I get that error trying to train ALS (from implicit) (with GPU) with 1024 latent factors. No issues for latent factors < 1024 (up to 992).
Then, if latent factors > 1024, training goes well, but too fast and the evaluation metrics (like precision@k) (from evaluate method of implicit) are too bad (low and not in line with the metrics of latent factors <= 992).
