Bringing Up DeepSeek-V4-Flash on AMD MI300X

(fergusfinn.com)

62 points by kkm3 hours ago

4 comments

  • maCDzP59 minutes ago
    I train on AMD MI250X and managed to get Gemma 4 31B to work - but it took a lot of work on the software side.
    • kkm57 minutes ago
      This is very interesting, planning to write about it?
  • kkm2 hours ago
    Also the vllm patch accompanying the blogpost: <a href="https:&#x2F;&#x2F;github.com&#x2F;doublewordai&#x2F;vllm-amd-blog-doubleword" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;doublewordai&#x2F;vllm-amd-blog-doubleword</a>
  • mezark2 hours ago
    We at doubleword are bullish for AMD for low-interactivity inference - it does just take a bigger lift on the software side...
  • benlm2 hours ago
    Nice work! Would DeepSeek V4 Pro on 8xMI300X work with these patches?