Loreto Parisi

**[UPDATE]** I did further checks, and the `classifier.train()` is a synchronized call, so basically the events can be intercepted in the `trainedWithDocument` callback, we need some exit condition like `obj.total...

By comparison Nvidia Megatron 11B, trained by Facebook AI in fairseq is provided as 19GB tar gz file hosted on their server farm: https://dl.fbaipublicfiles.com/fairseq/models/model_parallel/megatron_11b.tar.gz

@surrsurus thanks a lot. I will try to find out the API transcriptions at least for european languages.

@surrsurus another question. What about words not contained in the dictionary? The CMU has the [LOGIOS Lexicon Tool](http://www.speech.cs.cmu.edu/tools/lextool.html) (that is in perl) that generates a pronunciation dictionary to come up...

@surrsurus yes you are right, in fact I was stuck for that since looking at the perl code it outputs for sphinx text to speech tool rather than a standard...

@surrsurus Hello, I was trying to better understand the IPA format and its CMU relation. Now I can see that the CMU Dictionary has a split of the symbols with...

@surrsurus thank you. In fact I'm trying that approach simply doing for both grapheme and phoneme like ```python phonemes.append(list(split_line[1])) ``` I will update if it works as expected.

> All of the models are on huggingface already. https://huggingface.co/decapoda-research > > there's even an open, working pr to add support to the transformers lib. It's worth to note that...

yes confirmed. You first convert weights to ggml FP16 or FP32, then quantize to 4bit and run inference (cpu only).

yes this is exactly what I did from the download here.