Closed hemanth closed 4 months ago
Hi! It looks like you're making new API proposals, instead of commenting on the existing proposal. The best place to do that is https://github.com/WICG/proposals/, not another related proposal's repository.
Alternately, if you think there's something in the existing proposal that isn't serving your use case, it would be good to open an issue to explain your use case, without reference to any proposed solution or API surface. You should talk about what type of web application you are trying to build, why the current proposal is not sufficient for it, and what alternatives you could use instead. The result might end up best in this repository, if it's specifically about a deficit in the built-in AI proposal here, or in https://github.com/WICG/proposals/, if it's just somewhat related by virtue of being about AI.
https://whatwg.org/faq#adding-new-features is generally good guidance for this, and recommended reading.
Sorry, I mistakenly associated this repo with window.ai
. It is actually just for prompt-api.
I have created a new issue in WICG proposals.
window.ai.finetune
enables web applications to fine-tune nano/ any local AI models directly in the browser. This API allows developers to customize AI models based on specific datasets or raw text, enhancing the capabilities of client-side AI applications.Use Cases:
API:
window.ai.finetune(options: FineTuneOptions)
Initiates the fine-tuning process for a model based on the provided options.
options.model
: Specifies the base model to fine-tune. defaults to "nano".options.dataset
: An optional structured dataset for fine-tuning.options.rawText
: An optional string of raw text for fine-tuning.options.epochs
: An optional number of training epochs (default may vary by implementation).options.learningRate
: An optional learning rate for the fine-tuning process.Returns a Promise that resolves to a
FineTuneResult
object containing the ID of the fine-tuned model and performance metrics.