New measurements report 5.02, 5.07 and 5.27 bits per weight for same model under Q4_K_M label

A recent analysis examines a model identified by the Q4_K_M label. The study reports three distinct bits‑per‑weight measurements: 5.02, 5.07 and 5.27. All measurements pertain to the same underlying model architecture. The variation suggests differences in quantization or training settings. The findings are hosted on a GitHub repository. The repository is named picchio and is maintained by logxio. The data may inform developers optimizing model size and performance. Further investigation could clarify the source of the discrepancies.