[2506.13771] LittleBit: Ultra Low-Bit Quantization via Latent Factorization

Interesting to see improvements and research into quantization aware training (QAT) that can make some really tiny models.