Body tensors are byte-identical to the previous upload — strip and graft both copy tensor bytes verbatim. Only blk.40 changes, from Q8_0 (0.90 GB) to the same Q3_K-expert head the CyberTiel ladder ships (0.39 GB), plus the chat template moving to qwen3.8-froggeric-v22.5.0. Head precision was swept Q8_0 down to Q2_K with draft acceptance flat to Q3_K.
