wolfram/miqu-1-103b
Partial fine-tune (fewer layers) vs 152334H/miqu-1-70b-sf. 152334H/miqu-1-70b-sf matches part of wolfram/miqu-1-103b (723 of its 1083 tensors; the rest of wolfram/miqu-1-103b has no counterpart). On the shared part: Partial change: 50% of the weights changed (attention, mlp, norms), in 40 of 80 layers.
Tensors changed: Embeddings 0/1, Norms 80/161, MLP 120/240, Attention 160/320, lm_head 0/1
2 derivatives checked
- mradermacher/miqu-1-103b-i1-GGUF: Quantization
- mradermacher/miqu-1-103b-GGUF: Quantization