Supermix v88 (v88_corpus.selected.pt)
Standalone milestone release for v88_corpus.selected.pt from kai9987kai/Supermix β the 30.2M-parameter (5.69M active/token) MiMoMix MoE + Recursive Thinking Core model bridging supermix-v87 and supermix-v89.
Verified Training & Generalisation Metrics (Windows CPU)
- Parameters:
30,211,709total /5,693,933active per token (64routed experts, top-2,2MTP speculative draft layers). - Selected Checkpoint: step
21,000(selected_on = accuracy, in-training probe accuracy0.95,best_dev_loss = 0.025629). - Generalisation Tiers:
tier1_seen_responseloss0.014598,tier3_unseen_sentenceloss0.195604, all5/5v58generalisation checkspassed = True. - Receipts included:
results/generalisation_results.json,results/supervised_run.json.
Quick Start
pip install torch numpy flask huggingface_hub
huggingface-cli download Kai9987kai/supermix-v88 --local-dir .
python example_usage.py
- Downloads last month
- -
Model tree for Kai9987kai/supermix-v88
Base model
Kai9987kai/supermix-v87