Add cdist op #9391
Open
Add cdist op #9391
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
marigoold
requested review from
doombeaker,
jackalcooper,
hjchen2,
BBuf and
daquexian
as code owners
|
Code got formatted by CI. Please request CI again if you still want to have this PR merged. If the PR is from a forked repo, please download the patch files from the GitHub Actions web page and apply them locally. |
marigoold
requested review from
oneflow-ci-bot
and removed request for
oneflow-ci-bot
Speed stats: |
|
View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/9391/ |
Speed stats: |
This reverts commit 011712e.
Speed stats: |
Speed stats: |
Speed stats: |
|
View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/9391/ |
Speed stats: |
|
View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/9391/ |
|
Code got formatted by CI. Please request CI again if you still want to have this PR merged. If the PR is from a forked repo, please download the patch files from the GitHub Actions web page and apply them locally. |
Speed stats: |
Speed stats: |
Comment on lines
+3468
to
+3474
| // mm_for_euclid_dist has accuracy issue | ||
| // if (p == 2 && (mode == 1 || (mode == 0 && (r1 > 25 || r2 > 25)))) { | ||
| // shape output_shape(max_batch_shape); | ||
| // output_shape.emplace_back(r1); | ||
| // output_shape.emplace_back(r2); | ||
| // return JUST(Reshape(JUST(euclidean_dist(x1_expand, x2_expand)), output_shape)); | ||
| // } |
There was a problem hiding this comment.
删除无用的注释
这里的代码在 torch 里面是有的，只是当前还有精度问题，解决掉之后就解除注释了
Comment on lines
+3418
to
+3429
| Maybe<Tensor> euclidean_dist(const std::shared_ptr<Tensor>& x1, | ||
| const std::shared_ptr<Tensor>& x2) const { | ||
| const auto& x1_norm = JUST(ReduceSum(JUST(ScalarPow(x1, 2, false)), {-1}, true)); | ||
| const auto& x2_norm = JUST(ReduceSum(JUST(ScalarPow(x2, 2, false)), {-1}, true)); | ||
| const auto& x1_ones = JUST(OnesLike(x1_norm)); | ||
| const auto& x2_ones = JUST(OnesLike(x2_norm)); | ||
| const auto& x1_cat = JUST(Concat({JUST(ScalarMul(x1, -2, false)), x1_norm, x1_ones}, -1)); | ||
| const auto& x2_cat = JUST(Concat({x2, x2_ones, x2_norm}, -1)); | ||
| const auto& result = | ||
| JUST(MatMul(x1_cat, JUST(Transpose2dim(x2_cat, -1, -2)), false, false, 1.0)); | ||
| return Sqrt(JUST(ClampMin(result, 0.0))); | ||
| }; |
There was a problem hiding this comment.
这个函数的调用在下面注释了，没有用到了
| x2 = x2.to_global(placement=placement, sbp=sbp) | ||
| z = torch.cdist(x1, x2) | ||
| return z | ||
|
|
There was a problem hiding this comment.
我看这里测试都是相同dim的，代码里写了broadcast的逻辑，也测试一下broadcast的情况吧
youxiudeshouyeren
approved these changes
Apr 24, 2023
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.
cdist 对于两个输入 x1 (shape=[B, R1, C])，x2 (shape=[B, R2, C])，计算每个 batch 内 x1 和 x2 每一行向量之间距离的p范数，得到结果 result (shape=[B, R1, R2])。
torch 文档见 https://pytorch.org/docs/stable/generated/torch.cdist.html