Kaushik Ram Sadagopan

Hi @YJYJLee, could you provide the command you used to encounter this error? I checked out the main branch of the repo: https://github.com/facebookresearch/seamless_communication and ran this command: ``` m4t_predict /large_experiments/seamless/ust/data/TTS/vocoder_training/audio_wavs/multi_spkr/eng/eng_LJSpeech-1.1_0/LJ003-0001.wav...

m4t_evaluate also works for me with the medium model.

Are you sure you're using the latest version of main?

In fact, could you try my m4t_predict command above and see if it works for you?

The README for `prepare_data` answers this. (https://github.com/facebookresearch/stopes/tree/main/stopes/pipelines/prepare_data) Here's an example config: ``` binarization_config: binarize_workers: 60 max_examples_per_shard: 500000000 random_seed: 0 smallest_shard: 500000 executor_config: cluster: local log_folder: executor_logs preprocessing_config: max_tokens: null moses_config:...

You need to run the filtering pipeline to filter out data based on the following heuristics: length, deduplication, LASER margin score threshold, LID score thresholds, toxicity. It's not sufficient to...

Apologies, you are correct. Currently the filtering pipeline doesn't output the input config of the prepare_data pipeline which is inconvenient for user. We're working on completely refactoring the two pipelines...

Looks like you killed the process yourself on the command line, could you let it run entirely?

> I'm not super sure of the intention of this PR, can I ask a few questions: > > * Why having both inference/ and predict/ subdirectories, as they're seem...

predict falls under the CLI directory. Inference is a directory of its own, feel free to suggest a better name to make it clear.