Motif-Technologies/Motif-3
Full fine-tune vs Motif-Technologies/Motif-3-Base. Fine-tune: 100% of the weights changed, across 53 of 53 layers.
Tensors changed: Embeddings 2/2, Norms 324/324, Attention 846/855, MLP 942/951, other 103/103, lm_head 1/1
8 derivatives checked
- Motif-Technologies/Motif-3-NVFP4: Quantization
- 0ppxnhximxr/Motif-3-FP8: Full fine-tune
- AesSedai/Motif-3-GGUF: Not comparable yet
- Baekpica/Motif-3-Mixed-Quant-GGUF: Not comparable yet
- 0ppxnhximxr/Motif-3-GGUF: Not comparable yet
- avar6/Motif-3-gguf: Not comparable yet
- Baekpica/Motif-3-GGUF: Not comparable yet
- prometheusAIR/Motif-3-GGUF: Not comparable yet