wozeparrot
wozeparrot
**Release Type**: Official Release/GitHub (please choose appropriate option) Official **Version**: Version number and/or git branch 3.1.0.1-beta01-0359 **Platform(s)**: Does the problem occur on Windows, Android...? Windows **Describe the bug** A clear...
For now dataset has to be downloaded manually from https://www.openslr.org/12/. mlperf uses `dev-clean`. Then all the flacs have to be converted to wav using something like: ```fish for file in...
Language models are so much easier to write than RNN-T :sweat_smile: ### TODO - [ ] Dataset downloader - [x] Metrics - [x] ~~Do a run~~ Edit: initial run at...
This is the first in what will hopefully be a series of prs to get RNN-T training working. This adds the [LAMB optimizer](https://arxiv.org/abs/1904.00962) based on the official tensorflow implementation [here](https://github.com/tensorflow/addons/blob/master/tensorflow_addons/optimizers/lamb.py)....
Wooo, Bert training time. :rocket: Going to try to do a run where I overfit on the eval set just to make sure everything is working. Kinda annoying that its...
This is needed for #873 and maybe one other mlperf model This matches the pytorch implementation. There is a strange bug? I'm trying to fix with two specific high lr...
Inspired by #878 and kinda needed for mlperf training? not sure if mlperf specifies the checkpoint format. We might want to look into having a standardized way of storing model...
This adds a pretty simple quick start guide/tutorial that goes over the high level tinygrad API. ~~I do think it would be good to mention the JIT in this, probably...
Opening for visibility. This currently works for the CPU, LLVM, and CLANG backends. But as we are mainly targeting multigpu stuff this needs more thought. The main problem right now...
**Describe the bug** When cloning a large repository (in this case https://github.com/NixOS/nixpkgs) the cloned pack is corrupted. Bisect points to d4566d4aaa426b33015780c7cc18f887fc07cca4 as the bad commit, which is supposed to be...