Sandro

2,097 posts

Sandro

@pupposandro

Building @luceboxai GPUs. Local AI. Former Co-founder Cua (YC X25), previously Notion

Joined April 2021

Replies Articles Media

Pinned
Sandro
@pupposandro
Apr 20
Article
How we got 207 tok/s with Qwen3.5-27B on an RTX 3090
Co-author: @davideciffa We built a standalone C++/ggml speculative decoder for Qwen3.5-27B Q4_K_M with a DFlash block-diffusion draft. The demo video below shows a 207.6 tok/s run (5.46x over AR);...
Sandro
@pupposandro
Aug 12
First batch of GPUs for @luceboxai just arrived. Many more coming soon.
Sandro
@pupposandro
Aug 11
Lucebox Engine now serves DeepSeek V4 Flash 0731 from one 98.29 GB GGUF on a 128 GB AMD Strix Halo system. The artifact scores 82/92 on ds4-eval-92 and reaches 32.7 tok/s with DSpark. Across our fixed HumanEval, GSM8K, and MATH evaluation, it averages 27.9 tok/s. Thanks again to
Sandro
@pupposandro
Aug 11
Article
Lucebox × Geometric: DeepSeek V4 Flash 0731 reaches 32.7 tok/s on AMD Strix Halo
We added support for DeepSeek-V4-Flash-0731 to Lucebox Engine together with Geometric. The collaboration covers the adaptive ROCmFPX formats, HIP and CUDA kernels, embedded codebook loading, runtime...
Sandro
@pupposandro
Aug 11
Article
Lucebox × Geometric: DeepSeek V4 Flash 0731 reaches 32.7 tok/s on AMD Strix Halo
We added support for DeepSeek-V4-Flash-0731 to Lucebox Engine together with Geometric. The collaboration covers the adaptive ROCmFPX formats, HIP and CUDA kernels, embedded codebook loading, runtime...
Sandro
@pupposandro
Aug 9
My dad rocking a cool shirt today