Securelayer7/Qwythos-9B-Claude-Mythos-Uncensored-Abliterated-1M
LoRA-style (fewer layers) vs empero-ai/Qwythos-9B-Claude-Mythos-5-1M. Securelayer7/Qwythos-9B-Claude-Mythos-Uncensored-Abliterated-1M matches part of empero-ai/Qwythos-9B-Claude-Mythos-5-1M (427 of its 760 tensors; the rest of empero-ai/Qwythos-9B-Claude-Mythos-5-1M has no counterpart). On the shared part: Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, in_proj_qkv, o_proj, out_proj, q_proj, v_proj changed, in 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/105, Attention 72/224, MLP 32/96