how to add local model instead of the official ones

[abdelkareemkobo]

I am trying to use whisper to transcribe my data I want to use the following code

from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq

processor = AutoProcessor.from_pretrained("arbml/whisper-largev2-ar")

model = AutoModelForSpeechSeq2Seq.from_pretrained("arbml/whisper-largev2-ar")

result = model.transcribe("/kaggle/input/testing-audio/oh.wav")
print(result["text"])

It gives me "AttributeError: 'WhisperForConditionalGeneration' object has no attribute 'transcribe'"
I wonder if i can use this model with the following code from the current repo

import whisper

model = whisper.load_model("base")
result = model.transcribe("audio.mp3")
print(result["text"])

The model is in pytorch form

[phineas-pta]

your title is a bit misleading, u r using transformers interface

see my comment for example usage

#654 (comment)

[abdelkareemkobo]

I don't want ot use transformers interface I want to use the model in the their hub i donwloaded it and tryied to add to the official whisper repo to use the CLI and other stuff from it but can't do this.
I tried your code it's very helfpul and this really what i need! Thanks for it but i get an error while using the fine-tuned model from huggingface with
──────────────────
ValueError: You are trying to return timestamps, but the generation config is not properly set.Make sure to
initialize the generation config with the correct attributes that are needed such as no_timestamps_token_id.For
more details on how to generate the approtiate config, refer to
huggingface/transformers#21878 (comment)
──────────────────

from transformers import pipeline
import torch
device = "cuda" if torch.cuda.is_available() else "cpu"

model = pipeline(
    task="automatic-speech-recognition",
    model="arbml/whisper-medium-ar",
    device=device,
    chunk_length_s=30,
    generate_kwargs = {"num_beams":5}#same as setting as "open whisper" default
)
result = model("/content/001.mp3",return_timestamps=True,generate_kwargs={"language":"arabic","task":"transcribe"})

can you help me how to resolve this error?

[phineas-pta]

seem like your model isn't configured to return timestamps

you can simply remove return_timestamps=True

if you want to use whisper CLI, you have to edit source code to load fine-tuned checkpoint, you can inspire from this #830 (comment)