mryab

Joined 6 years ago

Karma

Submissions

Member since

May 30

Recent submissions

There is! See https://petals.ml/ for inference of models like BLOOM-176B over the internet or https://arxiv.org/abs/2301.11913 and https://arxiv.org/abs/2206.01288 that show you how to do pretraining from scratch in the same setting. Disclaimer: I'm a coauthor of these systems (including the one in OP)

Comment · 3 years ago

There absolutely are! Check out hivemind (https://github.com/learning-at-home/hivemind), a general library for deep learning over the Internet, or Petals (https://petals.ml/), a system that leverages Hivemind and allows you to run BLOOM-176B (or other large language models) that is distributed over many volunteer PCs. You can join it and host some layers of the model by running literally one command on a Linux machine with Docker and a recent enough GPU.

Disclaimer: I work on these projects, both are based on our research over the past three years

Comment · 5 years ago

For now, the only separation we have is that each worker is responsible for its own weights, since network security has not been our top priority. Still, we've been thinking about adding some security measures like proof-of-work for each node and detection of anomalous inputs/gradients (or simply NaN values). Right now we're running experiments on internal hardware, but before a public launch we'll make sure that malicious participants won't put everybody else's work to waste :)

Comment · 5 years ago

Thank you! This is definitely something we should look into in the future (hopefully with community help); as of now, training infrastructure and model convergence are the highest priorities. That said, we welcome all ideas of ways to motivate more volunteers to join the experiments, because Learning@home team comes from a distributed DL background with limited volunteer computing expertise.

Also, I believe that for some projects (e.g. GPT-3 replication effort) people would want to join the network regardless of the incentive mechanism, as demonstrated by Leela Chess Zero [1].

[1] http://lczero.org/

Comment · 5 years ago

Sure! It's @m_ryabinin

Comment · 5 years ago

This way, you never have to synchronize the weights of the entire model across the participants — you only need to send the gradients/activations to a set of peers. Slow connections are mitigated with asynchronous SGD and unreliable/disconnected experts can be discarded, which makes it more suitable for Internet-like networks.

Disclaimer: I work on this project. We're currently implementing a prototype, but it's not yet GPT-3 sized. Some issues like LR scheduling (crucial for Transformer convergence) and shared parameter averaging (for gating etc.) are tricky to implement for decentralized training over the Internet.

[1] https://learning-at-home.github.io/

Comment · 6 years ago

Here is a recent paper (disclaimer: I am the first author) named "Learning@home" which proposes something along these lines. Basically, we develop a system that allows you to train a network with thousands of "experts" distributed across hundreds or more of consumer-grade PCs. You don't have to fit 700GB of parameters on a single machine and there is significantly less network delay as for synchronous model parallel training. The only thing you sacrifice is the guarantee that all the batches will be processed by all required experts.

You can read it on ArXiv https://arxiv.org/abs/2002.04013v1 or browse the code here: https://github.com/learning-at-home/hivemind. It's not ready for widespread use yet, but the core functionality is stable and you can see what features we are working on now.