Aletheia-Bench/BatchOnline-GRPO-14B
Partial fine-tune vs deepseek-ai/DeepSeek-R1-Distill-Qwen-14B. Partial change: 88% of the weights changed (attention, mlp, norms), in 48 of 48 layers.
Tensors changed: Embeddings 0/1, Attention 332/336, MLP 141/144, Norms 90/97, lm_head 0/1
2 derivatives checked
- mradermacher/BatchOnline-GRPO-14B-i1-GGUF: Quantization
- mradermacher/BatchOnline-GRPO-14B-GGUF: Quantization
1 other repos hold byte-for-byte identical weight files.