ehristoforu/BigFalcon3-from10B
Partial fine-tune (fewer layers) vs tiiuae/Falcon3-10B-Instruct. tiiuae/Falcon3-10B-Instruct matches part of ehristoforu/BigFalcon3-from10B (363 of its 588 tensors; the rest of ehristoforu/BigFalcon3-from10B has no counterpart). On the shared part: Partial change: 69% of the weights changed (attention, mlp, norms), in 30 of 40 layers.
Tensors changed: Embeddings 0/1, Norms 60/81, MLP 90/120, Attention 120/160, lm_head 0/1
2 derivatives checked
- mradermacher/BigFalcon3-from10B-i1-GGUF: Quantization
- mradermacher/BigFalcon3-from10B-GGUF: Quantization