transformers VS gpt-neo

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.

In transformers, they tried really hard to have a single function or method to deal with both self and cross attention mechanisms, masking, positional and relative encodings, interpolation etc. While it allows a user to use the same function/method for any model, it has led to severe parameter bloat. Just compare the original implementation of llama by FAIR with the implementation by HF to get an idea.

pip install git+https://github.com/huggingface/transformers.git@main

By December 2020, EleutherAI had introduced The Pile, a comprehensive text dataset designed for training models. Subsequently, tech giants such as Microsoft, Meta, and Google used this dataset for training their models. In March 2021, they revealed GPT-Neo, an open-source model under Apache 2.0 license, which was unmatched in size at its launch. EleutherAI’s later projects include the release of GPT-J, a 6 billion parameter model, and GPT-NeoX, a 20 billion parameter model, unveiled in February 2022. Their work demonstrates the viability of high-quality open-source AI models.

Yeah, if that is your requirement you should definitely ignore chatterbot, as its older and probably not what your teacher wants. I'm looking at the gpt-neo docs right now: https://github.com/EleutherAI/gpt-neo

3.) EleutherAI's GPT-Neo and GPT-NeoX: EleutherAI is an independent research organization that aims to promote open research in artificial intelligence. They have released GPT-Neo, an open-source language model based on the GPT architecture, and are developing GPT-NeoX, a highly-scalable GPT-like model. You can find more information on their GitHub repositories: GPT-Neo: https://github.com/EleutherAI/gpt-neo GPT-NeoX: https://github.com/EleutherAI/gpt-neox

This is one of the first comprehensive plugins that has been rewritten to support multiple AI backends such as OpenAI GPT3+ and other custom sources in the future such as ChatGPT, GPT-J, GPT-neo and more.

GPT-Neo is newer than GPT-2 on the open source side of things. In my experience, it tends to give longer and more creative responses than GPT-2 but not on the level of GPT-3. I've not tried GPT-J or GPT-NeoX, but they're also open source and reportedly better than GPT-Neo (albeit less accessible).

Finally, yes-- it is trained via masked language modeling (text prediction). The approach has been fairly standard for years- the big difference with the GPT* models is the number of paramaters and volume of text-- we still haven't reached a ceiling with LLM parameters- they appear to keep improving with size. This training allows the model to learn a strong representation of language. Their training approach is published and open-source GPT* versions have already been made and released (https://github.com/EleutherAI/gpt-neo). However, the models are huge and can't be run locally for hobbyists. This gets at larger issues in democratization of ML.

- GPT-Neo 2.7B, released Mar/2021, and unmaintained/unsupported as of Aug/2021? or;