OpenNMT / OpenNMT/CTranslate2

Is it possible to convert CTranslate2 weights back to Hugging Face Transformers?

Open
#2,031 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C++
Stars
4.7k
Forks
536
Avg merge
12h 12m
Merged PRs (30d)
4

Description

I converted the Whisper weights to CTranslate2:

ct2-transformers-converter --model openai/whisper-large-v3 --output_dir whisper-large-v3-ct2
--copy_files tokenizer.json preprocessor_config.json --quantization float16

Is it possible to convert CTranslate2 weights back to Hugging Face Transformers?
whisper-large-v3-ct2 > openai/whisper-large-v3

Contributor guide

Open the contributing guide

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 names the ct2-transformers-converter entry point and the whisper-large-v3-ct2 and openai/whisper-large-v3 model paths; start by tracing the existing conversion flow. Done would be a supported, documented reverse conversion that produces Hugging Face Transformers weights from the CTranslate2 directory, demonstrated with the Whisper example.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp, huggingface
Domain
machine-learning, tooling
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.