Beta
Explore
Marketplace
Neural Labs
Chat
Wallet
Docs
mobe-factor-cache-qwen3-30b-a3b-b32 – Dataset by AverageMetaheuristicsEnjoyer | AlphaNeural AI
You can deploy this model and start earning money today!
AverageMetaheuristicsEnjoyer
/
mobe-factor-cache-qwen3-30b-a3b-b32
like
0
apache-2.0
us
mobe
moe-compression
Views
No views yet
Model card
Files and Versions
Community
API
MoBE factor cache — Qwen3-30B-A3B (num_B=32, T=768)
Trained MoBE (Mixture-of-Basis-Experts) factors for the MoE gate_proj / up_proj matrices of Qwen3-30B-A3B, plus the perplexity / activation-geometry results of the compression sweep that produced them.
Contents
mobe_factor_cache/
/layer_
/
__nB<num_B>
t
_E
_R
_C
.pt Each file is a torch.save payload with factors (A, B, W), target_std, meta (hyperparameters, final loss… See the full description on the dataset page:
https://huggingface.co/datasets/AverageMetaheuristicsEnjoyer/mobe-factor-cache-qwen3-30b-a3b-b32
.
