textstream
project: Train GPT model using text stream in browser in node (without loading the whole dataset into memory)- Support both
maxIter
andepochs
in thetrain
method. Switched fromdf.forEachAsync
towhile
loop +ds.iterator()
(it was not clear how to terminate long-runningforEachAsync
).
- Memory leakage in the
generate
method - Use
config.lr
in the Adam optimizer