Kirill Kozlov

Hi has anyone managed to find versions of libraries where quantization does not break the model trained on custom dataset?

> Hello, I'm trying to train YOLOv8-large in int4 format. I took the training recipe available at [sparsezoo](https://sparsezoo.neuralmagic.com/models/yolov8-l-coco-pruned85_quantized?hardware=deepsparse-c4.8xlarge&comparison=yolov8-l-coco-base) for training yolov8-large. I modified the num_bits to 4 everywhere. I also...

> UPD: Answer is that it can happen if focal_loss is True, but matched gets that flag as False :) @ArgoHA, what do you mean by that? Is it an...

> Did you check how ultralytics (for example) trains on the same dataset? @ArgoHA, Yes, ultralytics converges faster and the quality is better, I have high resolution images, maybe that...

> I tried 1280x1280 and got better results than ultralytics. Did you check your dataset is converted correctly? I usually save debug images after preprocessing with labels to see what...

> did you find what to do for this? i'm facing a similar issue I nullified all the components of loss except cls loss

> From a discussion in the Ultralytics Discord, a user was having issues with a model exported after `sparseml` training. I cooked up a quick process to add the metadata...

@Burhan-Q thx for your response, I have tried all available versions of Ultralytics sparseml DeepSparse, the problem seems to lie in the sparseml config itself, there is a problem with...

> You can try training with `overlap_masks=False` Thx for your advice, I ve tried, but got an error during training`TypeError: ERROR ❌ segment dataset incorrectly formatted or not a segment...

> What's the full error?