llm-jp/llm-jp-4-vl-9b-beta
Full fine-tune (fewer layers) vs llm-jp/llm-jp-4-8b-instruct. llm-jp/llm-jp-4-8b-instruct matches part of llm-jp/llm-jp-4-vl-9b-beta (291 of its 745 tensors; the rest of llm-jp/llm-jp-4-vl-9b-beta has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 32 of 32 layers.
Tensors changed: Embeddings 1/1, Norms 49/65, MLP 96/96, Attention 128/128, lm_head 1/1
2 derivatives checked
- tokimoa/llm-jp-4-vl-9b-beta-mlx-4bit: Quantization
- tokimoa/llm-jp-4-vl-9b-beta-mlx-bf16: Same weights, repacked