sarcascoder/parakh — extraction eval gate
Field-level OCR/extraction accuracy evaluation. Fails the PR if accuracy regresses past a threshold.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Stale
- License
- None
Inputs
| name | description | required | default |
|---|---|---|---|
| config | Path to parakh config YAML (relative to the workspace root). | yes | eval/config.yaml |
| model | Model adapter to use for this run (must be defined in your parakh config or available as a built-in adapter). | no | default |
| threshold | Minimum overall field-level accuracy (0.0–1.0) required to pass the gate. | no | 0.90 |
| reading-order-threshold | Minimum reading-order accuracy (0.0–1.0). Optional. Set to 0 to disable. | no | 0.85 |
| golden | Path to the golden dataset directory (overrides what is set in the config). | no | "" |
| python-version | Python version used to install and run parakh. | no | 3.11 |
| upload-report | If true, upload the parakh HTML report as a GitHub artifact (default: true). | no | true |
| fail-on-regression | If true, exit non-zero when accuracy is below threshold (default: true). | no | true |
Outputs
| name | description |
|---|---|
| accuracy | Overall field-level accuracy achieved on this run. |
| reading-order | Reading-order accuracy on this run. |
| failed-fields | Number of fields below the per-field threshold. |
| report-path | Path to the HTML report produced by parakh. |