Bug: whisper-13-v2-e3.bin doesn't load

ofersa commented 2 weeks ago

What happened?

A bug happened! I've build the bin using: from transformers import WhisperForConditionalGeneration import torch

model_suffix = 'whisper-13-v2-e3' model_name = f"ivrit-ai/{model_suffix}" model = WhisperForConditionalGeneration.from_pretrained(model_name)

model = model.float() torch.save(model.state_dict(), f"{model_suffix}.bin")

Steps to reproduce

step one...
step two...

What OS are you seeing the problem on?

No response

Relevant log output

options: {
  "path": "/Users/ofersa/Desktop/0000 21.01.17.mp4",
  "model_path": "/Users/ofersa/Library/Application Support/github.com.thewh1teagle.vibe/whisper-13-v2-e3.bin",
  "lang": "he",
  "verbose": false,
  "n_threads": 4,
  "init_prompt": "",
  "temperature": 0.4,
  "translate": null,
  "max_text_ctx": null
}

Caused by:
   0: failed to open model
   1: Failed to create a new whisper context.

Location:
    core/src/model.rs:49:6
App Version: 2.0.3
Commit Hash: 370d1ce20e8a69668c76fe44a51f5b71237d69e8
Arch: x86_64
Platform: macos
Kernel Version: 13.6.7
OS: macos
OS Version: 13.6.7
Models: ivrit-ai--whisper-large-v2-tuned-ggml-model.bin, ggml-medium.bin, whisper-13-v2-e3.bin, whisper-v2-d3-e3_f32.bin
Default Model: whisper-13-v2-e3.bin"

CPU feature detection is not supported on this architecture.

thewh1teagle commented 2 weeks ago

Hey, thanks for reporting! personally I don't know how to convert the transformers files into whisper compatible ggml format. I would check if there's ready to use model in huggingface.co/itayl If you found a way to convert the transformers to ggml format let me know, so I can add it to the instructions if someone else interested

ofersa commented 2 weeks ago

thanks for your answer. there is a previous version in https://huggingface.co/ivrit-ai/whisper-v2-d3-e3-ggml, but it's not the last version.

thewh1teagle / vibe