News for hackers. Stuff that matters.

jafermarq

jafermarq on Hacker News

Re:Flower 1.15.0 on Monday February 03, 2025 @11:53AM (#42920197)
Do you mean this example? https://github.com/adap/flower/tree/main/examples/quickstart...
Re:Flower 1.15.0 on Monday February 03, 2025 @11:29AM (#42919915)
Amazing to see the new docs page about Flower Network Communication: https://flower.ai/docs/framework/ref-flower-network-communic...

Show HN: Federation of robots collaboratively train an object manipulation model

Posted by jafermarq on Wednesday January 15, 2025 @01:04PM
from the show-hn dept.

https://github.com/adap/flower/tree/main/examples/quickstart-lerobot

This is a quickstart example using LeRobot and Flower that demonstrates how to train a diffusion model collaboratively across 10 individual nodes (each with its own dataset). This example uses the push-t dataset, where the task is to move a letter T object on top of another that is to remain static.The example it's pretty easy to run, and can do so efficiently if you have access to a recent...

( Read More... | no comments | 7 points | on HN )
Re:Raspberry Pi 5: Ready for Federated Vision on Thursday December 07, 2023 @11:32AM (#38558383)
In a world starving for GPUs, we need to start looking for alternatives. Luckily embedded processors keep getting both faster and cheaper -- making them a increasingly attractive for ML.

In this blog, we benchmark the Raspberry Pi 5 for on-device training on 50 vision models and compare it against its predecessor. Can the Raspberry Pi 5 spark a revolution where drones and small robots collaboratively train the foundational vision models of the future?

Re:Federated Finetuning of OpenAI's Whisper on Raspberry Pi 5 on Thursday November 16, 2023 @06:02PM (#38296901)
If the task requires labels yes. Alternatively, an auxiliary model (not part of the training) can generate pseudo labels and use those during training. A more general approach (for which a few works in FL have proven to work pretty well) would do instead un/semi-supervised training.
Re:Federated Finetuning of OpenAI's Whisper on Raspberry Pi 5 on Thursday November 16, 2023 @03:35PM (#38295072)
Yes. The example uses Whisper-tiny which is 39M, a perfect match for the downstream task of keyword spotting. Just one line needs to be changed in the code to run a larger Whisper model :)
Re:Federated Finetuning of OpenAI's Whisper on Raspberry Pi 5 on Thursday November 16, 2023 @03:31PM (#38295037)
yes. Finetuning a whisper model on a RPi 5 is ~2x faster than on the RPi 4. Other stages involving data pre-processing with HF dataset is again 2x-3x faster.
Re:Federated Finetuning of OpenAI's Whisper on Raspberry Pi 5 on Thursday November 16, 2023 @03:18PM (#38294883)
yeah. with FL it should be possible to make sense out of all data that is distributed across devices without ever having to move it to a central location (i.e. collect it). In the case of speech data, users participating in a federated setting would likely come from different backgrounds, which could be reflected in their accent or use of language.