@huggingface/tasks 0.14.0 → 0.15.0

package/dist/esm/snippets/python.js

@@ -1,9 +1,14 @@
+import { HF_HUB_INFERENCE_PROXY_TEMPLATE, openAIbaseUrl } from "../inference-providers.js";
 import { stringifyGenerationConfig, stringifyMessages } from "./common.js";
 import { getModelInputSnippet } from "./inputs.js";
-const snippetImportInferenceClient = (model, accessToken) => `from huggingface_hub import InferenceClient
-client = InferenceClient("${model.id}", token="${accessToken || "{API_TOKEN}"}")
-`;
-export const snippetConversational = (model, accessToken, opts) => {
+const snippetImportInferenceClient = (accessToken, provider) => `\
+from huggingface_hub import InferenceClient
+
+client = InferenceClient(
+	provider="${provider}",
+	api_key="${accessToken || "{API_TOKEN}"}"
+)`;
+export const snippetConversational = (model, accessToken, provider, opts) => {
     const streaming = opts?.streaming ?? true;
     const exampleMessages = getModelInputSnippet(model);
     const messages = opts?.messages ?? exampleMessages;
@@ -21,9 +26,8 @@ export const snippetConversational = (model, accessToken, opts) => {
         return [
             {
                 client: "huggingface_hub",
-                content: `from huggingface_hub import InferenceClient
-
-client = InferenceClient(api_key="${accessToken || "{API_TOKEN}"}")
+                content: `\
+${snippetImportInferenceClient(accessToken, provider)}
 
 messages = ${messagesStr}
 
@@ -42,7 +46,7 @@ for chunk in stream:
                 content: `from openai import OpenAI
 
 client = OpenAI(
-	base_url="https://api-inference.huggingface.co/v1/",
+	base_url="${openAIbaseUrl(provider)}",
 	api_key="${accessToken || "{API_TOKEN}"}"
 )
 
@@ -64,9 +68,8 @@ for chunk in stream:
         return [
             {
                 client: "huggingface_hub",
-                content: `from huggingface_hub import InferenceClient
-
-client = InferenceClient(api_key="${accessToken || "{API_TOKEN}"}")
+                content: `\
+${snippetImportInferenceClient(accessToken, provider)}
 
 messages = ${messagesStr}
 
@@ -83,7 +86,7 @@ print(completion.choices[0].message)`,
                 content: `from openai import OpenAI
 
 client = OpenAI(
-	base_url="https://api-inference.huggingface.co/v1/",
+	base_url="${openAIbaseUrl(provider)}",
 	api_key="${accessToken || "{API_TOKEN}"}"
 )
 
@@ -100,8 +103,11 @@ print(completion.choices[0].message)`,
         ];
     }
 };
-export const snippetZeroShotClassification = (model) => ({
-    content: `def query(payload):
+export const snippetZeroShotClassification = (model) => {
+    return [
+        {
+            client: "requests",
+            content: `def query(payload):
 	response = requests.post(API_URL, headers=headers, json=payload)
 	return response.json()
 
@@ -109,9 +115,14 @@ output = query({
     "inputs": ${getModelInputSnippet(model)},
     "parameters": {"candidate_labels": ["refund", "legal", "faq"]},
 })`,
-});
-export const snippetZeroShotImageClassification = (model) => ({
-    content: `def query(data):
+        },
+    ];
+};
+export const snippetZeroShotImageClassification = (model) => {
+    return [
+        {
+            client: "requests",
+            content: `def query(data):
 	with open(data["image_path"], "rb") as f:
 		img = f.read()
 	payload={
@@ -125,37 +136,78 @@ output = query({
     "image_path": ${getModelInputSnippet(model)},
     "parameters": {"candidate_labels": ["cat", "dog", "llama"]},
 })`,
-});
-export const snippetBasic = (model) => ({
-    content: `def query(payload):
+        },
+    ];
+};
+export const snippetBasic = (model) => {
+    return [
+        {
+            client: "requests",
+            content: `def query(payload):
 	response = requests.post(API_URL, headers=headers, json=payload)
 	return response.json()
 	
 output = query({
 	"inputs": ${getModelInputSnippet(model)},
 })`,
-});
-export const snippetFile = (model) => ({
-    content: `def query(filename):
-    with open(filename, "rb") as f:
-        data = f.read()
-    response = requests.post(API_URL, headers=headers, data=data)
-    return response.json()
+        },
+    ];
+};
+export const snippetFile = (model) => {
+    return [
+        {
+            client: "requests",
+            content: `def query(filename):
+		with open(filename, "rb") as f:
+			data = f.read()
+		response = requests.post(API_URL, headers=headers, data=data)
+		return response.json()
+	
+	output = query(${getModelInputSnippet(model)})`,
+        },
+    ];
+};
+export const snippetTextToImage = (model, accessToken, provider) => {
+    return [
+        {
+            client: "huggingface_hub",
+            content: `\
+${snippetImportInferenceClient(accessToken, provider)}
 
-output = query(${getModelInputSnippet(model)})`,
-});
-export const snippetTextToImage = (model, accessToken) => [
-    {
-        client: "huggingface_hub",
-        content: `${snippetImportInferenceClient(model, accessToken)}
 # output is a PIL.Image object
-image = client.text_to_image(${getModelInputSnippet(model)})`,
-    },
-    {
-        client: "requests",
-        content: `def query(payload):
+image = client.text_to_image(
+	${getModelInputSnippet(model)},
+	model="${model.id}"
+)`,
+        },
+        ...(provider === "fal-ai"
+            ? [
+                {
+                    client: "fal-client",
+                    content: `\
+import fal_client
+
+result = fal_client.subscribe(
+	# replace with correct id from fal.ai
+	"fal-ai/${model.id}",
+	arguments={
+		"prompt": ${getModelInputSnippet(model)},
+	},
+)
+print(result)
+`,
+                },
+            ]
+            : []),
+        ...(provider === "hf-inference"
+            ? [
+                {
+                    client: "requests",
+                    content: `\
+def query(payload):
 	response = requests.post(API_URL, headers=headers, json=payload)
 	return response.content
+
 image_bytes = query({
 	"inputs": ${getModelInputSnippet(model)},
 })
@@ -164,23 +216,33 @@ image_bytes = query({
 import io
 from PIL import Image
 image = Image.open(io.BytesIO(image_bytes))`,
-    },
-];
-export const snippetTabular = (model) => ({
-    content: `def query(payload):
-	response = requests.post(API_URL, headers=headers, json=payload)
-	return response.content
-response = query({
-	"inputs": {"data": ${getModelInputSnippet(model)}},
-})`,
-});
+                },
+            ]
+            : []),
+    ];
+};
+export const snippetTabular = (model) => {
+    return [
+        {
+            client: "requests",
+            content: `def query(payload):
+		response = requests.post(API_URL, headers=headers, json=payload)
+		return response.content
+	response = query({
+		"inputs": {"data": ${getModelInputSnippet(model)}},
+	})`,
+        },
+    ];
+};
 export const snippetTextToAudio = (model) => {
     // Transformers TTS pipeline and api-inference-community (AIC) pipeline outputs are diverged
     // with the latest update to inference-api (IA).
     // Transformers IA returns a byte object (wav file), whereas AIC returns wav and sampling_rate.
     if (model.library_name === "transformers") {
-        return {
-            content: `def query(payload):
+        return [
+            {
+                client: "requests",
+                content: `def query(payload):
 	response = requests.post(API_URL, headers=headers, json=payload)
 	return response.content
 
@@ -190,11 +252,14 @@ audio_bytes = query({
 # You can access the audio with IPython.display for example
 from IPython.display import Audio
 Audio(audio_bytes)`,
-        };
+            },
+        ];
     }
     else {
-        return {
-            content: `def query(payload):
+        return [
+            {
+                client: "requests",
+                content: `def query(payload):
 	response = requests.post(API_URL, headers=headers, json=payload)
 	return response.json()
 	
@@ -204,11 +269,15 @@ audio, sampling_rate = query({
 # You can access the audio with IPython.display for example
 from IPython.display import Audio
 Audio(audio, rate=sampling_rate)`,
-        };
+            },
+        ];
     }
 };
-export const snippetDocumentQuestionAnswering = (model) => ({
-    content: `def query(payload):
+export const snippetDocumentQuestionAnswering = (model) => {
+    return [
+        {
+            client: "requests",
+            content: `def query(payload):
  	with open(payload["image"], "rb") as f:
   		img = f.read()
 		payload["image"] = base64.b64encode(img).decode("utf-8")  
@@ -218,7 +287,9 @@ export const snippetDocumentQuestionAnswering = (model) => ({
 output = query({
     "inputs": ${getModelInputSnippet(model)},
 })`,
-});
+        },
+    ];
+};
 export const pythonSnippets = {
     // Same order as in tasks/src/pipelines.ts
     "text-classification": snippetBasic,
@@ -249,23 +320,26 @@ export const pythonSnippets = {
     "image-to-text": snippetFile,
     "zero-shot-image-classification": snippetZeroShotImageClassification,
 };
-export function getPythonInferenceSnippet(model, accessToken, opts) {
+export function getPythonInferenceSnippet(model, accessToken, provider, opts) {
     if (model.tags.includes("conversational")) {
         // Conversational model detected, so we display a code snippet that features the Messages API
-        return snippetConversational(model, accessToken, opts);
+        return snippetConversational(model, accessToken, provider, opts);
     }
     else {
-        let snippets = model.pipeline_tag && model.pipeline_tag in pythonSnippets
-            ? pythonSnippets[model.pipeline_tag]?.(model, accessToken) ?? { content: "" }
-            : { content: "" };
-        snippets = Array.isArray(snippets) ? snippets : [snippets];
+        const snippets = model.pipeline_tag && model.pipeline_tag in pythonSnippets
+            ? pythonSnippets[model.pipeline_tag]?.(model, accessToken, provider) ?? []
+            : [];
+        const baseUrl = provider === "hf-inference"
+            ? `https://api-inference.huggingface.co/models/${model.id}`
+            : HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", provider);
         return snippets.map((snippet) => {
             return {
                 ...snippet,
-                content: snippet.content.includes("requests")
-                    ? `import requests
+                content: snippet.client === "requests"
+                    ? `\
+import requests
 
-API_URL = "https://api-inference.huggingface.co/models/${model.id}"
+API_URL = "${baseUrl}"
 headers = {"Authorization": ${accessToken ? `"Bearer ${accessToken}"` : `f"Bearer {API_TOKEN}"`}}
 
 ${snippet.content}`
@@ -274,6 +348,3 @@ ${snippet.content}`
         });
     }
 }
-export function hasPythonInferenceSnippet(model) {
-    return !!model.pipeline_tag && model.pipeline_tag in pythonSnippets;
-}

package/dist/esm/snippets/types.d.ts

@@ -7,6 +7,6 @@ import type { ModelData } from "../model-data.js";
 export type ModelDataMinimal = Pick<ModelData, "id" | "pipeline_tag" | "mask_token" | "library_name" | "config" | "tags" | "inference">;
 export interface InferenceSnippet {
     content: string;
-    client?: string;
+    client: string;
 }
 //# sourceMappingURL=types.d.ts.map

package/dist/esm/snippets/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/snippets/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAElD;;;;GAIG;AACH,MAAM,MAAM,gBAAgB,GAAG,IAAI,CAClC,SAAS,EACT,IAAI,GAAG,cAAc,GAAG,YAAY,GAAG,cAAc,GAAG,QAAQ,GAAG,MAAM,GAAG,WAAW,CACvF,CAAC;AAEF,MAAM,WAAW,gBAAgB;IAChC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;CAChB"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/snippets/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAElD;;;;GAIG;AACH,MAAM,MAAM,gBAAgB,GAAG,IAAI,CAClC,SAAS,EACT,IAAI,GAAG,cAAc,GAAG,YAAY,GAAG,cAAc,GAAG,QAAQ,GAAG,MAAM,GAAG,WAAW,CACvF,CAAC;AAEF,MAAM,WAAW,gBAAgB;IAChC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;CACf"}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@huggingface/tasks",
   "packageManager": "pnpm@8.10.5",
-  "version": "0.14.0",
+  "version": "0.15.0",
   "description": "List of ML tasks for huggingface.co/tasks",
   "repository": "https://github.com/huggingface/huggingface.js.git",
   "publishConfig": {

package/src/index.ts

@@ -58,3 +58,5 @@ export type { LocalApp, LocalAppKey, LocalAppSnippet } from "./local-apps.js";
 
 export { DATASET_LIBRARIES_UI_ELEMENTS } from "./dataset-libraries.js";
 export type { DatasetLibraryUiElement, DatasetLibraryKey } from "./dataset-libraries.js";
+
+export * from "./inference-providers.js";

package/src/inference-providers.ts

@@ -0,0 +1,16 @@
+export const INFERENCE_PROVIDERS = ["hf-inference", "fal-ai", "replicate", "sambanova", "together"] as const;
+
+export type InferenceProvider = (typeof INFERENCE_PROVIDERS)[number];
+
+export const HF_HUB_INFERENCE_PROXY_TEMPLATE = `https://huggingface.co/api/inference-proxy/{{PROVIDER}}`;
+
+/**
+ * URL to set as baseUrl in the OpenAI SDK.
+ *
+ * TODO(Expose this from HfInference in the future?)
+ */
+export function openAIbaseUrl(provider: InferenceProvider): string {
+	return provider === "hf-inference"
+		? "https://api-inference.huggingface.co/v1/"
+		: HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", provider);
+}

package/src/snippets/curl.ts

@@ -1,20 +1,35 @@
+import { HF_HUB_INFERENCE_PROXY_TEMPLATE, type InferenceProvider } 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";
 
-export const snippetBasic = (model: ModelDataMinimal, accessToken: string): InferenceSnippet => ({
-	content: `curl https://api-inference.huggingface.co/models/${model.id} \\
+export const snippetBasic = (
+	model: ModelDataMinimal,
+	accessToken: string,
+	provider: InferenceProvider
+): InferenceSnippet[] => {
+	if (provider !== "hf-inference") {
+		return [];
+	}
+	return [
+		{
+			client: "curl",
+			content: `\
+curl https://api-inference.huggingface.co/models/${model.id} \\
 	-X POST \\
 	-d '{"inputs": ${getModelInputSnippet(model, true)}}' \\
 	-H 'Content-Type: application/json' \\
 	-H 'Authorization: Bearer ${accessToken || `{API_TOKEN}`}'`,
-});
+		},
+	];
+};
 
 export const snippetTextGeneration = (
 	model: ModelDataMinimal,
 	accessToken: string,
+	provider: InferenceProvider,
 	opts?: {
 		streaming?: boolean;
 		messages?: ChatCompletionInputMessage[];
@@ -22,8 +37,13 @@ export const snippetTextGeneration = (
 		max_tokens?: GenerationParameters["max_tokens"];
 		top_p?: GenerationParameters["top_p"];
 	}
-): InferenceSnippet => {
+): InferenceSnippet[] => {
 	if (model.tags.includes("conversational")) {
+		const baseUrl =
+			provider === "hf-inference"
+				? `https://api-inference.huggingface.co/models/${model.id}/v1/chat/completions`
+				: HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", provider) + "/v1/chat/completions";
+
 		// Conversational model detected, so we display a code snippet that features the Messages API
 		const streaming = opts?.streaming ?? true;
 		const exampleMessages = getModelInputSnippet(model) as ChatCompletionInputMessage[];
@@ -34,8 +54,10 @@ export const snippetTextGeneration = (
 			max_tokens: opts?.max_tokens ?? 500,
 			...(opts?.top_p ? { top_p: opts.top_p } : undefined),
 		};
-		return {
-			content: `curl 'https://api-inference.huggingface.co/models/${model.id}/v1/chat/completions' \\
+		return [
+			{
+				client: "curl",
+				content: `curl '${baseUrl}' \\
 -H 'Authorization: Bearer ${accessToken || `{API_TOKEN}`}' \\
 -H 'Content-Type: application/json' \\
 --data '{
@@ -52,34 +74,64 @@ export const snippetTextGeneration = (
 		})},
     "stream": ${!!streaming}
 }'`,
-		};
+			},
+		];
 	} else {
-		return snippetBasic(model, accessToken);
+		return snippetBasic(model, accessToken, provider);
 	}
 };
 
-export const snippetZeroShotClassification = (model: ModelDataMinimal, accessToken: string): InferenceSnippet => ({
-	content: `curl https://api-inference.huggingface.co/models/${model.id} \\
+export const snippetZeroShotClassification = (
+	model: ModelDataMinimal,
+	accessToken: string,
+	provider: InferenceProvider
+): InferenceSnippet[] => {
+	if (provider !== "hf-inference") {
+		return [];
+	}
+	return [
+		{
+			client: "curl",
+			content: `curl https://api-inference.huggingface.co/models/${model.id} \\
 	-X POST \\
 	-d '{"inputs": ${getModelInputSnippet(model, true)}, "parameters": {"candidate_labels": ["refund", "legal", "faq"]}}' \\
 	-H 'Content-Type: application/json' \\
 	-H 'Authorization: Bearer ${accessToken || `{API_TOKEN}`}'`,
-});
+		},
+	];
+};
 
-export const snippetFile = (model: ModelDataMinimal, accessToken: string): InferenceSnippet => ({
-	content: `curl https://api-inference.huggingface.co/models/${model.id} \\
+export const snippetFile = (
+	model: ModelDataMinimal,
+	accessToken: string,
+	provider: InferenceProvider
+): InferenceSnippet[] => {
+	if (provider !== "hf-inference") {
+		return [];
+	}
+	return [
+		{
+			client: "curl",
+			content: `curl https://api-inference.huggingface.co/models/${model.id} \\
 	-X POST \\
 	--data-binary '@${getModelInputSnippet(model, true, true)}' \\
 	-H 'Authorization: Bearer ${accessToken || `{API_TOKEN}`}'`,
-});
+		},
+	];
+};
 
 export const curlSnippets: Partial<
 	Record<
 		PipelineType,
-		(model: ModelDataMinimal, accessToken: string, opts?: Record<string, unknown>) => InferenceSnippet
+		(
+			model: ModelDataMinimal,
+			accessToken: string,
+			provider: InferenceProvider,
+			opts?: Record<string, unknown>
+		) => InferenceSnippet[]
 	>
 > = {
-	// Same order as in js/src/lib/interfaces/Types.ts
+	// Same order as in tasks/src/pipelines.ts
 	"text-classification": snippetBasic,
 	"token-classification": snippetBasic,
 	"table-question-answering": snippetBasic,
@@ -108,13 +160,10 @@ export const curlSnippets: Partial<
 export function getCurlInferenceSnippet(
 	model: ModelDataMinimal,
 	accessToken: string,
+	provider: InferenceProvider,
 	opts?: Record<string, unknown>
-): InferenceSnippet {
+): InferenceSnippet[] {
 	return model.pipeline_tag && model.pipeline_tag in curlSnippets
-		? curlSnippets[model.pipeline_tag]?.(model, accessToken, opts) ?? { content: "" }
-		: { content: "" };
-}
-
-export function hasCurlInferenceSnippet(model: Pick<ModelDataMinimal, "pipeline_tag">): boolean {
-	return !!model.pipeline_tag && model.pipeline_tag in curlSnippets;
+		? curlSnippets[model.pipeline_tag]?.(model, accessToken, provider, opts) ?? []
+		: [];
 }