Hi, I would like to train my own model with the same hyperparameters as the official english model (https://github.com/mozilla/DeepSpeech/releases/tag/v0.8.2) but i have no idea how long it would take with for example 1000 hours of data.
Someone would have any idea how long it would take with lower hardware? Thanks.
What is “lower hardware” ? We already document on the release page how much data we trained in how long on what hardware, so you should be able to figure out from that.
I can’t find the time spent to train the model on the release page. I would just like to know how much time they spent to train this model with 8 quadro RTX 6000 and how much time it could take with for example only 1 RTX 2080. Thanks
Looks like the length info disappeared. @reuben can correct me but it was something around a full week.
I train with 2x RTX 2080 Ti a french model of ~1000h in ~18h without Automatic Mixed Precision on ~30 epochs.
Ok, thanks for the answer!
@lissyx Sir, only one question, may I ask that in addition to GPU, how much RAM is on your machine for the training? for 1000 hrs data?
I’m unsure the amount of RAM is really relevant, my current system has 128GiB, but I have not had a look at the amount consumed by the training process.