One row in the benchmark result table.
Each BenchmarkRecord represents a single (config, backend) run.
-
platform
Hardware platform (e.g., 'cpu', 'gpu', 'tpu').
- Type:
str
-
backend
Backend name (e.g., 'numba', 'pallas', 'warp').
- Type:
str
-
label
Configuration label (from BenchmarkConfig.name).
- Type:
str
-
mean_ms
Mean execution time in milliseconds.
- Type:
float
-
std_ms
Standard deviation of execution time in milliseconds.
- Type:
float
-
min_ms
Minimum execution time in milliseconds.
- Type:
float
-
throughput
Optional throughput metric (e.g., GFLOPs/s). None if not
computed.
- Type:
float or None
-
success
Whether the benchmark run completed without error.
- Type:
bool
-
error
Error message if the run failed; None on success.
- Type:
str or None
-
kernel_kwargs
Kernel parameters used for this run (forwarded to the call
function).
- Type:
dict
-
data_kwargs
Data-description metadata for this run (not forwarded to the
kernel).
- Type:
dict
