ssmits/Falcon2-5.5B-Polish
Partial fine-tune (fewer layers) vs tiiuae/falcon-11B. ssmits/Falcon2-5.5B-Polish matches part of tiiuae/falcon-11B (172 of its 364 tensors; the rest of tiiuae/falcon-11B has no counterpart). On the shared part: Partial change: 10% of the weights changed (attention, mlp, norms), in 3 of 28 layers.
Tensors changed: Norms 6/58, MLP 6/56, Attention 6/56, Embeddings 0/1, lm_head 0/1
2 derivatives checked
- mradermacher/Falcon2-5.5B-Polish-i1-GGUF: Quantization
- mradermacher/Falcon2-5.5B-Polish-GGUF: Same weights, converted
6 other repos hold byte-for-byte identical weight files.