Thread-scaling observation
The recorded thread sweep decreases from 1.68 tok/s at one thread to 1.28 tok/s at six threads.
That row pattern is reported only for the captured run set.
research.altifigence.com > notebooks > PCCX v002
PCCX v002 Throughput Record
Decode throughput and thread-scaling rows from the current v002 benchmark CSV bundle.
Source scope: existing thread-scaling and experiment CSV rows in this repository.
1.68 tok/sBest recorded thread row: Recorded at 1 thread in thread_scaling_results.csv.
0.66-1.12 tok/s64-token experiment range: Five recorded rows with 21 input tokens and 64 output tokens.
separate from measurementsTarget handling: The notebook keeps target discussion outside measured rows.
Measurement scope
The page uses only the repository CSV rows and simple arithmetic.
Any future target or acceptance value should remain separate until supported by a fresh benchmark record.