slalom/qai-cli
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

include-metrics-from-training (#6)

Browse Source 
Reviewed-on: #6
This commit was merged in pull request #6.
This commit is contained in:
slalom
2026-06-12 18:23:25 +00:00
parent 522ddc74e2
commit a1ffbb77c5
13 changed files with 785 additions and 116 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
examples/meter-detection/README.md
View File

@@ -153,6 +153,20 @@ Or pass the job name explicitly:
qc-cli train status qc-cli-YYYYMMDD-HHMMSS
```
To submit the job, wait for completion, and automatically import metrics and register the model, run:
```bash
qc-cli train start --upload-metrics
```
The default polling interval is 30 seconds. It can be changed with `--poll-interval <seconds>`.
The metrics can be also submitted using:
```bash
qc-cli mlflow upload-metrics
```
## SageMaker Outputs
When the job completes, SageMaker packages the files written under `/opt/ml/model` into `model.tar.gz`.
@@ -163,10 +177,15 @@ This example writes:
best.pt
model.onnx
metrics.json
training_metrics.json
```
The archive is stored under the configured `s3.model_prefix`.
The `mlflow upload-metrics` command imports `training_metrics.json`, which provides per-epoch training and validation
losses, precision, recall, mAP@0.50, mAP@0.50:0.95, and learning rates. For object detection, mAP and precision/recall
are more meaningful than classification accuracy when assessing model quality.
## 6. Configure Qualcomm AI Hub
Authenticate with Qualcomm AI Hub: