ssmits/Falcon2-8B-Dutch
Partial fine-tune (fewer layers) vs ssmits/Falcon2-5.5B-Dutch. ssmits/Falcon2-5.5B-Dutch matches part of ssmits/Falcon2-8B-Dutch (172 of its 262 tensors; the rest of ssmits/Falcon2-8B-Dutch has no counterpart). On the shared part: Partial change: 61% of the weights changed (attention, mlp, norms), in 19 of 28 layers.
Tensors changed: lm_head 0/1, Norms 38/58, MLP 38/56, Attention 38/56, Embeddings 0/1
1 derivatives checked
- mradermacher/Falcon2-8B-Dutch-GGUF: Same weights, converted
6 other repos hold byte-for-byte identical weight files.