Alignment-Lab-AI/Mistral-nemo-3b-unhealed
Partial fine-tune (fewer layers) vs ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.3. Alignment-Lab-AI/Mistral-nemo-3b-unhealed matches part of ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.3 (138 of its 363 tensors; the rest of ArliAI/Mistral-Nemo-12B-ArliAI-RPMax-v1.3 has no counterpart). On the shared part: Partial change: 53% of the weights changed (attention, lm_head, mlp, norms), in 8 of 15 layers.
Tensors changed: Embeddings 0/1, Norms 16/31, MLP 24/45, Attention 32/60, lm_head 1/1
1 derivatives checked
- mradermacher/Mistral-nemo-3b-unhealed-GGUF: Same weights, converted