Skip to content

feat: add cambricon causal softmax op#35

Open
bitzyz wants to merge 1 commit intomasterfrom
feat/dev-causal-softmax-cambricon
Open

feat: add cambricon causal softmax op#35
bitzyz wants to merge 1 commit intomasterfrom
feat/dev-causal-softmax-cambricon

Conversation

@bitzyz
Copy link
Copy Markdown
Contributor

@bitzyz bitzyz commented Mar 27, 2026

  • Add Cambricon causal softmax operator
  • Align the test accuracy with infinicore

@bitzyz bitzyz self-assigned this Mar 27, 2026
@bitzyz bitzyz marked this pull request as draft March 27, 2026 03:23
@bitzyz bitzyz force-pushed the feat/dev-causal-softmax-cambricon branch 2 times, most recently from dbca812 to 7de4dcf Compare March 27, 2026 05:55
@bitzyz
Copy link
Copy Markdown
Contributor Author

bitzyz commented Mar 27, 2026

image

@bitzyz bitzyz force-pushed the feat/dev-causal-softmax-cambricon branch from 7de4dcf to d6bf09c Compare March 27, 2026 06:23
@bitzyz bitzyz marked this pull request as ready for review March 30, 2026 02:19
@bitzyz bitzyz requested review from Ziminli and voltjia March 30, 2026 06:40
@bitzyz bitzyz force-pushed the feat/dev-causal-softmax-cambricon branch from d6bf09c to c25096a Compare April 8, 2026 03:26
@bitzyz bitzyz changed the base branch from feat/dev-infra to master April 8, 2026 03:26
@bitzyz
Copy link
Copy Markdown
Contributor Author

bitzyz commented Apr 9, 2026

image

@voltjia
Copy link
Copy Markdown
Collaborator

voltjia commented Apr 23, 2026

本 PR 对 tolerances 的修改是根据 https://github.com/InfiniTensor/InfiniCore/blob/main/test/infiniop/causal_softmax.py 进行的。

Comment thread src/cambricon/common.h
Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

检查函数参数顺序,可以参考 CONTRIBUTING.md,其中一条是需要输入在前,输出在后。

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

检查参数顺序是否符合 CONTRIBUTING.md 要求。


~Operator() {}

void *default_workspace_{nullptr};
Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这里需要满足 CONTRIBUTING.md 要求,相邻两行之间空行。

cnrt_utils::GetLaunchConfig(input.device(), &core_per_cluster,
&cluster_count);
}
void operator()(const Tensor input, Tensor out) const override {
Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这里也需要满足要求空行。

@voltjia voltjia self-requested a review April 23, 2026 07:58
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants