Andrea Giammarchi 🍥

@WebReflection

Web, Mobile, IoT: all JS things since 00's - opinions are my own ™ Principal SW Engineer @anacondainc working on OSS stack 🦄

Milan, Italy

medium.com/@WebReflection

Replies Reposts Media

Pinned
Andrea Giammarchi 🍥
@WebReflection
Jul 11, 2023
coincident/server on IoT via Raspberry Pi Zero 2 W? Sure! github.com/WebReflection/…
00:00
5
Andrea Giammarchi 🍥
@WebReflection
2h
it took just about forever but it finally works: github.com/MiaAI-Lab/Deep… from @MiaAI_lab is running on my desk 🥳 HOW * Nvidia Sync for the cluster - sparkrun failed * sparkrun setup ssh after for any passwordless ssh * worker IP CX7 head -> workers * NFS IP CX7 workers -> head
GitHub - MiaAI-Lab/DeepSeek-v4.1-Flash-DGX-Sparks: DeepSeek-V4.1-Flash on 3-4x NVIDIA DGX Sparks
From github.com
1
Andrea Giammarchi 🍥
@WebReflection
Sep 25
I honestly don't understand why most software ship with dozen+ boolean flags when bitwise operations would carry a single summary of the whole state and make everything easier/faster to reason about. if you mention "only 31 options to deal with" I remind you BigInt works too 🤷
5
Andrea Giammarchi 🍥
@WebReflection
Sep 23
this is how I've solved possible ID saturation on most of my projects: don't just id++ if you want to be sure that such id is unique, use Int32Array or Uint32Array instead to let the ID cap and reset without issues, granting safety against MAX_SAFE_INTEGER
Change fetchSeq to a Uint32Array for better handling by WebReflection · Pull Request #4 · tarwin/...
From github.com
2
Andrea Giammarchi 🍥
@WebReflection
Sep 21
The @NVIDIAAI Sync is a decent tool but if you launch it from a DGX Spark itself it doesn’t configure it as available device for a cluster out of the box so if you have 3 you end up by default configuring the cluster for the other 2 but not 3 🤦
1