When translating a language in the first time, it auto download the opus-mt model, it works fine in previous setting.
However, I switched to an network with only 1MB/s download speed.
(And I used a new container so the model is not cached).
After 2 minutes of download, it shows below error:
[CRITICAL] WORKER TIMEOUT
The model to be downloaded is 298M, which would need about 5 minutes to download, but the current setting will timeout in 120 seconds, which is not enough.
After the worker crashed, it is restarted automatically, and it will download the model again, but it will keep failing because it is impossible to finish download in 120 seconds under the current network.
Can we increase the timeout limit for downloading a model?
Even better, can we resume from previous download progress, or don't treat it as timeout as long as it is making progress within last window time period?
When translating a language in the first time, it auto download the opus-mt model, it works fine in previous setting.
However, I switched to an network with only 1MB/s download speed.
(And I used a new container so the model is not cached).
After 2 minutes of download, it shows below error:
The model to be downloaded is 298M, which would need about 5 minutes to download, but the current setting will timeout in 120 seconds, which is not enough.
After the worker crashed, it is restarted automatically, and it will download the model again, but it will keep failing because it is impossible to finish download in 120 seconds under the current network.
Can we increase the timeout limit for downloading a model?
Even better, can we resume from previous download progress, or don't treat it as timeout as long as it is making progress within last window time period?