BAAI/AREX-Base
Partial fine-tune (fewer layers) vs Qwen/Qwen3.5-122B-A10B. BAAI/AREX-Base matches part of Qwen/Qwen3.5-122B-A10B (1164 of its 1949 tensors; the rest of Qwen/Qwen3.5-122B-A10B has no counterpart). On the shared part: Partial change: 88% of the weights changed (attention, lm_head, mlp, norms), in 48 of 48 layers.
Tensors changed: MLP 322/448, lm_head 1/1, Embeddings 0/4, Attention 319/444, Norms 121/267
1 derivatives checked
- bartowski/BAAI_AREX-Base-GGUF: Same weights, converted
2 other repos hold byte-for-byte identical weight files.