[PR #529] [MERGED] feat: add Blackwell-optimized vLLM infra for DeepSeek-R1 #582

New issue

Closed

opened 2026-10-26 09:59:15 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 09:59:15 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/awesome-deepseek-integration/pull/529
Author: @informatico-madrid
Created: 1/17/2026
Status: ✅ Merged
Merged: 2/23/2026
Merged by: @mowentian

Base: main ← Head: patch-1

📝 Commits (1)

037f21f Add Blackwell Linux Infra Optimizer to README

📊 Changes

1 file changed (+5 additions, -0 deletions)

View changed files

📝 README.md (+5 -0)

📄 Description

🚀 Description

This PR adds the Blackwell Linux Infra Optimizer to the Others section. It is a specialized infrastructure stack designed to run DeepSeek-R1 on the latest NVIDIA Blackwell (SM_120) architecture.

🛠️ Technical Highlights:

Kernel 6.14 Ready: Successfully resolves the current flash-attn symbol conflicts on new Linux kernels by utilizing a native FlashInfer backend within vLLM.
Hardware Native: Specifically optimized and compiled for SM_120 (RTX 5090 series).
Benchmark: Achieved a sustained throughput of 59.0 tokens/s on DeepSeek-R1-32B (AWQ) using a Dual 5090 setup.
Open Source: Provides a production-ready Docker environment for local-first deployment.

✅ Checklist

I have verified the link is working.
The project is non-commercial and open-source.
The description is concise and follows the repository's format.

_{🔄 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/deepseek-ai/awesome-deepseek-integration/pull/529 **Author:** [@informatico-madrid](https://github.com/informatico-madrid) **Created:** 1/17/2026 **Status:** ✅ Merged **Merged:** 2/23/2026 **Merged by:** [@mowentian](https://github.com/mowentian) **Base:** `main` ← **Head:** `patch-1` --- ### 📝 Commits (1) - [`037f21f`](https://github.com/deepseek-ai/awesome-deepseek-integration/commit/037f21f478f49de0154d25fd1bad01410eb39467) Add Blackwell Linux Infra Optimizer to README ### 📊 Changes **1 file changed** (+5 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `README.md` (+5 -0) </details> ### 📄 Description ### 🚀 Description This PR adds the **Blackwell Linux Infra Optimizer** to the `Others` section. It is a specialized infrastructure stack designed to run DeepSeek-R1 on the latest NVIDIA Blackwell (SM_120) architecture. ### 🛠️ Technical Highlights: - **Kernel 6.14 Ready**: Successfully resolves the current `flash-attn` symbol conflicts on new Linux kernels by utilizing a native **FlashInfer** backend within vLLM. - **Hardware Native**: Specifically optimized and compiled for **SM_120 (RTX 5090 series)**. - **Benchmark**: Achieved a sustained throughput of **59.0 tokens/s** on DeepSeek-R1-32B (AWQ) using a Dual 5090 setup. - **Open Source**: Provides a production-ready Docker environment for local-first deployment. ### ✅ Checklist - [x] I have verified the link is working. - [x] The project is non-commercial and open-source. - [x] The description is concise and follows the repository's format. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-26 09:59:15 +00:00