nlpguy/Mistral-NeMo-Minitron-Upscale-v3
Partial fine-tune (fewer layers) vs nvidia/Mistral-NeMo-Minitron-8B-Base. nvidia/Mistral-NeMo-Minitron-8B-Base matches part of nlpguy/Mistral-NeMo-Minitron-Upscale-v3 (363 of its 561 tensors; the rest of nlpguy/Mistral-NeMo-Minitron-Upscale-v3 has no counterpart). On the shared part: Partial change: 52% of the weights changed (attention, mlp, norms), in 32 of 40 layers.
Tensors changed: Embeddings 0/1, Norms 56/81, MLP 72/120, Attention 96/160, lm_head 0/1
2 derivatives checked
- mradermacher/Mistral-NeMo-Minitron-Upscale-v3-i1-GGUF: Quantization
- mradermacher/Mistral-NeMo-Minitron-Upscale-v3-GGUF: Quantization