[PR #327] [CLOSED] [Fix] Fix the bug causing infinite loop #1403

New issue

Closed

opened 2026-10-29 14:52:08 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-29 14:52:08 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/kvcache-ai/ktransformers/pull/327
Author: @jts250
Created: 2/15/2025
Status: ❌ Closed

Base: main ← Head: main

📝 Commits (5)

e2f3b49 Resolve the issue raised in #319.
0440cea refactor(kernel): simplify VALID_BLOCK_H and mask_h logic
dc9a8c6 refactor(kernel): simplify VALID_BLOCK_H and mask_h logic
66c2531 refactor(kernel): simplify VALID_BLOCK_H and mask_h logic
155a3c6 refactor(kernel): simplify VALID_BLOCK_H and mask_h logic

📊 Changes

1 file changed (+1 additions, -4 deletions)

View changed files

📝 ktransformers/operators/triton_attention.py (+1 -4)

📄 Description

Replaced mask_h = mask_h & (cur_head < q_head_num) with mask_h = mask_h | (cur_head < q_head_num).

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/kvcache-ai/ktransformers/pull/327 **Author:** [@jts250](https://github.com/jts250) **Created:** 2/15/2025 **Status:** ❌ Closed **Base:** `main` ← **Head:** `main` --- ### 📝 Commits (5) - [`e2f3b49`](https://github.com/kvcache-ai/ktransformers/commit/e2f3b49d902be2118a0bc592f73abcf70e323b85) Resolve the issue raised in #319. - [`0440cea`](https://github.com/kvcache-ai/ktransformers/commit/0440ceac2ada9e91b381ca12efb142e4e5f3b2fa) refactor(kernel): simplify VALID_BLOCK_H and mask_h logic - [`dc9a8c6`](https://github.com/kvcache-ai/ktransformers/commit/dc9a8c6b700d98d102790cf37726893d79a56748) refactor(kernel): simplify VALID_BLOCK_H and mask_h logic - [`66c2531`](https://github.com/kvcache-ai/ktransformers/commit/66c253175262e52af9710a57a5e856254ca1eb76) refactor(kernel): simplify VALID_BLOCK_H and mask_h logic - [`155a3c6`](https://github.com/kvcache-ai/ktransformers/commit/155a3c6499b7b4e888f60b129f1a42a859f97f05) refactor(kernel): simplify VALID_BLOCK_H and mask_h logic ### 📊 Changes **1 file changed** (+1 additions, -4 deletions) <details> <summary>View changed files</summary> 📝 `ktransformers/operators/triton_attention.py` (+1 -4) </details> ### 📄 Description Replaced mask_h = mask_h & (cur_head < q_head_num) with mask_h = mask_h | (cur_head < q_head_num). --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-29 14:52:08 +00:00