OpenMOSS / OpenMOSS/AnyGPT

Music Tokenizer Format

Open
#38 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
882
Forks
76
PR merge metrics
No merged PRs in 30d

Description

Hello, I was trying to train my own LLM on the encodec tokenizer and I wanted a bit of help. The LLM does not seem to learn the tokens and a drop-in replacement on SEED tokenizer works fine. The shape of the codes is [4, 250]. How do I format this in a sequence so it has causal dependency. Currently I have been doing it like the paper states. 4 codes per frame, and frame by frame. Is there something else I should look out for? In the outputs, it just repeats the 4 codes over and over again and the generated audio is just no sound.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

The issue does not name a file, entry point, or test. Start by reviewing the existing EnCodec token formatting and the referenced paper, then compare it with the working SEED tokenizer path. Done means identifying and documenting a causal sequence format that avoids repeated codes and produces audible output.

Written by the indexing model from the issue text.

Assessment

Domain
machine-learning
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.