netease-youdao/Confucius4-R2T2
Partial fine-tune vs Qwen/Qwen3-ASR-1.7B. Partial change: 84% of the weights changed (attention, embeddings, mlp, norms), in 28 of 28 layers.
Tensors changed: other 0/11, MLP 84/180, Norms 77/211, Attention 112/304, Embeddings 1/1
8 derivatives checked
- davidxifeng/Confucius4-R2T2-gguf: Same weights, converted
- netease-youdao/Confucius4-R2T2-GGUF: Same weights, converted (fewer layers)
- Nairod785/Confucius4-R2T2-Q4_K_M-GGUF: Quantization
- mradermacher/Confucius4-R2T2-GGUF: Same weights, converted (fewer layers)
- cstr/confucius4-r2t2-GGUF: Not comparable yet
- FIT17/Confucius4-R2T2-GGUF: Same weights, converted (fewer layers)
- mlx-community/Confucius4-R2T2-8bit: Quantization
- selcukkubur/Confucius4-R2T2-mlx-4bit: Quantization
1 other repos hold byte-for-byte identical weight files.