|
Download docs/task-progress.md from Intel/low_bit_open_llm_leaderboard: direct link, hf CLI and curl.
- Browser
- Download file 1.2 kB
-
https://huggingface.co/spaces/Intel/low_bit_open_llm_leaderboard/resolve/main/docs/task-progress.md
- Command line
-
hf download hf://spaces/Intel/low_bit_open_llm_leaderboard/docs/task-progress.md
-
curl -L -o task-progress.md https://huggingface.co/spaces/Intel/low_bit_open_llm_leaderboard/resolve/main/docs/task-progress.md
1.2 kB
| # Task progress | |
| The Quantization and Evaluation queues show Azure DevOps pipeline-step progress. | |
| The dispatcher fetches the Build Timeline API for active runs every 60 seconds, | |
| using the existing server-side `AZP_DEVOPS_PAT`. The queue UI refreshes every five | |
| seconds and reads an in-memory snapshot; browser refreshes do not call Azure. | |
| The app and dispatcher must run in the same process (as with `python app.py`). | |
| The bar counts completed Task records, including skipped steps, and displays the | |
| currently running task names. It measures pipeline steps, not elapsed time, | |
| AutoRound iterations, or evaluation samples. Steps can differ substantially in | |
| duration. Retried steps use their latest attempt. Snapshots older than three | |
| minutes are no longer displayed; missing credentials, unavailable timelines, and | |
| API failures produce an indeterminate bar with “Waiting for progress information”. | |
| For iteration/sample percentages, the external quantization/evaluation runner | |
| must additionally expose structured progress (completed units, total units, | |
| stage, run ID, and timestamp), through an authenticated endpoint or status artifact. | |
| The current integration requires no changes to that runner. | |