composer VS transformers

Mosaic's open source library is excellent: Composer https://github.com/mosaicml/composer.

* It gives you PyTorch DDP for free. Makes FSDP about as easy as can be, and provides best in class performance monitoring tools. https://docs.mosaicml.com/en/v0.12.1/notes/distributed_train...

Here's a nice intro to using Huggingface models: https://docs.mosaicml.com/en/v0.12.1/examples/finetune_huggi...

I'm just a huge fan of their developer experience. It's up there with Transformers and Datasets as the nicest tools to use.

You may consider using Composer Composer by MosaicML.

Which is why I'm excited to announce that we (MosaicML) just released an automatic way to avoid these errors. Namely, we just added automatic gradient accumulation to Composer, our open source library for faster + easier neural net training.

I would love to know your thoughts on PyTorch Lightning vs. other, even more lightweight libraries, if you have the time. PL strikes me as being less idiosyncratic than FastAI, but I'm still not sure whether it would be better in engineering work to go even more lightweight (when I'm not just writing the code myself) -- something that offers up just optimizations and a trainer, a la MosaicML's [Composer](https://github.com/mosaicml/composer) or Chris Hughes's [pytorch-accelerated](https://github.com/Chris-hughes10/pytorch-accelerated) .

This master's thesis sort of does it, but it doesn't have any fine-tuning yet so it completely wrecks the accuracy: https://github.com/joennlae/halutmatmul.

If someone worked on contributing this to Composer [1] I'd be down to help out. I can't justify building it all on my own right now since we're 100% focused on training speedup, but I could definitely meet and talk through it, help code tricky parts, review PRs, etc.

[1] https://github.com/mosaicml/composer

We're always looking for contributors for Composer. tl;dr it speeds up neural net training by a lot (e.g., 7x faster ResNet-50).

Looking at this: https://github.com/mosaicml/composer

transformers is a package by Hugging Face, that helps you interact with models on HF Hub (GitHub)

Is t5x an encoder/decoder architecture?

Some more general options.

The Flax ecosystem

https://github.com/google/flax?tab=readme-ov-file

or dm-haiku

https://github.com/google-deepmind/dm-haiku

were some of the best developed communities in the Jax AI field

Perhaps the “trax” repo? https://github.com/google/trax

Some HF examples https://github.com/huggingface/transformers/tree/main/exampl...

Sadly it seems much of the work is proprietary these days, but one example could be Grok-1, if you customize the details. https://github.com/xai-org/grok-1/blob/main/run.py

The HuggingFace transformers library already has support for a similar method called prompt lookup decoding that uses the existing context to generate an ngram model: https://github.com/huggingface/transformers/issues/27722

I don't think it would be that hard to switch it out for a pretrained ngram model.

Most of this tutorial is based on Hugging Face course about Transformers and on Niels Rogge's Transformers tutorials: make sure to check their work and give them a star on GitHub, if you please ❤️

* Superconvergence + LR range finder + Fast AI's Ranger21 optimizer was the goto optimizer for CNNs, and worked fabulously well, but on transformers, the learning rate range finder sadi 1e-3 was the best, whilst 1e-5 was better. However, the 1 cycle learning rate stuck. https://github.com/huggingface/transformers/issues/16013

Thanks! :) I'm pushing them into transformers, pytorch-gemma and collabing with the Gemma team to resolve all the issues :)

The RoPE fix should already be in transformers 4.38.2: https://github.com/huggingface/transformers/pull/29285

My main PR for transformers which fixes most of the issues (some still left): https://github.com/huggingface/transformers/pull/29402

If you want to tinker with the architecture Hugging Face has a FOSS implementation in transformers: https://github.com/huggingface/transformers/blob/main/src/tr...

If you want to reproduce the training pipeline, you couldn't do that even if you wanted to because you don't have access to thousands of A100s.