Habana Labs Opens Up The Code To Their AI Compiler, SynapseAI Core

[phoronix]

Phoronix: Habana Labs Opens Up The Code To Their AI Compiler, SynapseAI Core

Intel-owned Habana Labs now has the most open software stack among AI accelerators! While Habana Labs has long provided an open-source, upstream kernel driver for their Gaudi AI training and Goya AI inference accelerators, the user-space portions including their code compiler and run-time library have been closed-source. This has been a thorn for upstream kernel developers and their standards, but now Habana Labs has open-sourced their user-space components too...

https://www.phoronix.com/scan.php?page=news_item&px=Hababa-Labs-Open-Compiler-AI

[uid313]

Good that the kernel developers put pressure on these guys! 👍

Great to see it open source now! 👍

Open source! ❤

[StillStuckOnSI]

Given how many caveats and unimplemented features are listed in the readme, I hope this was just a rush job to get the ball rolling with the kernel (i.e. it will be fleshed out appropriately later) and not some kind of malicious compliance!

[Developer12]

Given how many caveats and unimplemented features are listed in the readme, I hope this was just a rush job to get the ball rolling with the kernel (i.e. it will be fleshed out appropriately later) and not some kind of malicious compliance!

Even if it is, in that case it's perfectly possible for someone to fork it and start adding features. There's hopefully enough here for people to see how things connect and figure out how they could independently implement the missing functionality. You could see a similar situation develop to that with AMD's GPUs.