/nested/_internal/ops.py

@@ -36,7 +36,8 @@ def check_schema(schema_str: str, func, *args, **kwargs) -> None:

    arg_type_check_fns = {
        "t": lambda x: isinstance(x, torch.Tensor) and not isinstance(x, NestedTensor),
-        "jt": lambda x: isinstance(x, NestedTensor),
+        "jt": lambda x: isinstance(x, NestedTensor) and x._lengths is None,
+        "jt_nc": lambda x: isinstance(x, NestedTensor),


This new type won't be too useful today as you won't actually be able to register two funcs to the same aten op overload even if those ops have different schemas.

For now, what you probably want to do is just to branch inside whatever op you are trying to implement.

Perhaps in the future we want to go the route of writing a general dispatching mechanism, I'm not sure. Or its also entirely possible that we'd revert the t vs jt distinction as well. We don't have many ops implemented, so I think its too early to decide today.

ah... I was trying to get out of adding an if statement to every single function we implement, but instead I will change the register_func call to accept a parameter of whether non-contiguous tensors are allowed or not (given most functions won't accept them at all)

and then add the two paths for the ones that allow them

on second thought, "jt_nc" might not be the best name, but it is the one that allows both the contiguous and noncont. versions to pass though to the actual kernel code vs "jt" only letting the contiguous ones pass to maintain compatibility

yeah I'm okay with your initial approach. I did something similar to that locally when I was playing around with this and it minimized the changes needed to existing op impls

soulitzer · 2023-11-03T03:33:41Z

torch/nested/__init__.py

+        if isinstance(length, SymInt):
+            length = torch.tensor([int(length)])
+
+        nt = jagged_from_tensor_and_lengths(tensor, start, length)


nit: You actually returned 3 things from this function, but the name suggests that you only expected to return one thing.

Do we actually want to return all 3 to the user?

yeah, I'll change it, I was trying to keep it similar to jagged_from_list which returns its offsets and is semantically a very similar function

Yeah I think its okay for the helper function jagged_from_tensor_and_lengths to also return 3, but we don't necessarily want to return all 3 in the public API.

torch/nested/_internal/nested_tensor.py

…sor" [ghstack-poisoned]

ghstack-source-id: 7df806f7755432a844b1c3435f723b160303e54f Pull Request resolved: #112770

jbschlosser

Nice work! I'm curious to see how this looks with SDPA :)

Do you mind adding a few tests into test/test_nestedtensor.py? I think we can check that things nicely error out for all ops when they encounter a non-contiguous jagged NT.

Along that line: I think we want to expand the definition of contiguous() to take lengths into account + add a corresponding test.

torch/nested/__init__.py

torch/nested/_internal/nested_tensor.py

torch/nested/__init__.py

torch/nested/_internal/nested_tensor.py

jbschlosser · 2023-11-03T18:00:56Z

torch/nested/__init__.py

+
+Keyword arguments:
+    layout (:class:`torch.layout`, optional): the desired layout of returned nested tensor.
+        Only strided and jagged layouts are supported. Default: if None, the jagged layout.


nit: None default should indicate strided, as is consistent with the behavior in other places (e.g. nested_tensor() / as_nested_tensor()).

also is None being handled? I didn't see it but I may have just missed it

it is being handled with a RuntimeError

…sor" [ghstack-poisoned]

ghstack-source-id: f712cd7556e787a7b93770f40188e08f4e058654 Pull Request resolved: #112770

…sor" [ghstack-poisoned]

ghstack-source-id: 1075ff48d892296e1cb96e92de4864afa19f1ba1 Pull Request resolved: #112770

torch/nested/_internal/ops.py

…sor" [ghstack-poisoned]

ghstack-source-id: bbcaec97e0ced5496675e725bfef03a66371c0e7 Pull Request resolved: #112770

…sor" cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ghstack-source-id: 94d57818c238208d051ea966a78a5062ead0f883 Pull Request resolved: #112770

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ghstack-source-id: 2bf3fe5a469f26ed06d95dbcb24028c47355f103 Pull Request resolved: #112770

jbschlosser

Looking really nice, thanks for all your hard work!

I'm just concerned about one thing: the format of offsets. Traditionally, the jagged layout defines it to be of shape B + 1 for batch size B, and it includes the final offset (e.g. offsets=[0, 3, 5, 7] with B=3 and a values buffer of shape [7, D]). For the contiguous case, this allows for lengths to be computed with offsets.diff().

From what I can tell, this PR redefines offsets to leave off the final offset, resulting in offsets.shape[0] == lengths.shape[0] == B.

I don't think we can change this for the contiguous case, as it's well-established at this point. For consistency, I'd also argue that the offsets shape should be the same across non-contiguous and contiguous cases. I admit this is a little strange, as the final offset doesn't have a ton of utility when lengths are defined separately.

@soulitzer / @cpuhrsch / @drisspg: what are your opinions on this?

jbschlosser · 2023-11-07T19:49:38Z

torch/nested/_internal/nested_tensor.py

+                ragged_size = get_tensor_id(offsets, coeff=1)
+            else:
+                ragged_size = get_tensor_id(lengths, coeff=1)
+        B = offsets.shape[0]


hm I don't think this is correct. The original (contiguous) jagged layout has offsets of shape B + 1, making lengths computable via offsets.diff(). For the non-contiguous case, I think we should maintain this shape, including the final offset even if it isn't strictly necessary.

jbschlosser · 2023-11-07T19:51:03Z

torch/nested/_internal/nested_tensor.py

@@ -232,6 +250,17 @@ def backward(ctx, gO: NestedTensor):  # type: ignore[override]
        return gO.values(), None, None


+# Not actually a view!
+class ViewNonContiguousNestedFromBuffer(torch.autograd.Function):


note: this will go away when we introduce proper dense -> jagged views, which I'm working on. No action needed for this PR

jbschlosser · 2023-11-07T19:52:34Z

torch/nested/_internal/nested_tensor.py

@@ -278,12 +307,50 @@ def jagged_from_list(
        offsets = torch.cat(
            [
                torch.zeros(1, dtype=torch.int64, device=values.device),
-                torch.tensor([s[0] for s in sizes], device=values.device).cumsum(dim=0),
+                torch.tensor([s[0] for s in sizes[:-1]], device=values.device).cumsum(


as mentioned above, we shouldn't change the offsets format; we should keep this as shape B + 1 with the final offset included

jbschlosser · 2023-11-07T19:53:44Z

torch/nested/_internal/nested_tensor.py

+    else:
+        raise RuntimeError(
+            "When constructing a jagged nested tensor using narrow(), "
+            "your start and length must be a Tensor that broadcasts to input.shape[0] x 1"


Suggested change

"your start and length must be a Tensor that broadcasts to input.shape[0] x 1"

"start and length must be Tensors that broadcast to input.shape[0]"

(I believe we removed the need for x 1 in the logic)

jbschlosser · 2023-11-07T19:54:56Z

torch/nested/_internal/nested_tensor.py

+        0, batch_size, dtype=torch.int64, device=tensor.device
+    )
+    # Jagged layout specifies that offsets are stored as int64 on the same device as values.
+    offsets = start_list + offset_lengths


we should add the "final offset" to match the old jagged offsets format with shape B + 1

jbschlosser · 2023-11-07T19:59:26Z

torch/nested/_internal/ops.py

+        split_offsets = torch.cat(
+            (
+                offsets,
+                torch.tensor(
+                    [values.shape[0]], device=offsets.device, dtype=offsets.dtype
+                ),
+            )
+        )
+        return torch.split(values, split_offsets.diff().tolist())


if we maintain the traditional jagged offsets format with shape B + 1, we can just use diff()

jbschlosser · 2023-11-07T20:02:10Z

torch/nested/__init__.py

+    r"""
+Constructs a nested tensor (which might be a view) from :attr:`tensor`, a strided tensor. This follows
+similar semantics to torch.Tensor.narrow, where in the :attr:`dim`-th dimension the new nested tensor
+(maybe view) shows only the elements in the interval `[start, start+length]`. As nested representations


Suggested change

(maybe view) shows only the elements in the interval `[start, start+length]`. As nested representations

(maybe view) shows only the elements in the interval `[start, start+length)`. As nested representations

extreme nit: exclusive upper bound for interval

jbschlosser · 2023-11-07T20:05:25Z

torch/nested/__init__.py

+Constructs a nested tensor (which might be a view) from :attr:`tensor`, a strided tensor. This follows
+similar semantics to torch.Tensor.narrow, where in the :attr:`dim`-th dimension the new nested tensor
+(maybe view) shows only the elements in the interval `[start, start+length]`. As nested representations


wording / formatting nit: rather than mentioning views in-line here, I think it'd be clearer to talk about view / copy behavior for each layout in a separate paragraph after this one

jbschlosser · 2023-11-07T20:57:38Z

from offline discussion: we're in agreement that offsets should consistently be in the old format: shape B + 1

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ani300 · 2023-11-08T18:56:39Z

I've updated all the offsets, hopefully I didn't miss any, but I can't seem to figure out the issue with the recompile happening on dynamo that breaks CI

ghstack-source-id: 20ea5dde599fc8326f9fff1faf742ff2cd62753e Pull Request resolved: #112770

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ghstack-source-id: 952a3948709ec3a3654da4c0b7866d290f042814 Pull Request resolved: #112770

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ghstack-source-id: b95f8e47918dff02a9ac86a08ec3d790be2fbea4 Pull Request resolved: #112770

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ghstack-source-id: 7517fb544cf352a8810067b2ad85ebfdd484893f Pull Request resolved: #112770

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ghstack-source-id: d55799b6027b641eee4815ed9e0a25a7bf9fcf99 Pull Request resolved: #112770

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ghstack-source-id: 2fbbc0e0cd2a59eb6c4dde4ba1d6cc83ffb52b42 Pull Request resolved: #112770

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ghstack-source-id: df56044c88154d07ff6850b347ac6be215c3ad26 Pull Request resolved: #112770

ani300 · 2023-11-13T16:37:52Z

@pytorchbot merge

pytorchmergebot · 2023-11-13T16:39:50Z

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here

Pull Request resolved: pytorch#112770 Approved by: https://github.com/cpuhrsch

cpuhrsch · 2023-11-16T18:31:32Z

torch/nested/__init__.py

@@ -187,3 +188,69 @@ def nested_tensor(tensor_list, *, dtype=None, layout=None, device=None, requires
        return nt
    else:
        raise RuntimeError(f"Specified layout is unsupported for nested tensors: {layout}")
+
+
+def narrow(tensor: Tensor, dim: int, start: Union[int, Tensor], length: Union[int, Tensor], layout=torch.strided) -> Tensor:


Hm, did the documentation for this make it onto the website?

I might have forgotten to commit the changes to the docs adding this function, as I don't see it in the commit

Implement narrow from a regular tensor to jagged tensor

0aa3aef

[ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 2, 2023

Implement narrow from a regular tensor to jagged tensor

71e7f48

ghstack-source-id: ec6d15238467d273092f99f675298e4aa8c42daa Pull Request resolved: #112770

ani300 requested review from jbschlosser and cpuhrsch November 2, 2023 18:54

ani300 changed the title ~~Implement narrow from a regular tensor to jagged tensor~~ [RFC] Implement narrow from a regular tensor to jagged tensor Nov 2, 2023

pytorchbot added the open source label Nov 2, 2023

cpuhrsch requested a review from soulitzer November 2, 2023 19:59

cpuhrsch reviewed Nov 2, 2023

View reviewed changes

cpuhrsch approved these changes Nov 3, 2023

View reviewed changes

soulitzer reviewed Nov 3, 2023

View reviewed changes

Update on "[RFC] Implement narrow from a regular tensor to jagged ten…

e8a7a62

…sor" [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 3, 2023

Implement narrow from a regular tensor to jagged tensor

23b39fb

ghstack-source-id: 7df806f7755432a844b1c3435f723b160303e54f Pull Request resolved: #112770

jbschlosser reviewed Nov 3, 2023

View reviewed changes

Update on "[RFC] Implement narrow from a regular tensor to jagged ten…

85322ba

…sor" [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 3, 2023

Implement narrow from a regular tensor to jagged tensor

5a13e20

ghstack-source-id: f712cd7556e787a7b93770f40188e08f4e058654 Pull Request resolved: #112770

Update on "[RFC] Implement narrow from a regular tensor to jagged ten…

b6c865d

…sor" [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 6, 2023

Implement narrow from a regular tensor to jagged tensor

5ccee64

ghstack-source-id: 1075ff48d892296e1cb96e92de4864afa19f1ba1 Pull Request resolved: #112770

jbschlosser reviewed Nov 6, 2023

View reviewed changes

torch/nested/_internal/ops.py Outdated Show resolved Hide resolved

Update on "[RFC] Implement narrow from a regular tensor to jagged ten…

b5f52a2

…sor" [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 7, 2023

Implement narrow from a regular tensor to jagged tensor

c2bac26

ghstack-source-id: bbcaec97e0ced5496675e725bfef03a66371c0e7 Pull Request resolved: #112770

github-actions bot added the module: dynamo label Nov 7, 2023

Update on "[RFC] Implement narrow from a regular tensor to jagged ten…

dade183

…sor" cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 7, 2023

Implement narrow from a regular tensor to jagged tensor

110f896

ghstack-source-id: 94d57818c238208d051ea966a78a5062ead0f883 Pull Request resolved: #112770

ani300 changed the title ~~[RFC] Implement narrow from a regular tensor to jagged tensor~~ Implement narrow from a regular tensor to jagged tensor Nov 7, 2023

ani300 added the release notes: nested tensor Changes that have a direct impact on nested tensors label Nov 7, 2023

Update on "Implement narrow from a regular tensor to jagged tensor"

9966fcb

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 7, 2023

Implement narrow from a regular tensor to jagged tensor

c5e9f22

ghstack-source-id: 2bf3fe5a469f26ed06d95dbcb24028c47355f103 Pull Request resolved: #112770

jbschlosser reviewed Nov 7, 2023

View reviewed changes

Update on "Implement narrow from a regular tensor to jagged tensor"

28989db

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 8, 2023

Implement narrow from a regular tensor to jagged tensor

943770a

ghstack-source-id: 20ea5dde599fc8326f9fff1faf742ff2cd62753e Pull Request resolved: #112770

Update on "Implement narrow from a regular tensor to jagged tensor"

d48db5a

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 8, 2023

Implement narrow from a regular tensor to jagged tensor

59887b1

ghstack-source-id: 952a3948709ec3a3654da4c0b7866d290f042814 Pull Request resolved: #112770

Update on "Implement narrow from a regular tensor to jagged tensor"

08fb0a6

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 8, 2023

Implement narrow from a regular tensor to jagged tensor

85dfb7b

ghstack-source-id: b95f8e47918dff02a9ac86a08ec3d790be2fbea4 Pull Request resolved: #112770

Update on "Implement narrow from a regular tensor to jagged tensor"

9db790c

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 8, 2023

Implement narrow from a regular tensor to jagged tensor

f2eb2df

ghstack-source-id: 7517fb544cf352a8810067b2ad85ebfdd484893f Pull Request resolved: #112770

Update on "Implement narrow from a regular tensor to jagged tensor"

11ca324

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 8, 2023

Implement narrow from a regular tensor to jagged tensor

d646f8c

ghstack-source-id: d55799b6027b641eee4815ed9e0a25a7bf9fcf99 Pull Request resolved: #112770

Update on "Implement narrow from a regular tensor to jagged tensor"

ba6c2b3

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 10, 2023

Implement narrow from a regular tensor to jagged tensor

6e462ef

ghstack-source-id: 2fbbc0e0cd2a59eb6c4dde4ba1d6cc83ffb52b42 Pull Request resolved: #112770

Update on "Implement narrow from a regular tensor to jagged tensor"

03a94ee

cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 aakhundov kadeng [ghstack-poisoned]

ani300 added a commit that referenced this pull request Nov 10, 2023

Implement narrow from a regular tensor to jagged tensor

063e90a

ghstack-source-id: df56044c88154d07ff6850b347ac6be215c3ad26 Pull Request resolved: #112770

github-actions bot added the ciflow/inductor label Nov 10, 2023

pytorch-bot bot added the ciflow/trunk Trigger trunk jobs on your pull request label Nov 13, 2023

pytorchmergebot added the merging label Nov 13, 2023

pytorchmergebot added Merged and removed merging labels Nov 13, 2023

pytorchmergebot closed this in 1aece43 Nov 13, 2023

Skylion007 pushed a commit to Skylion007/pytorch that referenced this pull request Nov 14, 2023

Implement narrow from a regular tensor to jagged tensor (pytorch#112770)

2373b1b

Pull Request resolved: pytorch#112770 Approved by: https://github.com/cpuhrsch

jbschlosser mentioned this pull request Nov 14, 2023

[Tracker] Move nested tensors to beta #112398

Open

44 tasks

cpuhrsch reviewed Nov 16, 2023

View reviewed changes

facebook-github-bot deleted the gh/ani300/2/head branch November 17, 2023 15:30

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Implement narrow from a regular tensor to jagged tensor #112770

Implement narrow from a regular tensor to jagged tensor #112770

ani300 commented Nov 2, 2023 •

edited by pytorch-bot bot

pytorch-bot bot commented Nov 2, 2023 •

edited

cpuhrsch Nov 2, 2023

jbschlosser Nov 2, 2023 •

edited

cpuhrsch Nov 3, 2023

soulitzer Nov 3, 2023

ani300 Nov 3, 2023

soulitzer Nov 3, 2023

ani300 Nov 3, 2023

ani300 Nov 3, 2023

ani300 Nov 3, 2023

jbschlosser Nov 6, 2023

soulitzer Nov 3, 2023

soulitzer Nov 3, 2023

ani300 Nov 3, 2023

soulitzer Nov 3, 2023

jbschlosser left a comment •

edited

jbschlosser Nov 3, 2023

ani300 Nov 3, 2023

jbschlosser left a comment

jbschlosser Nov 7, 2023

jbschlosser Nov 7, 2023

jbschlosser Nov 7, 2023

jbschlosser Nov 7, 2023

jbschlosser Nov 7, 2023

jbschlosser Nov 7, 2023

jbschlosser Nov 7, 2023

jbschlosser Nov 7, 2023

jbschlosser commented Nov 7, 2023

ani300 commented Nov 8, 2023

ani300 commented Nov 13, 2023

pytorchmergebot commented Nov 13, 2023

cpuhrsch Nov 16, 2023

ani300 Nov 16, 2023




		def narrow(tensor: Tensor, dim: int, start: Union[SymInt, Tensor], length: Union[SymInt, Tensor], layout=torch.jagged):

	"your start and length must be a Tensor that broadcasts to input.shape[0] x 1"
	"start and length must be Tensors that broadcast to input.shape[0]"

	(maybe view) shows only the elements in the interval `[start, start+length]`. As nested representations
	(maybe view) shows only the elements in the interval `[start, start+length)`. As nested representations

Implement narrow from a regular tensor to jagged tensor #112770

Implement narrow from a regular tensor to jagged tensor #112770

Conversation

ani300 commented Nov 2, 2023 • edited by pytorch-bot bot

pytorch-bot bot commented Nov 2, 2023 • edited

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/112770

✅ No Failures

Choose a reason for hiding this comment

jbschlosser Nov 2, 2023 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jbschlosser left a comment • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jbschlosser left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jbschlosser commented Nov 7, 2023

ani300 commented Nov 8, 2023

ani300 commented Nov 13, 2023

pytorchmergebot commented Nov 13, 2023

Merge started

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ani300 commented Nov 2, 2023 •

edited by pytorch-bot bot

pytorch-bot bot commented Nov 2, 2023 •

edited

jbschlosser Nov 2, 2023 •

edited

jbschlosser left a comment •

edited