Banaxi-Tech/BananaMind-Translate-V3.5-Pro

Partial fine-tune vs unsloth/gemma-4-E4B-it. Partial change: 50% of the weights changed (attention, mlp, other), in 42 of 42 layers.

Tensors changed: MLP 126/606, Norms 0/484, other 85/261, Attention 132/719, Embeddings 0/6

2 derivatives checked

• mradermacher/BananaMind-Translate-V3.5-Pro-i1-GGUF: Quantization (fewer layers)
• mradermacher/BananaMind-Translate-V3.5-Pro-GGUF: Same weights, converted (fewer layers)

On the Hub · Downloads on Model Pulse