Fulcrum-AI/Ryze-Embed-2
Partial fine-tune (fewer layers) vs Alibaba-NLP/gte-Qwen2-7B-instruct. Alibaba-NLP/gte-Qwen2-7B-instruct matches part of Fulcrum-AI/Ryze-Embed-2 (339 of its 627 tensors; the rest of Fulcrum-AI/Ryze-Embed-2 has no counterpart). On the shared part: Partial change: 67% of the weights changed (attention, mlp, norms), in 22 of 28 layers.
Tensors changed: Embeddings 0/1, Norms 44/57, MLP 66/84, Attention 154/196, lm_head 0/1
1 derivatives checked
- mradermacher/Ryze-Embed-2-GGUF: Quantization