ColossalAI VS transformers

> open-source a complete RLHF pipeline ... based on the LLaMA pre-trained model

I've gotten to where when I see "open source AI" I now know it's "well, except for $some_other_dependencies"

Anyway: https://scribe.rip/@yangyou_berkeley/colossalchat-an-open-so... and https://github.com/hpcaitech/ColossalAI#readme (Apache 2) can save you some medium.com heartache at least

Quick Read: https://www.marktechpost.com/2023/04/01/meet-colossalchat-an-open-source-ai-solution-for-cloning-chatgpt-with-a-complete-rlhf-pipeline/ Github: https://github.com/hpcaitech/ColossalAI Examples: https://chat.colossalai.org/

One of the current methods for training competing models is to have ChatGPT literally create prompt -> completion data sets. That's what was used for https://github.com/hpcaitech/ColossalAI. A model based off of the Llama weights released by facebook, then fine tuned on ChatGPT3.5 prompt + completions. So yes, there is a good chance that google is literally using ChatGPT in the training loop.

Here's the github from the article:

https://github.com/hpcaitech/ColossalAI

The article talks about their RLHF implementation briefly. There’s details on their RLHF implementation here: https://github.com/hpcaitech/ColossalAI/blob/a619a190df71ea3...

Here’s the project on GitHub: https://github.com/hpcaitech/ColossalAI

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.

I am aware that using mixed precision reduces the stability of weight and there will be little consistency but don't expect it to be this much. I have attached the graph of evaluation metrics. If someone can give me some insight into this issue, that would be great.