sophosympatheia/Midnight-Miqu-70B-v1.5
Full fine-tune vs migtissera/Tess-70B-v1.6. Fine-tune: 93% of the weights changed, across 80 of 80 layers.
Tensors changed: Norms 128/161, MLP 221/240, Attention 295/320
9 derivatives checked
- mradermacher/Midnight-Miqu-70B-v1.5-i1-GGUF: Quantization
- mradermacher/Midnight-Miqu-70B-v1.5-GGUF: Quantization
- backyardai/Midnight-Miqu-70B-v1.5-GGUF: Quantization
- honayb/Midnight-Miqu-70B-v1.5-i1-GGUF: Quantization
- hoborific/Midnight-Miqu-70B-v1.5-W4A16-AutoRound: Quantization
- AiAF/Midnight-Miqu-70B-v1.5_GGUF: Same weights, converted
- tensorblock/Midnight-Miqu-70B-v1.5-GGUF: Quantization
- mlx-community/Midnight-Miqu-70B-v1.5-MLX-8Bit: Quantization
- Sicarius-Prototyping/Midnight-Miqu-70B-v1.5_ChatML: Full fine-tune
2 other repos hold byte-for-byte identical weight files.