[GH-ISSUE #887] 能否使用A100推理Deepseek R1 fp8版本? #640

Open
opened 2026-10-29 14:42:17 +00:00 by bsakaguchi · 2 comments
Owner

Originally created by @huliangbing on GitHub (Mar 14, 2025).
Original GitHub issue: https://github.com/kvcache-ai/ktransformers/issues/887

Ktransformers很棒!我们已经部署成功!感谢!

清华大学推出了“FP8模型不再挑卡!DeepSeek推理成本减半速度翻番,清华团队开源「赤兔」推理引擎”https://github.com/thu-pacman/chitu,能否使用A100推理DS R1的fp8版,谢谢!

Originally created by @huliangbing on GitHub (Mar 14, 2025). Original GitHub issue: https://github.com/kvcache-ai/ktransformers/issues/887 Ktransformers很棒!我们已经部署成功!感谢! 清华大学推出了“FP8模型不再挑卡!DeepSeek推理成本减半速度翻番,清华团队开源「赤兔」推理引擎”https://github.com/thu-pacman/chitu,能否使用A100推理DS R1的fp8版,谢谢!
Author
Owner

@Atream commented on GitHub (Mar 15, 2025):

赤兔引擎的关键在于其fp8在线转换bf16并做乘法的triton kernel,我们会尽快将其适配过来。

<!-- gh-comment-id:2726445190 --> @Atream commented on GitHub (Mar 15, 2025): 赤兔引擎的关键在于其fp8在线转换bf16并做乘法的triton kernel,我们会尽快将其适配过来。
Author
Owner

@huliangbing commented on GitHub (Mar 16, 2025):

期待!👍👍👍

<!-- gh-comment-id:2727182070 --> @huliangbing commented on GitHub (Mar 16, 2025): 期待!👍👍👍
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
kvcache-ai/ktransformers#640
No description provided.