Tatiana Likhomanenko
Tatiana Likhomanenko
It is possible =) @vineelpratap @avidov
yep, it is in the PR for now, need to fix some CI stuff but you can try it https://github.com/facebookresearch/flashlight/pull/524. Please comment if you have any troubles to use it...
Seems your old bin still doesn't have proper classes thus you cannot load model. Are you sure to use old_binary that has w2l::TransformerCriterion? Also cc @vineelpratap.
Sorry, I don't get what happened. So you have converted model and running decoding in fl v0.3 and see the error on the forward pass for the transformer block, right?...
@DongChanS Please change this https://github.com/flashlight/flashlight/blob/master/flashlight/app/asr/criterion/TransformerCriterion.cpp#L284 to ``` yBatched = layer(i)->forward(std::vector({yBatched}), fl::Variable(af::array())).front(); ``` and this https://github.com/flashlight/flashlight/blob/master/flashlight/app/asr/criterion/TransformerCriterion.cpp#L296 to ``` yBatched = layer(i)->forward(tmp, fl::Variable(af::array())).front(); ``` I will send this fix later, but this...
Feel free to send PR on this =)
Could you first run separately one of these failed tests and post here the full error log?
Ahh, sorry. We had problems with some of the CPU tests before and the way we used mkldnn. Recently we have switched to onednn and fixed/improved all CPU implementations and...
could you try to run with `--showletters=true` and put here the log? Could you post also your training log/config?