Move inference snippets logic to inference package (#1247)

_Original slack message from @Wauplin ([private
link](https://huggingface.slack.com/archives/C04PJ0H35UM/p1740673808170509)):_

> I have an annoying problem I don't know how to handle in
`@hugingface.js`.
I'm working on updating the inference snippets under `@tasks` . Now that
we have a `makeUrl` helper for each provider (see [cohere
example](https://github.com/huggingface/huggingface.js/blob/b5230f9cbe71348014f39fd3a8149979cc10c196/packages/inference/src/providers/cohere.ts#L34)),
I want to use it in the CURL snippets to display to correct URL (for now
it's hardcoded
[here](https://github.com/huggingface/huggingface.js/blob/b5230f9cbe71348014f39fd3a8149979cc10c196/packages/tasks/src/snippets/curl.ts#L46)
which is incorrect for some providers). To do that, I need to make
`@tasks` depend on `@inference` . But `@inference` is already depending
on `@tasks` especially to type inference inputs/outputs. Is this
something npm/pnpm is able to settle (in Python it's not ^^).
> 
> The other solution I'm thinking about is to start having a `@snippets`
package depending on both `@tasks` and `@inference` but I'd like to
avoid that as much as possible 🙈 Any idea?

=> after some discussions we went for "let's move the `snippets` code to
`inference`" which this PR does. @julien-c @coyotte508

⚠️ **This is a breaking change for @huggingface.js/tasks.** Will require
to make a major release?

----

In practice I had to move only parts of the `./snippets` folder:
- the type definitions + utils remains in `@tasks` since it is used to
generate some local app snippets
- only the js/python/curl -specific parts are moved to `@inference` =>
i.e. only the logic to generate the snippets, not the helpers

e2e/ts/tsconfig.json

@@ -1,6 +1,6 @@
 {
 	"compilerOptions": {
-		"target": "ES2015",
+		"target": "ES2022",
 		"module": "commonjs",
 		"strict": true,
 		"esModuleInterop": true,

packages/inference/src/index.ts

@@ -2,3 +2,6 @@ export { HfInference, HfInferenceEndpoint } from "./HfInference";
 export { InferenceOutputError } from "./lib/InferenceOutputError";
 export * from "./types";
 export * from "./tasks";
+
+import * as snippets from "./snippets/index.js";
+export { snippets };

packages/tasks/src/snippets/curl.ts → packages/inference/src/snippets/curl.ts

@@ -1,9 +1,13 @@
-import { HF_HUB_INFERENCE_PROXY_TEMPLATE, type SnippetInferenceProvider } from "../inference-providers.js";
-import type { PipelineType } from "../pipelines.js";
-import type { ChatCompletionInputMessage, GenerationParameters } from "../tasks/index.js";
-import { stringifyGenerationConfig, stringifyMessages } from "./common.js";
-import { getModelInputSnippet } from "./inputs.js";
-import type { InferenceSnippet, ModelDataMinimal } from "./types.js";
+import { HF_HUB_INFERENCE_PROXY_TEMPLATE, type SnippetInferenceProvider } from "@huggingface/tasks";
+import type { PipelineType } from "@huggingface/tasks/src/pipelines.js";
+import type { ChatCompletionInputMessage, GenerationParameters } from "@huggingface/tasks/src/tasks/index.js";
+import {
+	type InferenceSnippet,
+	type ModelDataMinimal,
+	getModelInputSnippet,
+	stringifyGenerationConfig,
+	stringifyMessages,
+} from "@huggingface/tasks";
 
 export const snippetBasic = (
 	model: ModelDataMinimal,

packages/inference/src/snippets/index.ts

@@ -0,0 +1,5 @@
+import * as curl from "./curl.js";
+import * as python from "./python.js";
+import * as js from "./js.js";
+
+export { curl, python, js };

packages/tasks/src/snippets/js.ts → packages/inference/src/snippets/js.ts

@@ -1,9 +1,13 @@
-import { openAIbaseUrl, type SnippetInferenceProvider } from "../inference-providers.js";
-import type { PipelineType, WidgetType } from "../pipelines.js";
-import type { ChatCompletionInputMessage, GenerationParameters } from "../tasks/index.js";
-import { stringifyGenerationConfig, stringifyMessages } from "./common.js";
-import { getModelInputSnippet } from "./inputs.js";
-import type { InferenceSnippet, ModelDataMinimal } from "./types.js";
+import { openAIbaseUrl, type SnippetInferenceProvider } from "@huggingface/tasks";
+import type { PipelineType, WidgetType } from "@huggingface/tasks/src/pipelines.js";
+import type { ChatCompletionInputMessage, GenerationParameters } from "@huggingface/tasks/src/tasks/index.js";
+import {
+	type InferenceSnippet,
+	type ModelDataMinimal,
+	getModelInputSnippet,
+	stringifyGenerationConfig,
+	stringifyMessages,
+} from "@huggingface/tasks";
 
 const HFJS_METHODS: Partial<Record<WidgetType, string>> = {
 	"text-classification": "textClassification",

packages/tasks/src/snippets/python.ts → packages/inference/src/snippets/python.ts

@@ -1,9 +1,13 @@
-import { openAIbaseUrl, type SnippetInferenceProvider } from "../inference-providers.js";
-import type { PipelineType, WidgetType } from "../pipelines.js";
-import type { ChatCompletionInputMessage, GenerationParameters } from "../tasks/index.js";
-import { stringifyGenerationConfig, stringifyMessages } from "./common.js";
-import { getModelInputSnippet } from "./inputs.js";
-import type { InferenceSnippet, ModelDataMinimal } from "./types.js";
+import { openAIbaseUrl, type SnippetInferenceProvider } from "@huggingface/tasks";
+import type { PipelineType, WidgetType } from "@huggingface/tasks/src/pipelines.js";
+import type { ChatCompletionInputMessage, GenerationParameters } from "@huggingface/tasks/src/tasks/index.js";
+import {
+	type InferenceSnippet,
+	type ModelDataMinimal,
+	getModelInputSnippet,
+	stringifyGenerationConfig,
+	stringifyMessages,
+} from "@huggingface/tasks";
 
 const HFH_INFERENCE_CLIENT_METHODS: Partial<Record<WidgetType, string>> = {
 	"audio-classification": "audio_classification",

packages/inference/test/vcr.ts

@@ -191,9 +191,15 @@ async function vcr(
 				statusText: response.statusText,
 				headers: Object.fromEntries(
 					// Remove varying headers as much as possible
-					[...response.headers.entries()].filter(
-						([key]) => key !== "date" && key !== "content-length" && !key.startsWith("x-") && key !== "via"
-					)
+					(() => {
+						const entries: [string, string][] = [];
+						response.headers.forEach((value, key) => {
+							if (key !== "date" && key !== "content-length" && !key.startsWith("x-") && key !== "via") {
+								entries.push([key, value]);
+							}
+						});
+						return entries;
+					})()
 				),
 			},
 		};

packages/tasks-gen/scripts/generate-snippets-fixtures.ts

@@ -19,14 +19,14 @@ import { existsSync as pathExists } from "node:fs";
 import * as fs from "node:fs/promises";
 import * as path from "node:path/posix";
 
-import type { SnippetInferenceProvider, InferenceSnippet } from "@huggingface/tasks";
-import { snippets } from "@huggingface/tasks";
+import { snippets } from "@huggingface/inference";
+import type { SnippetInferenceProvider, InferenceSnippet, ModelDataMinimal } from "@huggingface/tasks";
 
 type LANGUAGE = "sh" | "js" | "py";
 
 const TEST_CASES: {
 	testName: string;
-	model: snippets.ModelDataMinimal;
+	model: ModelDataMinimal;
 	languages: LANGUAGE[];
 	providers: SnippetInferenceProvider[];
 	opts?: Record<string, unknown>;

packages/tasks/src/index.ts

@@ -45,11 +45,15 @@ export type {
 } from "./widget-example.js";
 export { SPECIAL_TOKENS_ATTRIBUTES } from "./tokenizer-data.js";
 
-import * as snippets from "./snippets/index.js";
 export * from "./gguf.js";
 
-export { snippets };
-export type { InferenceSnippet } from "./snippets/index.js";
+export {
+	type InferenceSnippet,
+	type ModelDataMinimal,
+	stringifyGenerationConfig,
+	stringifyMessages,
+	getModelInputSnippet,
+} from "./snippets/index.js";
 
 export { SKUS, DEFAULT_MEMORY_OPTIONS } from "./hardware.js";
 export type { HardwareSpec, SkuType } from "./hardware.js";

packages/tasks/src/snippets/index.ts

@@ -1,7 +1,3 @@
-import * as inputs from "./inputs.js";
-import * as curl from "./curl.js";
-import * as python from "./python.js";
-import * as js from "./js.js";
+export * from "./common.js";
+export * from "./inputs.js";
 export * from "./types.js";
-
-export { inputs, curl, python, js };