wangzhang/gpt-oss-20b-abliterated
LoRA-style vs openai/gpt-oss-20b. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of k_proj, o_proj, q_proj, v_proj changed, in 16 of 24 layers; norms, biases and embeddings are untouched.
Tensors changed: Norms 0/49, MLP 0/96, Attention 47/216, lm_head 0/1, Embeddings 0/1
4 derivatives checked
- mradermacher/gpt-oss-20b-abliterated-i1-GGUF: Quantization
- mradermacher/gpt-oss-20b-abliterated-GGUF: Quantization
- wangzhang/gpt-oss-20b-abliterated-GGUF: Same weights, converted
- 3MPER0RR/gpt-oss-20b-3MPER0RR-abliterated: Abliterated
1 other repos hold byte-for-byte identical weight files.