shuttleai/shuttle-3.5-ckpts
Adapter vs Qwen/Qwen3-32B. shuttleai/shuttle-3.5-ckpts is a LoRA adapter, not full weights: rank 64, on down_proj, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj in 64 layers (448 adapted matrices). It modifies Qwen/Qwen3-32B at load time.
1 other repos hold byte-for-byte identical weight files.