@huggingface/inference 3.6.0 → 3.6.2

package/src/snippets/templates.exported.ts

@@ -0,0 +1,72 @@
+// Generated file - do not edit directly
+export const templates: Record<string, Record<string, Record<string, string>>> = {
+  "js": {
+    "fetch": {
+      "basic": "async function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.json();\n\treturn result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});",
+      "basicAudio": "async function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"audio/flac\"\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.json();\n\treturn result;\n}\n\nquery({ inputs: {{ providerInputs.asObj.inputs }} }).then((response) => {\n    console.log(JSON.stringify(response));\n});",
+      "basicImage": "async function query(data) {\n\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"image/jpeg\"\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.json();\n\treturn 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\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst 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\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\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\tconst response = await fetch(\n\t\t\"{{ fullUrl }}\",\n\t\t{\n\t\t\theaders: {\n\t\t\t\tAuthorization: \"{{ authorizationHeader }}\",\n\t\t\t\t\"Content-Type\": \"application/json\",\n\t\t\t},\n\t\t\tmethod: \"POST\",\n\t\t\tbody: JSON.stringify(data),\n\t\t}\n\t);\n\tconst result = await response.blob();\n\treturn 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\t\t\"{{ fullUrl }}\",\n        {\n            headers: {\n\t\t\t\tAuthorization: \"{{ 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\tmodel: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tprovider: \"{{ 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\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ 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\tdata,\n\tmodel: \"{{ model.id }}\",\n\tprovider: \"{{ 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\tif (chunk.choices && chunk.choices.length > 0) {\n\t\tconst newContent = chunk.choices[0].delta.content;\n\t\tout += newContent;\n\t\tconsole.log(newContent);\n\t}  \n}",
+      "textToImage": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst image = await client.textToImage({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n\tparameters: { num_inference_steps: 5 },\n});\n/// Use the generated image (it's a Blob)",
+      "textToVideo": "import { InferenceClient } from \"@huggingface/inference\";\n\nconst client = new InferenceClient(\"{{ accessToken }}\");\n\nconst image = await client.textToVideo({\n    provider: \"{{ provider }}\",\n    model: \"{{ model.id }}\",\n\tinputs: {{ inputs.asObj.inputs }},\n});\n// Use the generated video (it's a Blob)"
+    },
+    "openai": {
+      "conversational": "import { OpenAI } from \"openai\";\n\nconst client = new OpenAI({\n\tbaseURL: \"{{ baseUrl }}\",\n\tapiKey: \"{{ accessToken }}\",\n});\n\nconst chatCompletion = await client.chat.completions.create({\n\tmodel: \"{{ providerModelId }}\",\n{{ inputs.asTsString }}\n});\n\nconsole.log(chatCompletion.choices[0].message);",
+      "conversationalStream": "import { OpenAI } from \"openai\";\n\nconst client = new OpenAI({\n\tbaseURL: \"{{ baseUrl }}\",\n\tapiKey: \"{{ 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\tif (chunk.choices && chunk.choices.length > 0) {\n\t\tconst newContent = chunk.choices[0].delta.content;\n\t\tout += newContent;\n\t\tconsole.log(newContent);\n\t}  \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 }} \\\n    -H 'Authorization: {{ authorizationHeader }}' \\\n    -H 'Content-Type: application/json' \\\n    -d '{\n{{ providerInputs.asCurlString }},\n        \"stream\": false\n    }'",
+      "conversationalStream": "curl {{ fullUrl }} \\\n    -H 'Authorization: {{ authorizationHeader }}' \\\n    -H 'Content-Type: application/json' \\\n    -d '{\n{{ providerInputs.asCurlString }},\n        \"stream\": true\n    }'",
+      "zeroShotClassification": "curl {{ fullUrl }} \\\n    -X POST \\\n    -d '{\"inputs\": {{ providerInputs.asObj.inputs }}, \"parameters\": {\"candidate_labels\": [\"refund\", \"legal\", \"faq\"]}}' \\\n    -H 'Content-Type: application/json' \\\n    -H 'Authorization: {{ authorizationHeader }}'"
+    }
+  }
+} as const;

package/src/tasks/cv/textToVideo.ts

@@ -5,17 +5,13 @@ import { omit } from "../../utils/omit";
 import { isUrl } from "../../lib/isUrl";
 import { InferenceOutputError } from "../../lib/InferenceOutputError";
 import { typedInclude } from "../../utils/typedInclude";
+import { makeRequestOptions } from "../../lib/makeRequestOptions";
+import { pollFalResponse, type FalAiQueueOutput } from "../../providers/fal-ai";
 
 export type TextToVideoArgs = BaseArgs & TextToVideoInput;
 
 export type TextToVideoOutput = Blob;
 
-interface FalAiOutput {
-	video: {
-		url: string;
-	};
-}
-
 interface ReplicateOutput {
 	output: string;
 }
@@ -39,25 +35,13 @@ export async function textToVideo(args: TextToVideoArgs, options?: Options): Pro
 		args.provider === "fal-ai" || args.provider === "replicate" || args.provider === "novita"
 			? { ...omit(args, ["inputs", "parameters"]), ...args.parameters, prompt: args.inputs }
 			: args;
-	const res = await request<FalAiOutput | ReplicateOutput | NovitaOutput>(payload, {
+	const res = await request<FalAiQueueOutput | ReplicateOutput | NovitaOutput>(payload, {
 		...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 as FalAiOutput).video.url);
-		return await urlResponse.blob();
+		const { url, info } = await makeRequestOptions(args, { ...options, task: "text-to-video" });
+		return await pollFalResponse(res as FalAiQueueOutput, url, info.headers as Record<string, string>);
 	} else if (args.provider === "novita") {
 		const isValidOutput =
 			typeof res === "object" &&

package/src/types.ts

@@ -94,19 +94,22 @@ export type RequestArgs = BaseArgs &
 	};
 
 export interface ProviderConfig {
-	baseUrl: string;
+	makeBaseUrl: ((task?: InferenceTask) => string) | (() => string);
 	makeBody: (params: BodyParams) => Record<string, unknown>;
 	makeHeaders: (params: HeaderParams) => Record<string, string>;
 	makeUrl: (params: UrlParams) => string;
 	clientSideRoutingOnly?: boolean;
 }
 
+export type AuthMethod = "none" | "hf-token" | "credentials-include" | "provider-key";
+
 export interface HeaderParams {
 	accessToken?: string;
-	authMethod: "none" | "hf-token" | "credentials-include" | "provider-key";
+	authMethod: AuthMethod;
 }
 
 export interface UrlParams {
+	authMethod: AuthMethod;
 	baseUrl: string;
 	model: string;
 	task?: InferenceTask;