ARA CommonsProduct
Research infrastructure, in one console
Connect the machines or cloud servers your team already uses. The platform launches runs, watches them, recovers failures, and records each result with the exact code and data behind it. The console below uses sample data; click around.
lr-sweep / lr=1e-4exp-101 · gpu-node-03done
lr-sweep / lr=3e-4exp-102 · gpu-node-04done
lr-sweep / lr=1e-3exp-103 · gpu-node-04failed
batch-size / 64exp-104 · gpu-node-01running
batch-size / 128exp-105 · gpu-node-02running
batch-size / 256exp-106 · —queued
data-v13 / baselineexp-107 · gpu-node-05failed
lr-sweep / lr=1e-3
exp-103 · failed- Machine
- gpu-node-04
- Last checkpoint
- step 3000
- Code version
- 9f3a1c2
- Data version
- v12
- Result (loss)
- —
Failed: Out of memory at step 3120
[queue] exp-103 submitted
[start] on gpu-node-04
[error] Out of memory at step 3120
[saved] step 3000
Your machines or ours
Run on your own servers, a cloud account, or a mix. Offered as a cloud service or installed on-premises.
Checkable results
Every number links to the code version and data version that produced it, so a colleague can verify it without re-running.
Shared history
Teammates can see what has already been tried before starting new work, which avoids repeated runs.