Test the model
Goes to EWACS through this console, so nothing about the engine is
exposed to the browser. Answers are slow on CPU — a few seconds to a minute.
Forge — training
Loss over the run
Host
Swap devices
Controls
Collect
Recover
Source
Rebuild deletes the documents and re-queues the units, so the source is
harvested again with the current filters — nothing is permanently lost.
Purge also deletes the units, so the scan progress goes with them.
Both show you exactly what dies before anything happens.
Corpus
Where documents die
Accepted by language
Queue
Coverage — what the corpus holds
Coverage, not knowledge. How much Symfony
source we hold is a fact the corpus can state exactly; how much a model
knows is a question only an eval answers, and a percentage that conflated
the two would be worse than no number.
By origin — repository or package
Training samples
Jobs
| # | Kind | Agent | State | Secs | What it did | When |
|---|
Datasets and evals
| Dataset | Train | Eval | Tokens | Licences |
|---|
| Model | Dataset | n | Exact% | Contains% | F1 | When |
|---|
Inspect
The rows behind the numbers — every counting bug here so far was
only visible by reading stored output rather than trusting a total.
Leave both blank for the 100 most recent. Try lang
env tree
python json
dockerfile, or path
__init__ migrations
celery.| Path | Lang | Licence | Bytes | State |
|---|
Log
| When | Actor | What |
|---|