MarsupialAI/Lusca-33B
Full fine-tune (fewer layers) vs mistralai/Mistral-Small-Instruct-2409. mistralai/Mistral-Small-Instruct-2409 matches part of MarsupialAI/Lusca-33B (507 of its 759 tensors; the rest of MarsupialAI/Lusca-33B has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 56 of 56 layers.
Tensors changed: Embeddings 1/1, Norms 82/113, MLP 168/168, Attention 224/224, lm_head 1/1
2 derivatives checked
- mradermacher/Lusca-33B-i1-GGUF: Quantization
- mradermacher/Lusca-33B-GGUF: Quantization