@huggingface/inference 3.6.0 → 3.6.2

package/dist/index.cjs

@@ -1,9 +1,7 @@
 "use strict";
-var __create = Object.create;
 var __defProp = Object.defineProperty;
 var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
 var __getOwnPropNames = Object.getOwnPropertyNames;
-var __getProtoOf = Object.getPrototypeOf;
 var __hasOwnProp = Object.prototype.hasOwnProperty;
 var __export = (target, all) => {
   for (var name2 in all)
@@ -17,14 +15,6 @@ var __copyProps = (to, from, except, desc) => {
   }
   return to;
 };
-var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__getProtoOf(mod)) : {}, __copyProps(
-  // If the importer is in node compatibility mode or this is not an ESM
-  // file that has been converted to a CommonJS file using a Babel-
-  // compatible transform (i.e. "__esModule" has not been set), then set
-  // "default" to the CommonJS "module.exports" for node compatibility.
-  isNodeMode || !mod || !mod.__esModule ? __defProp(target, "default", { value: mod, enumerable: true }) : target,
-  mod
-));
 var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
 
 // src/index.ts
@@ -114,6 +104,9 @@ var HF_ROUTER_URL = "https://router.huggingface.co";
 
 // src/providers/black-forest-labs.ts
 var BLACK_FOREST_LABS_AI_API_BASE_URL = "https://api.us1.bfl.ai";
+var makeBaseUrl = () => {
+  return BLACK_FOREST_LABS_AI_API_BASE_URL;
+};
 var makeBody = (params) => {
   return params.args;
 };
@@ -128,7 +121,7 @@ var makeUrl = (params) => {
   return `${params.baseUrl}/v1/${params.model}`;
 };
 var BLACK_FOREST_LABS_CONFIG = {
-  baseUrl: BLACK_FOREST_LABS_AI_API_BASE_URL,
+  makeBaseUrl,
   makeBody,
   makeHeaders,
   makeUrl
@@ -136,6 +129,9 @@ var BLACK_FOREST_LABS_CONFIG = {
 
 // src/providers/cerebras.ts
 var CEREBRAS_API_BASE_URL = "https://api.cerebras.ai";
+var makeBaseUrl2 = () => {
+  return CEREBRAS_API_BASE_URL;
+};
 var makeBody2 = (params) => {
   return {
     ...params.args,
@@ -149,7 +145,7 @@ var makeUrl2 = (params) => {
   return `${params.baseUrl}/v1/chat/completions`;
 };
 var CEREBRAS_CONFIG = {
-  baseUrl: CEREBRAS_API_BASE_URL,
+  makeBaseUrl: makeBaseUrl2,
   makeBody: makeBody2,
   makeHeaders: makeHeaders2,
   makeUrl: makeUrl2
@@ -157,6 +153,9 @@ var CEREBRAS_CONFIG = {
 
 // src/providers/cohere.ts
 var COHERE_API_BASE_URL = "https://api.cohere.com";
+var makeBaseUrl3 = () => {
+  return COHERE_API_BASE_URL;
+};
 var makeBody3 = (params) => {
   return {
     ...params.args,
@@ -170,14 +169,40 @@ var makeUrl3 = (params) => {
   return `${params.baseUrl}/compatibility/v1/chat/completions`;
 };
 var COHERE_CONFIG = {
-  baseUrl: COHERE_API_BASE_URL,
+  makeBaseUrl: makeBaseUrl3,
   makeBody: makeBody3,
   makeHeaders: makeHeaders3,
   makeUrl: makeUrl3
 };
 
+// src/lib/InferenceOutputError.ts
+var InferenceOutputError = class extends TypeError {
+  constructor(message) {
+    super(
+      `Invalid inference output: ${message}. Use the 'request' method with the same parameters to do a custom call with no type checking.`
+    );
+    this.name = "InferenceOutputError";
+  }
+};
+
+// src/lib/isUrl.ts
+function isUrl(modelOrUrl) {
+  return /^http(s?):/.test(modelOrUrl) || modelOrUrl.startsWith("/");
+}
+
+// src/utils/delay.ts
+function delay(ms) {
+  return new Promise((resolve) => {
+    setTimeout(() => resolve(), ms);
+  });
+}
+
 // src/providers/fal-ai.ts
 var FAL_AI_API_BASE_URL = "https://fal.run";
+var FAL_AI_API_BASE_URL_QUEUE = "https://queue.fal.run";
+var makeBaseUrl4 = (task) => {
+  return task === "text-to-video" ? FAL_AI_API_BASE_URL_QUEUE : FAL_AI_API_BASE_URL;
+};
 var makeBody4 = (params) => {
   return params.args;
 };
@@ -187,17 +212,64 @@ var makeHeaders4 = (params) => {
   };
 };
 var makeUrl4 = (params) => {
-  return `${params.baseUrl}/${params.model}`;
+  const baseUrl = `${params.baseUrl}/${params.model}`;
+  if (params.authMethod !== "provider-key" && params.task === "text-to-video") {
+    return `${baseUrl}?_subdomain=queue`;
+  }
+  return baseUrl;
 };
 var FAL_AI_CONFIG = {
-  baseUrl: FAL_AI_API_BASE_URL,
+  makeBaseUrl: makeBaseUrl4,
   makeBody: makeBody4,
   makeHeaders: makeHeaders4,
   makeUrl: makeUrl4
 };
+async function pollFalResponse(res, url, headers) {
+  const requestId = res.request_id;
+  if (!requestId) {
+    throw new InferenceOutputError("No request ID found in the response");
+  }
+  let status = res.status;
+  const parsedUrl = new URL(url);
+  const baseUrl = `${parsedUrl.protocol}//${parsedUrl.host}${parsedUrl.host === "router.huggingface.co" ? "/fal-ai" : ""}`;
+  const modelId = new URL(res.response_url).pathname;
+  const queryParams = parsedUrl.search;
+  const statusUrl = `${baseUrl}${modelId}/status${queryParams}`;
+  const resultUrl = `${baseUrl}${modelId}${queryParams}`;
+  while (status !== "COMPLETED") {
+    await delay(500);
+    const statusResponse = await fetch(statusUrl, { headers });
+    if (!statusResponse.ok) {
+      throw new InferenceOutputError("Failed to fetch response status from fal-ai API");
+    }
+    try {
+      status = (await statusResponse.json()).status;
+    } catch (error) {
+      throw new InferenceOutputError("Failed to parse status response from fal-ai API");
+    }
+  }
+  const resultResponse = await fetch(resultUrl, { headers });
+  let result;
+  try {
+    result = await resultResponse.json();
+  } catch (error) {
+    throw new InferenceOutputError("Failed to parse result response from fal-ai API");
+  }
+  if (typeof result === "object" && !!result && "video" in result && typeof result.video === "object" && !!result.video && "url" in result.video && typeof result.video.url === "string" && isUrl(result.video.url)) {
+    const urlResponse = await fetch(result.video.url);
+    return await urlResponse.blob();
+  } else {
+    throw new InferenceOutputError(
+      "Expected { video: { url: string } } result format, got instead: " + JSON.stringify(result)
+    );
+  }
+}
 
 // src/providers/fireworks-ai.ts
 var FIREWORKS_AI_API_BASE_URL = "https://api.fireworks.ai";
+var makeBaseUrl5 = () => {
+  return FIREWORKS_AI_API_BASE_URL;
+};
 var makeBody5 = (params) => {
   return {
     ...params.args,
@@ -214,13 +286,16 @@ var makeUrl5 = (params) => {
   return `${params.baseUrl}/inference`;
 };
 var FIREWORKS_AI_CONFIG = {
-  baseUrl: FIREWORKS_AI_API_BASE_URL,
+  makeBaseUrl: makeBaseUrl5,
   makeBody: makeBody5,
   makeHeaders: makeHeaders5,
   makeUrl: makeUrl5
 };
 
 // src/providers/hf-inference.ts
+var makeBaseUrl6 = () => {
+  return `${HF_ROUTER_URL}/hf-inference`;
+};
 var makeBody6 = (params) => {
   return {
     ...params.args,
@@ -240,7 +315,7 @@ var makeUrl6 = (params) => {
   return `${params.baseUrl}/models/${params.model}`;
 };
 var HF_INFERENCE_CONFIG = {
-  baseUrl: `${HF_ROUTER_URL}/hf-inference`,
+  makeBaseUrl: makeBaseUrl6,
   makeBody: makeBody6,
   makeHeaders: makeHeaders6,
   makeUrl: makeUrl6
@@ -248,6 +323,9 @@ var HF_INFERENCE_CONFIG = {
 
 // src/providers/hyperbolic.ts
 var HYPERBOLIC_API_BASE_URL = "https://api.hyperbolic.xyz";
+var makeBaseUrl7 = () => {
+  return HYPERBOLIC_API_BASE_URL;
+};
 var makeBody7 = (params) => {
   return {
     ...params.args,
@@ -264,7 +342,7 @@ var makeUrl7 = (params) => {
   return `${params.baseUrl}/v1/chat/completions`;
 };
 var HYPERBOLIC_CONFIG = {
-  baseUrl: HYPERBOLIC_API_BASE_URL,
+  makeBaseUrl: makeBaseUrl7,
   makeBody: makeBody7,
   makeHeaders: makeHeaders7,
   makeUrl: makeUrl7
@@ -272,6 +350,9 @@ var HYPERBOLIC_CONFIG = {
 
 // src/providers/nebius.ts
 var NEBIUS_API_BASE_URL = "https://api.studio.nebius.ai";
+var makeBaseUrl8 = () => {
+  return NEBIUS_API_BASE_URL;
+};
 var makeBody8 = (params) => {
   return {
     ...params.args,
@@ -294,7 +375,7 @@ var makeUrl8 = (params) => {
   return params.baseUrl;
 };
 var NEBIUS_CONFIG = {
-  baseUrl: NEBIUS_API_BASE_URL,
+  makeBaseUrl: makeBaseUrl8,
   makeBody: makeBody8,
   makeHeaders: makeHeaders8,
   makeUrl: makeUrl8
@@ -302,6 +383,9 @@ var NEBIUS_CONFIG = {
 
 // src/providers/novita.ts
 var NOVITA_API_BASE_URL = "https://api.novita.ai";
+var makeBaseUrl9 = () => {
+  return NOVITA_API_BASE_URL;
+};
 var makeBody9 = (params) => {
   return {
     ...params.args,
@@ -322,7 +406,7 @@ var makeUrl9 = (params) => {
   return params.baseUrl;
 };
 var NOVITA_CONFIG = {
-  baseUrl: NOVITA_API_BASE_URL,
+  makeBaseUrl: makeBaseUrl9,
   makeBody: makeBody9,
   makeHeaders: makeHeaders9,
   makeUrl: makeUrl9
@@ -330,6 +414,9 @@ var NOVITA_CONFIG = {
 
 // src/providers/replicate.ts
 var REPLICATE_API_BASE_URL = "https://api.replicate.com";
+var makeBaseUrl10 = () => {
+  return REPLICATE_API_BASE_URL;
+};
 var makeBody10 = (params) => {
   return {
     input: params.args,
@@ -346,7 +433,7 @@ var makeUrl10 = (params) => {
   return `${params.baseUrl}/v1/models/${params.model}/predictions`;
 };
 var REPLICATE_CONFIG = {
-  baseUrl: REPLICATE_API_BASE_URL,
+  makeBaseUrl: makeBaseUrl10,
   makeBody: makeBody10,
   makeHeaders: makeHeaders10,
   makeUrl: makeUrl10
@@ -354,6 +441,9 @@ var REPLICATE_CONFIG = {
 
 // src/providers/sambanova.ts
 var SAMBANOVA_API_BASE_URL = "https://api.sambanova.ai";
+var makeBaseUrl11 = () => {
+  return SAMBANOVA_API_BASE_URL;
+};
 var makeBody11 = (params) => {
   return {
     ...params.args,
@@ -370,7 +460,7 @@ var makeUrl11 = (params) => {
   return params.baseUrl;
 };
 var SAMBANOVA_CONFIG = {
-  baseUrl: SAMBANOVA_API_BASE_URL,
+  makeBaseUrl: makeBaseUrl11,
   makeBody: makeBody11,
   makeHeaders: makeHeaders11,
   makeUrl: makeUrl11
@@ -378,6 +468,9 @@ var SAMBANOVA_CONFIG = {
 
 // src/providers/together.ts
 var TOGETHER_API_BASE_URL = "https://api.together.xyz";
+var makeBaseUrl12 = () => {
+  return TOGETHER_API_BASE_URL;
+};
 var makeBody12 = (params) => {
   return {
     ...params.args,
@@ -400,7 +493,7 @@ var makeUrl12 = (params) => {
   return params.baseUrl;
 };
 var TOGETHER_CONFIG = {
-  baseUrl: TOGETHER_API_BASE_URL,
+  makeBaseUrl: makeBaseUrl12,
   makeBody: makeBody12,
   makeHeaders: makeHeaders12,
   makeUrl: makeUrl12
@@ -408,6 +501,9 @@ var TOGETHER_CONFIG = {
 
 // src/providers/openai.ts
 var OPENAI_API_BASE_URL = "https://api.openai.com";
+var makeBaseUrl13 = () => {
+  return OPENAI_API_BASE_URL;
+};
 var makeBody13 = (params) => {
   if (!params.chatCompletion) {
     throw new Error("OpenAI only supports chat completions.");
@@ -427,21 +523,16 @@ var makeUrl13 = (params) => {
   return `${params.baseUrl}/v1/chat/completions`;
 };
 var OPENAI_CONFIG = {
-  baseUrl: OPENAI_API_BASE_URL,
+  makeBaseUrl: makeBaseUrl13,
   makeBody: makeBody13,
   makeHeaders: makeHeaders13,
   makeUrl: makeUrl13,
   clientSideRoutingOnly: true
 };
 
-// src/lib/isUrl.ts
-function isUrl(modelOrUrl) {
-  return /^http(s?):/.test(modelOrUrl) || modelOrUrl.startsWith("/");
-}
-
 // package.json
 var name = "@huggingface/inference";
-var version = "3.6.0";
+var version = "3.6.2";
 
 // src/providers/consts.ts
 var HARDCODED_MODEL_ID_MAPPING = {
@@ -580,7 +671,8 @@ function makeRequestOptionsFromResolvedModel(resolvedModel, args, options) {
     return "none";
   })();
   const url = endpointUrl ? chatCompletion2 ? endpointUrl + `/v1/chat/completions` : endpointUrl : providerConfig.makeUrl({
-    baseUrl: authMethod !== "provider-key" ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", provider) : providerConfig.baseUrl,
+    authMethod,
+    baseUrl: authMethod !== "provider-key" ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", provider) : providerConfig.makeBaseUrl(task),
     model: resolvedModel,
     chatCompletion: chatCompletion2,
     task
@@ -844,16 +936,6 @@ async function* streamingRequest(args, options) {
   }
 }
 
-// src/lib/InferenceOutputError.ts
-var InferenceOutputError = class extends TypeError {
-  constructor(message) {
-    super(
-      `Invalid inference output: ${message}. Use the 'request' method with the same parameters to do a custom call with no type checking.`
-    );
-    this.name = "InferenceOutputError";
-  }
-};
-
 // src/utils/pick.ts
 function pick(o, props) {
   return Object.assign(
@@ -1068,13 +1150,6 @@ async function objectDetection(args, options) {
   return res;
 }
 
-// src/utils/delay.ts
-function delay(ms) {
-  return new Promise((resolve) => {
-    setTimeout(() => resolve(), ms);
-  });
-}
-
 // src/tasks/cv/textToImage.ts
 function getResponseFormatArg(provider) {
   switch (provider) {
@@ -1246,12 +1321,8 @@ async function textToVideo(args, options) {
     task: "text-to-video"
   });
   if (args.provider === "fal-ai") {
-    const isValidOutput = typeof res === "object" && !!res && "video" in res && typeof res.video === "object" && !!res.video && "url" in res.video && typeof res.video.url === "string" && isUrl(res.video.url);
-    if (!isValidOutput) {
-      throw new InferenceOutputError("Expected { video: { url: string } }");
-    }
-    const urlResponse = await fetch(res.video.url);
-    return await urlResponse.blob();
+    const { url, info } = await makeRequestOptions(args, { ...options, task: "text-to-video" });
+    return await pollFalResponse(res, url, info.headers);
   } else if (args.provider === "novita") {
     const isValidOutput = typeof res === "object" && !!res && "video" in res && typeof res.video === "object" && !!res.video && "video_url" in res.video && typeof res.video.video_url === "string" && isUrl(res.video.video_url);
     if (!isValidOutput) {
@@ -1673,10 +1744,116 @@ __export(snippets_exports, {
 // src/snippets/getInferenceSnippets.ts
 var import_tasks = require("@huggingface/tasks");
 var import_jinja = require("@huggingface/jinja");
-var import_fs = __toESM(require("fs"), 1);
-var import_path = __toESM(require("path"), 1);
-var import_node_fs = require("fs");
-var import_meta = {};
+
+// src/snippets/templates.exported.ts
+var templates = {
+  "js": {
+    "fetch": {
+      "basic": 'async function query(data) {\n	const response = await fetch(\n		"{{ fullUrl }}",\n		{\n			headers: {\n				Authorization: "{{ authorizationHeader }}",\n				"Content-Type": "application/json",\n			},\n			method: "POST",\n			body: JSON.stringify(data),\n		}\n	);\n	const result = await response.json();\n	return result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});',
+      "basicAudio": 'async function query(data) {\n	const response = await fetch(\n		"{{ fullUrl }}",\n		{\n			headers: {\n				Authorization: "{{ authorizationHeader }}",\n				"Content-Type": "audio/flac"\n			},\n			method: "POST",\n			body: JSON.stringify(data),\n		}\n	);\n	const result = await response.json();\n	return result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});',
+      "basicImage": 'async function query(data) {\n	const response = await fetch(\n		"{{ fullUrl }}",\n		{\n			headers: {\n				Authorization: "{{ authorizationHeader }}",\n				"Content-Type": "image/jpeg"\n			},\n			method: "POST",\n			body: JSON.stringify(data),\n		}\n	);\n	const result = await response.json();\n	return result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});',
+      "textToAudio": '{% if model.library_name == "transformers" %}\nasync function query(data) {\n	const response = await fetch(\n		"{{ fullUrl }}",\n		{\n			headers: {\n				Authorization: "{{ authorizationHeader }}",\n				"Content-Type": "application/json",\n			},\n			method: "POST",\n			body: JSON.stringify(data),\n		}\n	);\n	const result = await response.blob();\n    return result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    // Returns a byte object of the Audio wavform. Use it directly!\n});\n{% else %}\nasync function query(data) {\n	const response = await fetch(\n		"{{ fullUrl }}",\n		{\n			headers: {\n				Authorization: "{{ authorizationHeader }}",\n				"Content-Type": "application/json",\n			},\n			method: "POST",\n			body: JSON.stringify(data),\n		}\n	);\n    const result = await response.json();\n    return result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});\n{% endif %} ',
+      "textToImage": 'async function query(data) {\n	const response = await fetch(\n		"{{ fullUrl }}",\n		{\n			headers: {\n				Authorization: "{{ authorizationHeader }}",\n				"Content-Type": "application/json",\n			},\n			method: "POST",\n			body: JSON.stringify(data),\n		}\n	);\n	const result = await response.blob();\n	return result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    // Use image\n});',
+      "zeroShotClassification": 'async function query(data) {\n    const response = await fetch(\n		"{{ fullUrl }}",\n        {\n            headers: {\n				Authorization: "{{ authorizationHeader }}",\n                "Content-Type": "application/json",\n            },\n            method: "POST",\n            body: JSON.stringify(data),\n        }\n    );\n    const result = await response.json();\n    return result;\n}\n\nquery({\n    inputs: {{ providerInputs.asObj.inputs }},\n    parameters: { candidate_labels: ["refund", "legal", "faq"] }\n}).then((response) => {\n    console.log(JSON.stringify(response));\n});'
+    },
+    "huggingface.js": {
+      "basic": 'import { InferenceClient } from "@huggingface/inference";\n\nconst client = new InferenceClient("{{ accessToken }}");\n\nconst output = await client.{{ methodName }}({\n	model: "{{ model.id }}",\n	inputs: {{ inputs.asObj.inputs }},\n	provider: "{{ provider }}",\n});\n\nconsole.log(output);',
+      "basicAudio": 'import { InferenceClient } from "@huggingface/inference";\n\nconst client = new InferenceClient("{{ accessToken }}");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n	data,\n	model: "{{ model.id }}",\n	provider: "{{ provider }}",\n});\n\nconsole.log(output);',
+      "basicImage": 'import { InferenceClient } from "@huggingface/inference";\n\nconst client = new InferenceClient("{{ accessToken }}");\n\nconst data = fs.readFileSync({{inputs.asObj.inputs}});\n\nconst output = await client.{{ methodName }}({\n	data,\n	model: "{{ model.id }}",\n	provider: "{{ provider }}",\n});\n\nconsole.log(output);',
+      "conversational": 'import { InferenceClient } from "@huggingface/inference";\n\nconst client = new InferenceClient("{{ accessToken }}");\n\nconst chatCompletion = await client.chatCompletion({\n    provider: "{{ provider }}",\n    model: "{{ model.id }}",\n{{ inputs.asTsString }}\n});\n\nconsole.log(chatCompletion.choices[0].message);',
+      "conversationalStream": 'import { InferenceClient } from "@huggingface/inference";\n\nconst client = new InferenceClient("{{ accessToken }}");\n\nlet out = "";\n\nconst stream = await client.chatCompletionStream({\n    provider: "{{ provider }}",\n    model: "{{ model.id }}",\n{{ inputs.asTsString }}\n});\n\nfor await (const chunk of stream) {\n	if (chunk.choices && chunk.choices.length > 0) {\n		const newContent = chunk.choices[0].delta.content;\n		out += newContent;\n		console.log(newContent);\n	}  \n}',
+      "textToImage": `import { InferenceClient } from "@huggingface/inference";
+
+const client = new InferenceClient("{{ accessToken }}");
+
+const image = await client.textToImage({
+    provider: "{{ provider }}",
+    model: "{{ model.id }}",
+	inputs: {{ inputs.asObj.inputs }},
+	parameters: { num_inference_steps: 5 },
+});
+/// Use the generated image (it's a Blob)`,
+      "textToVideo": `import { InferenceClient } from "@huggingface/inference";
+
+const client = new InferenceClient("{{ accessToken }}");
+
+const image = await client.textToVideo({
+    provider: "{{ provider }}",
+    model: "{{ model.id }}",
+	inputs: {{ inputs.asObj.inputs }},
+});
+// Use the generated video (it's a Blob)`
+    },
+    "openai": {
+      "conversational": 'import { OpenAI } from "openai";\n\nconst client = new OpenAI({\n	baseURL: "{{ baseUrl }}",\n	apiKey: "{{ accessToken }}",\n});\n\nconst chatCompletion = await client.chat.completions.create({\n	model: "{{ providerModelId }}",\n{{ inputs.asTsString }}\n});\n\nconsole.log(chatCompletion.choices[0].message);',
+      "conversationalStream": 'import { OpenAI } from "openai";\n\nconst client = new OpenAI({\n	baseURL: "{{ baseUrl }}",\n	apiKey: "{{ accessToken }}",\n});\n\nlet out = "";\n\nconst stream = await client.chat.completions.create({\n    provider: "{{ provider }}",\n    model: "{{ model.id }}",\n{{ inputs.asTsString }}\n});\n\nfor await (const chunk of stream) {\n	if (chunk.choices && chunk.choices.length > 0) {\n		const newContent = chunk.choices[0].delta.content;\n		out += newContent;\n		console.log(newContent);\n	}  \n}'
+    }
+  },
+  "python": {
+    "fal_client": {
+      "textToImage": '{% if provider == "fal-ai" %}\nimport fal_client\n\nresult = fal_client.subscribe(\n    "{{ providerModelId }}",\n    arguments={\n        "prompt": {{ inputs.asObj.inputs }},\n    },\n)\nprint(result)\n{% endif %} '
+    },
+    "huggingface_hub": {
+      "basic": 'result = client.{{ methodName }}(\n    inputs={{ inputs.asObj.inputs }},\n    model="{{ model.id }}",\n)',
+      "basicAudio": 'output = client.{{ methodName }}({{ inputs.asObj.inputs }}, model="{{ model.id }}")',
+      "basicImage": 'output = client.{{ methodName }}({{ inputs.asObj.inputs }}, model="{{ model.id }}")',
+      "conversational": 'completion = client.chat.completions.create(\n    model="{{ model.id }}",\n{{ inputs.asPythonString }}\n)\n\nprint(completion.choices[0].message) ',
+      "conversationalStream": 'stream = client.chat.completions.create(\n    model="{{ model.id }}",\n{{ inputs.asPythonString }}\n    stream=True,\n)\n\nfor chunk in stream:\n    print(chunk.choices[0].delta.content, end="") ',
+      "documentQuestionAnswering": 'output = client.document_question_answering(\n    "{{ inputs.asObj.image }}",\n    question="{{ inputs.asObj.question }}",\n    model="{{ model.id }}",\n) ',
+      "imageToImage": '# output is a PIL.Image object\nimage = client.image_to_image(\n    "{{ inputs.asObj.inputs }}",\n    prompt="{{ inputs.asObj.parameters.prompt }}",\n    model="{{ model.id }}",\n) ',
+      "importInferenceClient": 'from huggingface_hub import InferenceClient\n\nclient = InferenceClient(\n    provider="{{ provider }}",\n    api_key="{{ accessToken }}",\n)',
+      "textToImage": '# output is a PIL.Image object\nimage = client.text_to_image(\n    {{ inputs.asObj.inputs }},\n    model="{{ model.id }}",\n) ',
+      "textToVideo": 'video = client.text_to_video(\n    {{ inputs.asObj.inputs }},\n    model="{{ model.id }}",\n) '
+    },
+    "openai": {
+      "conversational": 'from openai import OpenAI\n\nclient = OpenAI(\n    base_url="{{ baseUrl }}",\n    api_key="{{ accessToken }}"\n)\n\ncompletion = client.chat.completions.create(\n    model="{{ providerModelId }}",\n{{ inputs.asPythonString }}\n)\n\nprint(completion.choices[0].message) ',
+      "conversationalStream": 'from openai import OpenAI\n\nclient = OpenAI(\n    base_url="{{ baseUrl }}",\n    api_key="{{ accessToken }}"\n)\n\nstream = client.chat.completions.create(\n    model="{{ providerModelId }}",\n{{ inputs.asPythonString }}\n    stream=True,\n)\n\nfor chunk in stream:\n    print(chunk.choices[0].delta.content, end="")'
+    },
+    "requests": {
+      "basic": 'def query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.json()\n\noutput = query({\n    "inputs": {{ providerInputs.asObj.inputs }},\n}) ',
+      "basicAudio": 'def query(filename):\n    with open(filename, "rb") as f:\n        data = f.read()\n    response = requests.post(API_URL, headers={"Content-Type": "audio/flac", **headers}, data=data)\n    return response.json()\n\noutput = query({{ providerInputs.asObj.inputs }})',
+      "basicImage": 'def query(filename):\n    with open(filename, "rb") as f:\n        data = f.read()\n    response = requests.post(API_URL, headers={"Content-Type": "image/jpeg", **headers}, data=data)\n    return response.json()\n\noutput = query({{ providerInputs.asObj.inputs }})',
+      "conversational": 'def query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.json()\n\nresponse = query({\n{{ providerInputs.asJsonString }}\n})\n\nprint(response["choices"][0]["message"])',
+      "conversationalStream": 'def query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload, stream=True)\n    for line in response.iter_lines():\n        if not line.startswith(b"data:"):\n            continue\n        if line.strip() == b"data: [DONE]":\n            return\n        yield json.loads(line.decode("utf-8").lstrip("data:").rstrip("/n"))\n\nchunks = query({\n{{ providerInputs.asJsonString }},\n    "stream": True,\n})\n\nfor chunk in chunks:\n    print(chunk["choices"][0]["delta"]["content"], end="")',
+      "documentQuestionAnswering": 'def query(payload):\n    with open(payload["image"], "rb") as f:\n        img = f.read()\n        payload["image"] = base64.b64encode(img).decode("utf-8")\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.json()\n\noutput = query({\n    "inputs": {\n        "image": "{{ inputs.asObj.image }}",\n        "question": "{{ inputs.asObj.question }}",\n    },\n}) ',
+      "imageToImage": 'def query(payload):\n    with open(payload["inputs"], "rb") as f:\n        img = f.read()\n        payload["inputs"] = base64.b64encode(img).decode("utf-8")\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.content\n\nimage_bytes = query({\n{{ providerInputs.asJsonString }}\n})\n\n# You can access the image with PIL.Image for example\nimport io\nfrom PIL import Image\nimage = Image.open(io.BytesIO(image_bytes)) ',
+      "importRequests": '{% if importBase64 %}\nimport base64\n{% endif %}\n{% if importJson %}\nimport json\n{% endif %}\nimport requests\n\nAPI_URL = "{{ fullUrl }}"\nheaders = {"Authorization": "{{ authorizationHeader }}"}',
+      "tabular": 'def query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.content\n\nresponse = query({\n    "inputs": {\n        "data": {{ providerInputs.asObj.inputs }}\n    },\n}) ',
+      "textToAudio": '{% if model.library_name == "transformers" %}\ndef query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.content\n\naudio_bytes = query({\n    "inputs": {{ providerInputs.asObj.inputs }},\n})\n# You can access the audio with IPython.display for example\nfrom IPython.display import Audio\nAudio(audio_bytes)\n{% else %}\ndef query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.json()\n\naudio, sampling_rate = query({\n    "inputs": {{ providerInputs.asObj.inputs }},\n})\n# You can access the audio with IPython.display for example\nfrom IPython.display import Audio\nAudio(audio, rate=sampling_rate)\n{% endif %} ',
+      "textToImage": '{% if provider == "hf-inference" %}\ndef query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.content\n\nimage_bytes = query({\n    "inputs": {{ providerInputs.asObj.inputs }},\n})\n\n# You can access the image with PIL.Image for example\nimport io\nfrom PIL import Image\nimage = Image.open(io.BytesIO(image_bytes))\n{% endif %}',
+      "zeroShotClassification": 'def query(payload):\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.json()\n\noutput = query({\n    "inputs": {{ providerInputs.asObj.inputs }},\n    "parameters": {"candidate_labels": ["refund", "legal", "faq"]},\n}) ',
+      "zeroShotImageClassification": 'def query(data):\n    with open(data["image_path"], "rb") as f:\n        img = f.read()\n    payload={\n        "parameters": data["parameters"],\n        "inputs": base64.b64encode(img).decode("utf-8")\n    }\n    response = requests.post(API_URL, headers=headers, json=payload)\n    return response.json()\n\noutput = query({\n    "image_path": {{ providerInputs.asObj.inputs }},\n    "parameters": {"candidate_labels": ["cat", "dog", "llama"]},\n}) '
+    }
+  },
+  "sh": {
+    "curl": {
+      "basic": "curl {{ fullUrl }} \\\n    -X POST \\\n    -H 'Authorization: {{ authorizationHeader }}' \\\n    -H 'Content-Type: application/json' \\\n    -d '{\n{{ providerInputs.asCurlString }}\n    }'",
+      "basicAudio": "curl {{ fullUrl }} \\\n    -X POST \\\n    -H 'Authorization: {{ authorizationHeader }}' \\\n    -H 'Content-Type: audio/flac' \\\n    --data-binary @{{ providerInputs.asObj.inputs }}",
+      "basicImage": "curl {{ fullUrl }} \\\n    -X POST \\\n    -H 'Authorization: {{ authorizationHeader }}' \\\n    -H 'Content-Type: image/jpeg' \\\n    --data-binary @{{ providerInputs.asObj.inputs }}",
+      "conversational": `curl {{ fullUrl }} \\
+    -H 'Authorization: {{ authorizationHeader }}' \\
+    -H 'Content-Type: application/json' \\
+    -d '{
+{{ providerInputs.asCurlString }},
+        "stream": false
+    }'`,
+      "conversationalStream": `curl {{ fullUrl }} \\
+    -H 'Authorization: {{ authorizationHeader }}' \\
+    -H 'Content-Type: application/json' \\
+    -d '{
+{{ providerInputs.asCurlString }},
+        "stream": true
+    }'`,
+      "zeroShotClassification": `curl {{ fullUrl }} \\
+    -X POST \\
+    -d '{"inputs": {{ providerInputs.asObj.inputs }}, "parameters": {"candidate_labels": ["refund", "legal", "faq"]}}' \\
+    -H 'Content-Type: application/json' \\
+    -H 'Authorization: {{ authorizationHeader }}'`
+    }
+  }
+};
+
+// src/snippets/getInferenceSnippets.ts
 var PYTHON_CLIENTS = ["huggingface_hub", "fal_client", "requests", "openai"];
 var JS_CLIENTS = ["fetch", "huggingface.js", "openai"];
 var SH_CLIENTS = ["curl"];
@@ -1685,20 +1862,12 @@ var CLIENTS = {
   python: [...PYTHON_CLIENTS],
   sh: [...SH_CLIENTS]
 };
-var rootDirFinder = () => {
-  let currentPath = typeof import_meta !== "undefined" && import_meta.url ? import_path.default.normalize(new URL(import_meta.url).pathname) : __dirname;
-  while (currentPath !== "/") {
-    if ((0, import_node_fs.existsSync)(import_path.default.join(currentPath, "package.json"))) {
-      return currentPath;
-    }
-    currentPath = import_path.default.normalize(import_path.default.join(currentPath, ".."));
-  }
-  return "/";
-};
-var templatePath = (language, client, templateName) => import_path.default.join(rootDirFinder(), "src", "snippets", "templates", language, client, `${templateName}.jinja`);
-var hasTemplate = (language, client, templateName) => (0, import_node_fs.existsSync)(templatePath(language, client, templateName));
+var hasTemplate = (language, client, templateName) => templates[language]?.[client]?.[templateName] !== void 0;
 var loadTemplate = (language, client, templateName) => {
-  const template = import_fs.default.readFileSync(templatePath(language, client, templateName), "utf8");
+  const template = templates[language]?.[client]?.[templateName];
+  if (!template) {
+    throw new Error(`Template not found: ${language}/${client}/${templateName}`);
+  }
   return (data) => new import_jinja.Template(template).render({ ...data });
 };
 var snippetImportPythonInferenceClient = loadTemplate("python", "huggingface_hub", "importInferenceClient");