serious question
I think I speak for a lot of people when I ask you: why are you still focusing your energy on the totally antiquated, worst asr model ever created? I mean, Whisper is 3-4 years old. There are a litany of orders of magnitude better models that you would be doing a real service to the community by focusing on instead of the completely dead, completely useless Whisper crap-model.
I'm sorry to hear about that experience with Whisper.Gicen the rise of more accurate and smaller models like Parakeet, I do have the background knowledge myself to understand what you're saying.
The thing is, despite Whisper's age, it's still a widely used model. Fine-tunes aren't uncommon; in fact, whenever I look for new Whisper fine-tunes to quantize, I sort by most downloads and choose the first model I see that I haven't already done and am able to do.
Quite often, I see these fine-tunes have tens of thousands of downloads in a month, telling me that there's still many people using Whisper models. As such, I still find Whisper models to be worth my time as the numbers are showing me so.
I hope you find this explanation helpful to your question.
Sure, it's helpful, but I think, given the popularity of your work, this is a chicken-and-egg issue. If you were to shift your focus to Parakeet-tdt-0.6b-v3, Canary-1b-v2, Qwen3-ASR-0.6b & 1.7b, just to name a select few--the fine-tunes would follow. All of those three models I named--and those are just a tiny representation of the huge array of super-sophisticated, far superior-than-Whisper models that have dropped in only the last six months or so--are, again, orders of magnitude better than ANY Whisper variant or fine-tune in existence. That's just a fact. So if you were to just start, you would a) help the lack of education problem, where folks who are not devs think ASR = Whisper; b) you would be doing everyone a genuine service by releasing single-file, quantized models for these superior, popular models, and c) you would be doing yourself a service by doing work that is actually timely and relevant to the current state of ASR technology. Whisper is objectively none of those things. It doesn't deserve the clear passion and effort you put into your work. At least consider branching out. It genuinely makes me a bit sad or disillusioned that you have such passion and talent and are misdirecting it on such a shit model..