vadimayegorov/Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-oQ2BENCHMARKED-mlx-fp16

Same weights, converted vs Hunterx/Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-oQ2BENCHMARKED. Same weights stored in another dtype (U32/BF16 → U32/F16).

Tensors changed: Embeddings 0/3, Norms 0/301, Attention 0/1533, MLP 0/828, lm_head 0/2

On the Hub · Downloads on Model Pulse