vllm-sr/elephant-embeddings-v1-text-small

Full fine-tune vs vllm-sr/mmbert-32k-yarn. Fine-tune: 100% of the weights changed, across 22 of 22 layers.

Tensors changed: Embeddings 2/2, Norms 1/44, Attention 44/44, MLP 44/44

3 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse