run-0636
performance
citable URL: https://halobench.com/records/run-0636/ — this address never moves; the anchor /records/#run-0636 keeps resolving
date 2026-08-27 · window 2026-08-27T20:20:00Z → 2026-08-27T20:46:35Z
suite llama-bench-decode-depth · config cfg-0178
cell depth 204,800 tokens (machine-written trailer)
Metrics
| metric | value |
|---|---|
| decode_tps | 5.73 |
| prefill_tps | 63.9 |
Guard chain
waiver — Throughput ladder; output-sanity inherited from run-0632 (tau2 capability, same binary/model/quant, valid outputs). No separate 4/4 needle guard for the perf cells.
Cited by — computed at build time, never stored
claims clm-0125