Delta-Vector/MS3.2-Austral-Winton
LoRA-style vs Delta-Vector/MS3.2-Austral-24B-KTO. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj changed, in 40 of 40 layers; norms, biases and embeddings are untouched.
Tensors changed: lm_head 0/1, Embeddings 0/1, Norms 0/81, MLP 120/120, Attention 160/160
5 derivatives checked
- mradermacher/MS3.2-Austral-Winton-i1-GGUF: Quantization
- bartowski/Delta-Vector_MS3.2-Austral-Winton-GGUF: Same weights, converted
- mradermacher/MS3.2-Austral-Winton-GGUF: Quantization
- OddTheGreat/NeutralGear_24B_V.2: Full fine-tune
- NewEden/Austral-24b-GRPO: LoRA-style