Skip to content

Refresh ascend link #87

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Sep 30, 2025
+1 −1
Merged

Refresh ascend link #87

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion _posts/2025-09-29-deepseek-v3-2.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -102,7 +102,7 @@ We are also excited to announce out-of-the-box Blackwell support for this model.
We are barely touching the surface of the optimization for DSA and related sparse attention in vLLM. In the coming weeks:

* We plan to expand the architectures supported beyond Hopper and Blackwell.
* We will expand the support to other hardwares such as AMD and TPU. With vLLM's extensible systems, developers can add support for models directly. For example, [vllm-ascend](https://github.com/vllm-project/vllm-ascend/tree/v0.9.1-dev/examples/deepseek.md) and [vllm-mlu](https://github.com/Cambricon/vllm-mlu) already support DeepSeek V3.2!
* We will expand the support to other hardwares such as AMD and TPU. With vLLM's extensible systems, developers can add support for models directly. For example, [vllm-ascend](https://github.com/vllm-project/vllm-ascend/releases/tag/v0.11.0rc0) and [vllm-mlu](https://github.com/Cambricon/vllm-mlu) already support DeepSeek V3.2!
* We continuously test large-scale wide EP serving and disaggregation.
* You will soon be able to run an end-to-end RL loop with this model.
* We will explore the "masked MHA mode for short sequence prefilling" from DeepSeek.
Expand Down