BBQGOD/DeepSeek-GRM-27B

Full fine-tune vs google/gemma-2-27b. Fine-tune: 96% of the weights changed, across 46 of 46 layers.

Tensors changed: Embeddings 0/1, Attention 184/184, Norms 185/185, MLP 138/138

2 derivatives checked

• mradermacher/DeepSeek-GRM-27B-i1-GGUF: Quantization
• mradermacher/DeepSeek-GRM-27B-GGUF: Quantization

On the Hub · Downloads on Model Pulse