You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 57199f0
Browse filesBrowse the repository at this point in the historyBrowse files
Copy file name to clipboardExpand all lines: docs/results-and-ingestion.md
+6Lines changed: 6 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -96,6 +96,12 @@ Single-node GPU count is `tp * pp * pcp_size`. DCP does not multiply the physica
96
96
97
97
InferenceX-app treats routing fields as columns or config dimensions and stores numeric measurements in `benchmark_results.metrics` JSONB. The mapper supports v1 shared topology, v2 split prefill/decode topology, and nested v3 AgentX metrics. Unknown numeric metrics are retained and warned about, which permits schema growth without silently losing numeric data.
98
98
99
+
### Power telemetry stop and validation
100
+
101
+
The single-node AMD monitor in [`benchmark_lib.sh`](../benchmarks/benchmark_lib.sh) waits for every observed GPU to have a positive, numeric power sample at or beyond the first whole second after the stop request. `AMD_MONITOR_STOP_TIMEOUT_S` bounds the wait (default `30`; `0` skips it). Streams without usable epoch timestamps use the legacy fixed tail wait. A timeout does not certify coverage: the aggregator still rejects an unbracketed benchmark window. AgentX cancellation skips the coverage wait and stops the monitor, including when cancellation arrives during a normal stop.
102
+
103
+
[`single_node.py`](../infx/results/power/single_node.py) excludes unusable power rows outside the formal window from integration and bracketing. Within the boundary search band, these rows are counted per GPU in `power_validation.json` under `boundary_degenerate_rows`, even when no usable samples remain. In-window validation is unchanged; these counters are diagnostic evidence, not replacement measurements.
0 commit comments