powermove72/Shark-Coder
Partial fine-tune vs S-miguel/The-Trinity-Coder-7B. Partial change: 50% of the weights changed (attention, embeddings, mlp, norms), in 16 of 32 layers.
Tensors changed: MLP 48/96, Attention 64/128, Norms 32/65, Embeddings 1/1, lm_head 0/1
1 derivatives checked
- mradermacher/Shark-Coder-GGUF: Same weights, converted