Skip to content
View Picom-code's full-sized avatar
24 followers · 11 following

Achievements

Achievement: StarstruckAchievement: QuickdrawAchievement: Pair ExtraordinaireAchievement: Pull SharkAchievement: YOLO

Achievements

Achievement: StarstruckAchievement: QuickdrawAchievement: Pair ExtraordinaireAchievement: Pull SharkAchievement: YOLO

Highlights

  • Pro

Block or report Picom-code

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Picom-code/README.md

Pico Moorhead

local AI infrastructure · your models, your machine

DGX Spark RoCE Context Throughput

pico-dgx-ds4: DeepSeek V4.1 Flash on 4x DGX Spark

pico-dgx-ds4

DeepSeek V4.1 Flash on four DGX Sparks over a 200 GbE RoCE switch, built so many people can share one cluster. A fork of blaesus/dsgx10: faster single-user, 2x+ the multi-user throughput, 1M context, 16 slots, image input.

  1 user ........ 90 chat / 144 code tok/s
  16 users ...... 376 tok/s, a different prompt each
  context ....... 1M
  prefill ....... 10k tok/s

EZLM: your own AI, running on your computer

ezlm

Your own model, running on your own computer. EZLM scans your machine, picks the smartest model it can run well, and hands you a private URL and key. No account, no cloud, no command line.

Download for Windows · Download for Mac · picom-code.github.io/ezlm

Day job — technical marketing engineer at NVIDIA: AI agents, MCP, NeMo Agent Toolkit, Brev.

cuda · python · rust · typescript · pytorch

Pinned Loading

  1. pico-dgx-ds4 pico-dgx-ds4 Public

    DeepSeek V4.1 Flash on 4x DGX Spark: 90/144 tok/s chat/code single-user, 376 tok/s at 16 distinct users, 1M context, images. Fork of blaesus/dsgx10.

    Cuda 112 19

  2. ezlm ezlm Public

    EZLM: your own AI model, running on your own computer. Beta downloads.

    HTML 3