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
It looks like the timings are being ended for each preprocessor just as each one is run, rather than after it returns results. This means all the times are super short, and not useful for determining things like how long LMM remote calls are taking to execute. Log from recent query on unicorn below. Note that all the [Timing] tagged lines are immediately after the orchestrator sends the request, and the timestamps of the actual work being done by those same preprocessors comes only after the Timing information has already been output:
rchestrator-1 | 2024-12-07T23:11:07.740406431Z Running preprocessors in parallel...
orchestrator-1 | 2024-12-07T23:11:07.740433091Z Now on priority group 1
orchestrator-1 | 2024-12-07T23:11:07.745274019Z Sending to preprocessor "content-categoriser"
orchestrator-1 | 2024-12-07T23:11:07.749962151Z Sending to preprocessor "graphic-caption"
orchestrator-1 | 2024-12-07T23:11:07.753502938Z Sending to preprocessor "collage-detector-preprocessor"
orchestrator-1 | 2024-12-07T23:11:07.757077038Z Sending to preprocessor "line-charts"
orchestrator-1 | 2024-12-07T23:11:07.760584553Z Sending to preprocessor "autour-preprocessor"
orchestrator-1 | 2024-12-07T23:11:07.765583227Z [Timing] Preprocessor "content-categoriser": Execution Time: 20.177 ms, CPU Time: 20.176 ms, Normalized CPU Usage: 6.25%
orchestrator-1 | 2024-12-07T23:11:07.765600881Z [Timing] Preprocessor "graphic-caption": Execution Time: 15.613 ms, CPU Time: 16.213 ms, Normalized CPU Usage: 6.49%
orchestrator-1 | 2024-12-07T23:11:07.765606782Z [Timing] Preprocessor "collage-detector-preprocessor": Execution Time: 12.079 ms, CPU Time: 12.823 ms, Normalized CPU Usage: 6.635%
orchestrator-1 | 2024-12-07T23:11:07.765623954Z [Timing] Preprocessor "line-charts": Execution Time: 8.532 ms, CPU Time: 9.053 ms, Normalized CPU Usage: 6.632%
orchestrator-1 | 2024-12-07T23:11:07.765657888Z [Timing] Preprocessor "autour-preprocessor": Execution Time: 5.048 ms, CPU Time: 5.359 ms, Normalized CPU Usage: 6.635%
content-categoriser-1 | 2024-12-07T23:11:07.767434874Z [2024-12-07 23:11:07 +0000] [7] [DEBUG] POST /preprocessor
content-categoriser-1 | 2024-12-07T23:11:07.767831088Z DEBUG:root:Received request
collage-detector-preprocessor-1 | 2024-12-07T23:11:07.768269551Z [2024-12-07 23:11:07 +0000] [7] [DEBUG] POST /preprocessor
line-charts-1 | 2024-12-07T23:11:07.768272616Z [2024-12-07 23:11:07 +0000] [7] [DEBUG] POST /preprocessor
autour-preprocessor-1 | 2024-12-07T23:11:07.768732380Z [2024-12-07 23:11:07 +0000] [7] [DEBUG] POST /preprocessor
collage-detector-preprocessor-1 | 2024-12-07T23:11:07.769115319Z DEBUG:root:Received request
autour-preprocessor-1 | 2024-12-07T23:11:07.769549624Z DEBUG:root:Received request
line-charts-1 | 2024-12-07T23:11:07.769764618Z DEBUG:root:Received request
graphic-caption-1 | 2024-12-07T23:11:07.775121194Z [2024-12-07 23:11:07 +0000] [7] [DEBUG] POST /preprocessor
graphic-caption-1 | 2024-12-07T23:11:07.775386532Z DEBUG:root:Received request
content-categoriser-1 | 2024-12-07T23:11:07.778872506Z DEBUG:root:OLLAMA_URL https://ollama.pegasus.cim.mcgill.ca/ollama/api/generate
content-categoriser-1 | 2024-12-07T23:11:07.778904436Z DEBUG:root:OLLAMA_API_KEY looks properly formatted: sk-[redacted]
content-categoriser-1 | 2024-12-07T23:11:07.778960462Z DEBUG:root:serializing json from request_data dictionary
content-categoriser-1 | 2024-12-07T23:11:07.780270140Z DEBUG:root:serialization complete
content-categoriser-1 | 2024-12-07T23:11:07.780296289Z DEBUG:root:Posting request to ollama model llama3.2-vision:latest
content-categoriser-1 | 2024-12-07T23:11:07.781486654Z DEBUG:urllib3.connectionpool:Starting new HTTPS connection (1): ollama.pegasus.cim.mcgill.ca:443
graphic-caption-1 | 2024-12-07T23:11:07.782034202Z DEBUG:root:OLLAMA_URL https://ollama.pegasus.cim.mcgill.ca/ollama/api/generate
graphic-caption-1 | 2024-12-07T23:11:07.782057165Z DEBUG:root:OLLAMA_API_KEY looks properly formatted: sk-[redacted]
graphic-caption-1 | 2024-12-07T23:11:07.782065471Z DEBUG:root:serializing json from request_data dictionary
graphic-caption-1 | 2024-12-07T23:11:07.782835547Z DEBUG:root:Posting request to ollama model llama3.2-vision:latest
The text was updated successfully, but these errors were encountered:
It looks like the timings are being ended for each preprocessor just as each one is run, rather than after it returns results. This means all the times are super short, and not useful for determining things like how long LMM remote calls are taking to execute. Log from recent query on unicorn below. Note that all the [Timing] tagged lines are immediately after the orchestrator sends the request, and the timestamps of the actual work being done by those same preprocessors comes only after the Timing information has already been output:
The text was updated successfully, but these errors were encountered: