Decode · tok/s
38.4
Prefill
—
tok/100W
16
Full GLM-5.3, EXL3 3.0bpw, TensorFold glm-dsa-tp4 at ddfad35. Single stream: 38.4 tok/s code, 24.9 prose.
First-person 4× GB10 ring. Rank is the quoted 38.4 code. 24.9 prose stays in the note. 63.0 is the C4 aggregate. Separate from the Int4/Int8 30.0 prose pair.
