- Manage runs
- Runtime characteristics
How do I quickly delete failed runs?
Guild saves every run whether it succeeds or not. This lets you troubleshoot issues But over time failed runs can accumulate and you’ll want to delete them.
Use this command to delete failed runs:
guild runs delete --error
Use this command to delete terminated runs (i.e. runs that were
stopped by the user by typing
guild runs delete --terminated
You can delete both failed and terminated runs by using both
‑‑terminated options at the same time, or using
this short form:
guild runs delete -ET
Guild will let you confirm the list of runs before deleting them.
You can later restore a deleted run using (cmd:runs-restore)[runs restore].
If a source is referenced multiple times, does Guild download each occurrence?
No, Guild will only download the source once. There is no performance penalty for referencing a resource source multiple times.
How much overhead does Guild incur when running an operation?
Guild runs operations in a separate OS process to ensure that the operation is isolated. As of Guild 0.3.0, the additional overhead incurred when running an operation is as follows:
Additional time: typically less than 100 milliseconds but may be more on slower systems or loaded systems
Additional resident RAM: less than 40 MB
How do I know which library version I’m using?
The check command shows software library versions:
- Guild AI
- CUDA and cuDNN
To show this information, run:
To show more information, use the
guild check --verbose