HectorHe/DeepSeek-V2-Lite-sft-commonsense15k

Full fine-tune vs deepseek-ai/DeepSeek-V2-Lite. Fine-tune: 98% of the weights changed, across 27 of 27 layers.

Tensors changed: lm_head 1/1, Embeddings 0/1, Norms 24/82, MLP 2808/2808, Attention 108/108

1 derivatives checked

• mradermacher/DeepSeek-V2-Lite-sft-commonsense15k-GGUF: Quantization

On the Hub · Downloads on Model Pulse