Downloads · 30 days
0
qywu/Qwen3-30B-A3B-MoE-LoRA-Zero-Rank256
Qwen3-30B-A3B-MoE-LoRA-Zero-Rank256 is a machine learning model from qywu. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Zero-delta LoRA adapters for Qwen3-30B-A3B-Instruct-2507 in the sharedouter format, rank=256.
Downloads · 30 days
0
Access
Public
Updated Aug 24, 2026
Repo size
34 GB
Likes
0
Public
Click a slice to open those files.
.safetensors7.4 GB · 100%
From the Hugging Face model README
Zero-delta LoRA adapters for Qwen3-30B-A3B-Instruct-2507 in the shared_outer format, rank=256.
All LoRA delta matrices are zero-initialized — model output is identical to the base model. Used for performance testing (rank=256 kernel benchmarks) without requiring a trained adapter.
shared_outer/: shared outer B matrix (288 tensors, 6.9 GB). A fully
independent per-expert variant was previously shipped and has been removed.These carry no signal — they exist only to exercise the rank=256 kernel path. Load them exactly like the trained adapters in Qwen3-30B-A3B-MoE-LoRA-Password-Adapters; generations will match the base model.