SteelStorage/BA-Zephyria-39b
Partial fine-tune (fewer layers) vs unsloth/Mistral-Small-Instruct-2409. unsloth/Mistral-Small-Instruct-2409 matches part of SteelStorage/BA-Zephyria-39b (507 of its 894 tensors; the rest of SteelStorage/BA-Zephyria-39b has no counterpart). On the shared part: Partial change: 26% of the weights changed (attention, mlp, norms), in 15 of 56 layers.
Tensors changed: Embeddings 0/1, Norms 30/113, MLP 45/168, Attention 60/224, lm_head 0/1
4 derivatives checked
- mradermacher/BA-Zephyria-39b-i1-GGUF: Quantization
- mradermacher/BA-Zephyria-39b-GGUF: Quantization
- TheDrummer/Tunguska-39B-v1-GGUF: Full fine-tune
- TheDrummer/Tunguska-39B-v1: Full fine-tune