yolov5 VS transformers

Ref https://www.youtube.com/watch?v=0GwnxFNfZhM https://github.com/ultralytics/yolov5 https://dev.to/gfstealer666/kaaraich-yolo-alkrithuemainkaartrwcchcchabwatthu-object-detection-3lef https://www.kaggle.com/datasets/devdgohil/the-oxfordiiit-pet-dataset/data

!git clone https://github.com/ultralytics/yolov5.git ## Navigate to the model %cd yolov5/ ## Install requirements !pip install -r requirements.txt ## Download the YOLOv5 model !wget https://github.com/ultralytics/yolov5/releases/download/v6.0/yolov5s.pt

Specifically what I did was take the coco128.yaml, added 6 new classes from Dataset A (which have already been converted to YOLO Darknet TXT), from index 0-5 and subsequently adjusted the indices of the other COCO classes. The I proceeded to train and validate on Dataset A for 20 epochs.

I am using the default YOLOv5m6 model here with sahi/yolov5 library for my object detection project. I want to change just some of labels - for example when YOLO detects a human, I want it to label the human as "threat", not "person". Is there any way I can do it just changing some code, or I should train the model from scratch by just changing labels?

You should add them without annotations. Go through this.

You are correct and if you check the firmware, it's yet another famous 3rd party project without attribution, namely https://github.com/ultralytics/yolov5

On one side, being a Linux user for 24 years now, it annoys me that they rip off code and claiming it as theirs again, thus violating licenses, but on the other thanks to k3d's exploit I'm able to tinker more with the machine and if needed do (selective) updates by hand then with a closed source system. It's not just "klipper", with klipper, fluidd and moonraker, it's also ffmpeg and mjpegstreamer. It's gonna be interesting since they also use a project that isn't just GPL, but APGL (in short "If your software gives service online, you have to publish the source code of it and any library that it borrows functions from.") - they use yolov5 (for AI).

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.