Single ckpt file to use as a local checkpoint about whisper-jax

[sanchit-gandhi]

Hey @senceryucel - could you put the checkpoint into a folder, and then pass the path to this folder to the pipeline? The pipeline will then complain that you don't have a config, feature extractor or tokenizer defined, in which case you should also save these to the dir:

from transformers import AutoConfig, AutoProcessor

# set the path to your output dir
output_dir = ...
# define the HF Hub identifier of the original model, e.g. "openai/whisper-large-v2" for the large-v2 model
model_id = "openai/whisper-large-v2"

config = AutoConfig.from_pretrained(model_id)
processor = AutoProcessor.from_pretrained(model_id)

config.save_pretrained(output_dir)
processor.save_pretrained(output_dir)

You should now be able to pass checkpoint=output_dir to the pipeline.