nvidia/llama-nv-embed-reasoning-3b

Full fine-tune vs meta-llama/Llama-3.2-3B. Fine-tune: 100% of the weights changed, across 28 of 28 layers.

Tensors changed: Embeddings 1/1, Norms 48/57, MLP 84/84, Attention 112/112

3 derivatives checked

• mradermacher/llama-nv-embed-reasoning-3b-GGUF: Same weights, converted
• DataScience-UIBK/OBLIQ-IR-3B-no-distill: LoRA-style
• DataScience-UIBK/OBLIQ-IR-3B: LoRA-style

On the Hub · Downloads on Model Pulse