llm-jp/llm-jp-4-vl-9b
Full fine-tune (fewer layers) vs llm-jp/llm-jp-4-8b-thinking. llm-jp/llm-jp-4-8b-thinking matches part of llm-jp/llm-jp-4-vl-9b (289 of its 745 tensors; the rest of llm-jp/llm-jp-4-vl-9b has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 32 of 32 layers.
Tensors changed: Norms 51/65, MLP 96/96, Attention 128/128
2 derivatives checked
- tokinasin/llm-jp-4-vl-9b-GGUF-v0.1: Same weights, converted (fewer layers)
- mlx-community/llm-jp-4-vl-9b-mlx-4bit: Quantization