ZackC
15 posts

ZackC
@zackC_EE
Birder/Photographer | EE @ Waterloo | 18


I implemented @GoogleResearch's TurboQuant as a CUDA-native compression engine on Blackwell B200. 5x KV cache compression on Qwen 2.5-1.5B, near-loseless attention scores, generating live from compressed memory. 5 custom cuTile CUDA kernels ft: - fused attention (with QJL corrections) - online softmax -on-chip cache decompression - pipelined TMA loads Try it out: devtechjr.github.io/turboquant_cut… s/o @blelbach and the cuTile team at @nvidia for lending me Blackwell GPU access :) cc @sundeep @GavinSherry





I implemented @GoogleResearch's TurboQuant as a CUDA-native compression engine on Blackwell B200. 5x KV cache compression on Qwen 2.5-1.5B, near-loseless attention scores, generating live from compressed memory. 5 custom cuTile CUDA kernels ft: - fused attention (with QJL corrections) - online softmax -on-chip cache decompression - pipelined TMA loads Try it out: devtechjr.github.io/turboquant_cut… s/o @blelbach and the cuTile team at @nvidia for lending me Blackwell GPU access :) cc @sundeep @GavinSherry








Yesterday I made the 3d printed stencil generator and many were saying that it wont work for smaller components or on more densely populated boards. I am going to test that today. I believe it could work down to 0201 with a 0.2mm nozzle and hole compensation in the slicer.



