Skip to content

bugfix: fix precision issue of TileLang operator chunk_gated_delta_rule_fwd_h.#1510

Merged
yingxudeng merged 1 commit into
jd-opensource:preview/qwen3.5-qwen3.6from
fengz72:preview/qwen3.5-qwen3.6
May 22, 2026
Merged

bugfix: fix precision issue of TileLang operator chunk_gated_delta_rule_fwd_h.#1510
yingxudeng merged 1 commit into
jd-opensource:preview/qwen3.5-qwen3.6from
fengz72:preview/qwen3.5-qwen3.6

Conversation

@fengz72
Copy link
Copy Markdown

@fengz72 fengz72 commented May 21, 2026

fix precision issue of TileLang operator chunk_gated_delta_rule_fwd_h

Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

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

Code Review

This pull request modifies the chunk_gated_delta_rule_fwd_h kernel for Ascend targets by adding a T.barrier_all() synchronization call before entering the computation scope. This change ensures proper synchronization across threads before the gated delta rule forward pass logic is executed. There are no review comments to address, and I have no further feedback to provide.

@yingxudeng yingxudeng changed the title bugfix: fix precision issue of TileLang operator chunk_gated_delta_ru… bugfix: fix precision issue of TileLang operator chunk_gated_delta_rule_fwd_h. May 22, 2026
@yingxudeng yingxudeng merged commit 82a407d into jd-opensource:preview/qwen3.5-qwen3.6 May 22, 2026
10 checks passed
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