Decode · tok/s
54.7
Prefill
—
tok/100W
22.8
CYBER-FROST 3.8, 3.87 bpw, with the repo's own bench script: code: 54.7 tok/s. DEPLOY.md says 53.2. devops: 49.0 tok/s. prose: 46.9 tok/s.
Same one-Spark re-run. Rank is the quoted 54.7 code. 49.0 and 46.9 stay in the note. Separate from the /cyber NVFP4 29.29 card. Post calls the finetune de-refused.