fastT5 VS FasterTransformer

I've tried https://github.com/Ki6an/fastT5 but it works with CPU only.

I am working on a project where I fine-tuned a Pegasus model on the Reddit dataset. Now, I need to convert the fine-tuned model to ONNX for the deployment stage. I have followed this guide from Huggingface to convert to the ONNX model for unsupported architects. I got it done but the ONNX model can't generate text. Turned out that Pegasus is an encoder-decoder model and most guides are for either encoder-model (e.g. BERT) or decoder-model (e.g. GPT2). I found the only example of converting an encoder-decoder model to ONNX from here https://github.com/Ki6an/fastT5.

Microsoft Onnx Runtime T5 export tool / FastT5: to support caching, it exports 2 times the decoder part, one with cache, and one without (for the first generated token). So the memory footprint is doubled, which makes the solution difficult to use for these large transformer models.

Here is the fastT5 model code for reference code:https://github.com/Ki6an/fastT5/blob/master/fastT5/onnx_models.py

for more information on the project refer to the repository here.

https://docs.nvidia.com/ai-enterprise/overview/0.1.0/platfor...

RIVA: NVIDIA® Riva, a premium edition of NVIDIA AI Enterprise software, is a GPU-accelerated speech and translation AI SDK

FasterTransformer: https://github.com/NVIDIA/FasterTransformer an

There could be some spectrum of this expertise. For instance, https://github.com/NVIDIA/FasterTransformer, https://github.com/microsoft/DeepSpeed

Have anyone tried to optimize the forward and backward using custom Cuda code or fused kernel to speed up the training time of current LLMs? I only have seen FasterTransformer ( NVIDIA/FasterTransformer) and other similar tools but they're only focusing on inference.

Replit built Ghostwriter on the open source scene based on Salesforce’s Codegen, using Nvidia’s FasterTransformer and Triton server for highly optimized decoders, and the knowledge distillation process of the CodeGen model from two billion parameters to a faster model of one billion parameters.

Nvidia FasterTransformer is a mix of Pytorch and CUDA/C++ dedicated code. The performance boost is huge on T5, they report a 10X speedup like TensorRT. However, the speedup is computed on a translation task where sequences are 25 tokens long on average. In our experience, improvement on very short sequences tend to decrease by large margins on longer ones. Still we plan to dig deeper into this project as it implements very interesting ideas.

On the other side of the spectrum, there is Nvidia demos (here or there) showing us how to build manually a full Transformer graph (operator by operator) in TensorRT to get best performance from their hardware. It’s out of reach for many NLP practitioners and it’s time consuming to debug/maintain/adapt to a slightly different architecture (I tried). Plus, there is a secret: the very optimized model only works for specific sequence lengths and batch sizes. Truth is that, so far (and it will improve soon), it’s mainly for MLPerf benchmark (the one used to compare DL hardware), marketing content, and very specialized engineers.