JACK-POTTSSON/Qwen3.8-27B-OBLITERATEDpintus
Partial fine-tune (fewer layers) vs Qwen/Qwen3.8-27B. Qwen/Qwen3.8-27B matches part of JACK-POTTSSON/Qwen3.8-27B-OBLITERATEDpintus (1199 of its 2049 tensors; the rest of JACK-POTTSSON/Qwen3.8-27B-OBLITERATEDpintus has no counterpart). On the shared part: Partial change: 5% of the weights changed (lm_head), in 0 of 64 layers.
Tensors changed: Norms 0/325, Attention 0/560, MLP 0/307, Embeddings 0/5, lm_head 1/1, other 0/1
2 other repos hold byte-for-byte identical weight files.