Bringing Up DeepSeek-V4-Flash on AMD MI300X

 (fergusfinn.com)

33 points | by kkm 2 hours ago

3 comments

  • mezark 36 minutes ago
    We at doubleword are bullish for AMD for low-interactivity inference - it does just take a bigger lift on the software side...
  • kkm 37 minutes ago
    Also the vllm patch accompanying the blogpost: https://github.com/doublewordai/vllm-amd-blog-doubleword
  • benlm 44 minutes ago
    Nice work! Would DeepSeek V4 Pro on 8xMI300X work with these patches?