jinaai/jina-reranker-v3
Full fine-tune vs Qwen/Qwen3-0.6B. Fine-tune: 100% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 1/1, Norms 113/113, MLP 84/84, Attention 112/112
2 derivatives checked
- jinaai/jina-reranker-v3-GGUF: Same weights, converted (fewer layers)
- jinaai/jina-reranker-v3-mlx: Same weights, repacked
1 other repos hold byte-for-byte identical weight files.