INSAIT-Institute/RAFT-14B

Partial fine-tune vs deepseek-ai/DeepSeek-R1-Distill-Qwen-14B. Partial change: 89% of the weights changed (attention, mlp, norms), in 48 of 48 layers.

Tensors changed: Embeddings 0/1, Attention 333/336, MLP 143/144, Norms 88/97, lm_head 0/1

2 derivatives checked

• mradermacher/RAFT-14B-i1-GGUF: Quantization
• mradermacher/RAFT-14B-GGUF: Quantization

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

On the Hub · Downloads on Model Pulse