serve VS transformers

What's wrong with just using Torchserve[1]? We've been using it to serve embedding models in production.

[1] https://pytorch.org/serve/

Looks like you are already up to speed loading LLaMa models which is great. Assuming this is Hugging Face PyTorch checkpoint, I think it should be possible to spin up a TorchServe instance which has in-built support for API access and HF Transformers. Since scale and latency aren’t a big concern for you, this should be good enough start.

You could split the service in two, flask for the web part and a service to serve the model, I haven't used it but there is https://pytorch.org/serve/

I'm attempting to set up inference for a torchserve container, and it's really tough to figure out what's not allowing me to connect to my network with the ports that I'm trying to expose. I'm using Google Kubernetes Engine and Helm via tweaking one of the tutorials at [torchserve](github.com/pytorch/serve). Specifically, it's the GKE tutorial [here](https://github.com/pytorch/serve/tree/master/kubernetes).

I did a Space to showcase a bit the speedups we can have in a end-to-end case with TorchServe to deploy the model on a cloud instance (AWS EC2 g4dn, using one T4 GPU): https://huggingface.co/spaces/fxmarty/bettertransformer-demo

For 2), I am aware of a few options. Triton inference server is an obvious one as is the ‘transformer-deploy’ version from LDS. My only reservation here is that they require the model compilation or are architecture specific. I am aware of others like Bento, Ray serving and TorchServe. Ideally I would have something that allows any (PyTorch model) to be used without the extra compilation effort (or at least optionally) and has some convenience things like ease of use, easy to deploy, easy to host multiple models and can perform some dynamic batching. Anyway, I am really interested to hear people's experience here as I know there are now quite a few options! Any help is appreciated! Disclaimer - I have no affiliation or are connected in any way with the libraries or companies listed here. These are just the ones I know of. Thanks in advance.

If you built the model using pytorch or tensorflow, I'd suggest using torchserve or TF serving to serve the model as its own "microservice," then query it from your django app. Among other things, it will make retraining and updating your model a lot easier.

GitHub

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