Comment by volotat

The model is way too small and undertrained to make any generalization claims. I want to wait until it reads the whole corpus I gave and then test it on some simple established benchmarks to see how it will behave.