[Inference] Better types for HfInference #1121
Conversation
| // it("request - openai-community/gpt2", async () => { | ||
| // expect( | ||
| // await hf.request({ | ||
| // model: "openai-community/gpt2", | ||
| // inputs: "one plus two equals", | ||
| // }) | ||
| // ).toMatchObject([ | ||
| // { | ||
| // generated_text: expect.any(String), | ||
| // }, | ||
| // ]); | ||
| // }); |
There was a problem hiding this comment.
Still need to find a way around request and streamingRequest... They use a generic type T which messes thing up I think
| export class HfInference { | ||
| private readonly accessToken: string; | ||
| private readonly defaultOptions: Options; | ||
| export class HfInference implements Task { |
There was a problem hiding this comment.
export class HfInference implements Task {That bit makes sure HfInference defines a method for every task (textToImage, chatCompletion, etc)
| static mapInferenceFn<TOut, TArgs>(instance: HfInference, func: (...args: [TArgs, Options?]) => TOut) { | ||
| return function (...[args, options]: Parameters<(...args: [TArgs, Options?]) => TOut>): TOut { | ||
| return func({ ...args, accessToken: instance.accessToken }, { ...instance.defaultOptions, ...(options ?? {}) }); | ||
| }; | ||
| } |
There was a problem hiding this comment.
This helper functions adds some default arguments when calling HfInference.<task>() while keeping the exact type for inputs and outputs
There was a problem hiding this comment.
So jealous about how Typescript's seems so much powerful than Python's typing system... 😒
| @@ -40,3 +42,5 @@ export * from "./multimodal/visualQuestionAnswering"; | |||
| // Tabular tasks | |||
| export * from "./tabular/tabularRegression"; | |||
| export * from "./tabular/tabularClassification"; | |||
|
|
|||
| export const speechToText = automaticSpeechRecognition; | |||
There was a problem hiding this comment.
(unrelated) aliasing automaticSpeechRecognition as speechToText
There was a problem hiding this comment.
did it exist before or that's new?
There was a problem hiding this comment.
(why not i guess, though we should try to push the "official" task names otherwise it's hard to standardize one way of doing things)
There was a problem hiding this comment.
(why not i guess, though we should try to push the "official" task names otherwise it's hard to standardize one way of doing things)
yeah, good point
There was a problem hiding this comment.
Do you want to edit HfInferenceEndpoint too in the same file?
Also note the script packages/inference/scripts/generate-dts.ts used for generating definition files, you may need to change it now
(and we could switch from tsup to tshy removing the need for the script entirely maybe)
(unfinished, but I would love some inputs)
cc @coyotte508
Before this PR, Typescript was unable to determine actual types for the inputs of methods on the
HfInferenceclientHfInference.textToImagewould lack typings forparametersfor exampleThis PR ensures that all tasks that are exported in
src/tasks/indexare re-exported by the clientHfInference, and introduces a helper function that keeps every task's types along the way