@huggingface/inference 3.3.0 → 3.3.1

package/README.md

@@ -49,6 +49,7 @@ You can send inference requests to third-party providers with the inference clie
 Currently, we support the following providers:
 - [Fal.ai](https://fal.ai)
 - [Fireworks AI](https://fireworks.ai)
+- [Nebius](https://studio.nebius.ai)
 - [Replicate](https://replicate.com)
 - [Sambanova](https://sambanova.ai)
 - [Together](https://together.xyz)
@@ -71,12 +72,13 @@ When authenticated with a third-party provider key, the request is made directly
 Only a subset of models are supported when requesting third-party providers. You can check the list of supported models per pipeline tasks here:
 - [Fal.ai supported models](https://huggingface.co/api/partners/fal-ai/models)
 - [Fireworks AI supported models](https://huggingface.co/api/partners/fireworks-ai/models)
+- [Nebius supported models](https://huggingface.co/api/partners/nebius/models)
 - [Replicate supported models](https://huggingface.co/api/partners/replicate/models)
 - [Sambanova supported models](https://huggingface.co/api/partners/sambanova/models)
 - [Together supported models](https://huggingface.co/api/partners/together/models)
 - [HF Inference API (serverless)](https://huggingface.co/models?inference=warm&sort=trending)
 
-❗**Important note:** To be compatible, the third-party API must adhere to the "standard" shape API we expect on HF model pages for each pipeline task type. 
+❗**Important note:** To be compatible, the third-party API must adhere to the "standard" shape API we expect on HF model pages for each pipeline task type.
 This is not an issue for LLMs as everyone converged on the OpenAI API anyways, but can be more tricky for other tasks like "text-to-image" or "automatic-speech-recognition" where there exists no standard API. Let us know if any help is needed or if we can make things easier for you!
 
 👋**Want to add another provider?** Get in touch if you'd like to add support for another Inference provider, and/or request it on https://huggingface.co/spaces/huggingface/HuggingDiscussions/discussions/49
@@ -463,7 +465,7 @@ await hf.zeroShotImageClassification({
   model: 'openai/clip-vit-large-patch14-336',
   inputs: {
     image: await (await fetch('https://placekitten.com/300/300')).blob()
-  },  
+  },
   parameters: {
     candidate_labels: ['cat', 'dog']
   }

package/dist/index.cjs

@@ -103,6 +103,9 @@ var HF_ROUTER_URL = "https://router.huggingface.co";
 // src/providers/fal-ai.ts
 var FAL_AI_API_BASE_URL = "https://fal.run";
 
+// src/providers/nebius.ts
+var NEBIUS_API_BASE_URL = "https://api.studio.nebius.ai";
+
 // src/providers/replicate.ts
 var REPLICATE_API_BASE_URL = "https://api.replicate.com";
 
@@ -122,7 +125,7 @@ function isUrl(modelOrUrl) {
 
 // package.json
 var name = "@huggingface/inference";
-var version = "3.3.0";
+var version = "3.3.1";
 
 // src/providers/consts.ts
 var HARDCODED_MODEL_ID_MAPPING = {
@@ -135,6 +138,7 @@ var HARDCODED_MODEL_ID_MAPPING = {
   "fal-ai": {},
   "fireworks-ai": {},
   "hf-inference": {},
+  nebius: {},
   replicate: {},
   sambanova: {},
   together: {}
@@ -191,7 +195,7 @@ async function makeRequestOptions(args, options) {
   const { accessToken, endpointUrl, provider: maybeProvider, model: maybeModel, ...remainingArgs } = args;
   let otherArgs = remainingArgs;
   const provider = maybeProvider ?? "hf-inference";
-  const { forceTask, includeCredentials, taskHint, wait_for_model, use_cache, dont_load_model, chatCompletion: chatCompletion2 } = options ?? {};
+  const { forceTask, includeCredentials, taskHint, chatCompletion: chatCompletion2 } = options ?? {};
   if (endpointUrl && provider !== "hf-inference") {
     throw new Error(`Cannot use endpointUrl with a third-party provider.`);
   }
@@ -229,17 +233,6 @@ async function makeRequestOptions(args, options) {
   if (!binary) {
     headers["Content-Type"] = "application/json";
   }
-  if (provider === "hf-inference") {
-    if (wait_for_model) {
-      headers["X-Wait-For-Model"] = "true";
-    }
-    if (use_cache === false) {
-      headers["X-Use-Cache"] = "false";
-    }
-    if (dont_load_model) {
-      headers["X-Load-Model"] = "0";
-    }
-  }
   if (provider === "replicate") {
     headers["Prefer"] = "wait";
   }
@@ -258,7 +251,7 @@ async function makeRequestOptions(args, options) {
     method: "POST",
     body: binary ? args.data : JSON.stringify({
       ...otherArgs,
-      ...chatCompletion2 || provider === "together" ? { model } : void 0
+      ...chatCompletion2 || provider === "together" || provider === "nebius" ? { model } : void 0
     }),
     ...credentials ? { credentials } : void 0,
     signal: options?.signal
@@ -275,6 +268,19 @@ function makeUrl(params) {
       const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : FAL_AI_API_BASE_URL;
       return `${baseUrl}/${params.model}`;
     }
+    case "nebius": {
+      const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : NEBIUS_API_BASE_URL;
+      if (params.taskHint === "text-to-image") {
+        return `${baseUrl}/v1/images/generations`;
+      }
+      if (params.taskHint === "text-generation") {
+        if (params.chatCompletion) {
+          return `${baseUrl}/v1/chat/completions`;
+        }
+        return `${baseUrl}/v1/completions`;
+      }
+      return baseUrl;
+    }
     case "replicate": {
       const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : REPLICATE_API_BASE_URL;
       if (params.model.includes(":")) {
@@ -341,11 +347,8 @@ async function loadTaskInfo() {
 async function request(args, options) {
   const { url, info } = await makeRequestOptions(args, options);
   const response = await (options?.fetch ?? fetch)(url, info);
-  if (options?.retry_on_error !== false && response.status === 503 && !options?.wait_for_model) {
-    return request(args, {
-      ...options,
-      wait_for_model: true
-    });
+  if (options?.retry_on_error !== false && response.status === 503) {
+    return request(args, options);
   }
   if (!response.ok) {
     const contentType = response.headers.get("Content-Type");
@@ -474,11 +477,8 @@ function newMessage() {
 async function* streamingRequest(args, options) {
   const { url, info } = await makeRequestOptions({ ...args, stream: true }, options);
   const response = await (options?.fetch ?? fetch)(url, info);
-  if (options?.retry_on_error !== false && response.status === 503 && !options?.wait_for_model) {
-    return yield* streamingRequest(args, {
-      ...options,
-      wait_for_model: true
-    });
+  if (options?.retry_on_error !== false && response.status === 503) {
+    return yield* streamingRequest(args, options);
   }
   if (!response.ok) {
     if (response.headers.get("Content-Type")?.startsWith("application/json")) {
@@ -769,13 +769,27 @@ async function objectDetection(args, options) {
 }
 
 // src/tasks/cv/textToImage.ts
+function getResponseFormatArg(provider) {
+  switch (provider) {
+    case "fal-ai":
+      return { sync_mode: true };
+    case "nebius":
+      return { response_format: "b64_json" };
+    case "replicate":
+      return void 0;
+    case "together":
+      return { response_format: "base64" };
+    default:
+      return void 0;
+  }
+}
 async function textToImage(args, options) {
-  const payload = args.provider === "together" || args.provider === "fal-ai" || args.provider === "replicate" ? {
+  const payload = !args.provider || args.provider === "hf-inference" || args.provider === "sambanova" ? args : {
     ...omit(args, ["inputs", "parameters"]),
     ...args.parameters,
-    ...args.provider !== "replicate" ? { response_format: "base64" } : void 0,
+    ...getResponseFormatArg(args.provider),
     prompt: args.inputs
-  } : args;
+  };
   const res = await request(payload, {
     ...options,
     taskHint: "text-to-image"
@@ -1155,8 +1169,8 @@ async function chatCompletion(args, options) {
     taskHint: "text-generation",
     chatCompletion: true
   });
-  const isValidOutput = typeof res === "object" && Array.isArray(res?.choices) && typeof res?.created === "number" && typeof res?.id === "string" && typeof res?.model === "string" && /// Together.ai does not output a system_fingerprint
-  (res.system_fingerprint === void 0 || typeof res.system_fingerprint === "string") && typeof res?.usage === "object";
+  const isValidOutput = typeof res === "object" && Array.isArray(res?.choices) && typeof res?.created === "number" && typeof res?.id === "string" && typeof res?.model === "string" && /// Together.ai and Nebius do not output a system_fingerprint
+  (res.system_fingerprint === void 0 || res.system_fingerprint === null || typeof res.system_fingerprint === "string") && typeof res?.usage === "object";
   if (!isValidOutput) {
     throw new InferenceOutputError("Expected ChatCompletionOutput");
   }
@@ -1290,6 +1304,7 @@ var HfInferenceEndpoint = class {
 var INFERENCE_PROVIDERS = [
   "fal-ai",
   "fireworks-ai",
+  "nebius",
   "hf-inference",
   "replicate",
   "sambanova",

package/dist/index.js

@@ -48,6 +48,9 @@ var HF_ROUTER_URL = "https://router.huggingface.co";
 // src/providers/fal-ai.ts
 var FAL_AI_API_BASE_URL = "https://fal.run";
 
+// src/providers/nebius.ts
+var NEBIUS_API_BASE_URL = "https://api.studio.nebius.ai";
+
 // src/providers/replicate.ts
 var REPLICATE_API_BASE_URL = "https://api.replicate.com";
 
@@ -67,7 +70,7 @@ function isUrl(modelOrUrl) {
 
 // package.json
 var name = "@huggingface/inference";
-var version = "3.3.0";
+var version = "3.3.1";
 
 // src/providers/consts.ts
 var HARDCODED_MODEL_ID_MAPPING = {
@@ -80,6 +83,7 @@ var HARDCODED_MODEL_ID_MAPPING = {
   "fal-ai": {},
   "fireworks-ai": {},
   "hf-inference": {},
+  nebius: {},
   replicate: {},
   sambanova: {},
   together: {}
@@ -136,7 +140,7 @@ async function makeRequestOptions(args, options) {
   const { accessToken, endpointUrl, provider: maybeProvider, model: maybeModel, ...remainingArgs } = args;
   let otherArgs = remainingArgs;
   const provider = maybeProvider ?? "hf-inference";
-  const { forceTask, includeCredentials, taskHint, wait_for_model, use_cache, dont_load_model, chatCompletion: chatCompletion2 } = options ?? {};
+  const { forceTask, includeCredentials, taskHint, chatCompletion: chatCompletion2 } = options ?? {};
   if (endpointUrl && provider !== "hf-inference") {
     throw new Error(`Cannot use endpointUrl with a third-party provider.`);
   }
@@ -174,17 +178,6 @@ async function makeRequestOptions(args, options) {
   if (!binary) {
     headers["Content-Type"] = "application/json";
   }
-  if (provider === "hf-inference") {
-    if (wait_for_model) {
-      headers["X-Wait-For-Model"] = "true";
-    }
-    if (use_cache === false) {
-      headers["X-Use-Cache"] = "false";
-    }
-    if (dont_load_model) {
-      headers["X-Load-Model"] = "0";
-    }
-  }
   if (provider === "replicate") {
     headers["Prefer"] = "wait";
   }
@@ -203,7 +196,7 @@ async function makeRequestOptions(args, options) {
     method: "POST",
     body: binary ? args.data : JSON.stringify({
       ...otherArgs,
-      ...chatCompletion2 || provider === "together" ? { model } : void 0
+      ...chatCompletion2 || provider === "together" || provider === "nebius" ? { model } : void 0
     }),
     ...credentials ? { credentials } : void 0,
     signal: options?.signal
@@ -220,6 +213,19 @@ function makeUrl(params) {
       const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : FAL_AI_API_BASE_URL;
       return `${baseUrl}/${params.model}`;
     }
+    case "nebius": {
+      const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : NEBIUS_API_BASE_URL;
+      if (params.taskHint === "text-to-image") {
+        return `${baseUrl}/v1/images/generations`;
+      }
+      if (params.taskHint === "text-generation") {
+        if (params.chatCompletion) {
+          return `${baseUrl}/v1/chat/completions`;
+        }
+        return `${baseUrl}/v1/completions`;
+      }
+      return baseUrl;
+    }
     case "replicate": {
       const baseUrl = shouldProxy ? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider) : REPLICATE_API_BASE_URL;
       if (params.model.includes(":")) {
@@ -286,11 +292,8 @@ async function loadTaskInfo() {
 async function request(args, options) {
   const { url, info } = await makeRequestOptions(args, options);
   const response = await (options?.fetch ?? fetch)(url, info);
-  if (options?.retry_on_error !== false && response.status === 503 && !options?.wait_for_model) {
-    return request(args, {
-      ...options,
-      wait_for_model: true
-    });
+  if (options?.retry_on_error !== false && response.status === 503) {
+    return request(args, options);
   }
   if (!response.ok) {
     const contentType = response.headers.get("Content-Type");
@@ -419,11 +422,8 @@ function newMessage() {
 async function* streamingRequest(args, options) {
   const { url, info } = await makeRequestOptions({ ...args, stream: true }, options);
   const response = await (options?.fetch ?? fetch)(url, info);
-  if (options?.retry_on_error !== false && response.status === 503 && !options?.wait_for_model) {
-    return yield* streamingRequest(args, {
-      ...options,
-      wait_for_model: true
-    });
+  if (options?.retry_on_error !== false && response.status === 503) {
+    return yield* streamingRequest(args, options);
   }
   if (!response.ok) {
     if (response.headers.get("Content-Type")?.startsWith("application/json")) {
@@ -714,13 +714,27 @@ async function objectDetection(args, options) {
 }
 
 // src/tasks/cv/textToImage.ts
+function getResponseFormatArg(provider) {
+  switch (provider) {
+    case "fal-ai":
+      return { sync_mode: true };
+    case "nebius":
+      return { response_format: "b64_json" };
+    case "replicate":
+      return void 0;
+    case "together":
+      return { response_format: "base64" };
+    default:
+      return void 0;
+  }
+}
 async function textToImage(args, options) {
-  const payload = args.provider === "together" || args.provider === "fal-ai" || args.provider === "replicate" ? {
+  const payload = !args.provider || args.provider === "hf-inference" || args.provider === "sambanova" ? args : {
     ...omit(args, ["inputs", "parameters"]),
     ...args.parameters,
-    ...args.provider !== "replicate" ? { response_format: "base64" } : void 0,
+    ...getResponseFormatArg(args.provider),
     prompt: args.inputs
-  } : args;
+  };
   const res = await request(payload, {
     ...options,
     taskHint: "text-to-image"
@@ -1100,8 +1114,8 @@ async function chatCompletion(args, options) {
     taskHint: "text-generation",
     chatCompletion: true
   });
-  const isValidOutput = typeof res === "object" && Array.isArray(res?.choices) && typeof res?.created === "number" && typeof res?.id === "string" && typeof res?.model === "string" && /// Together.ai does not output a system_fingerprint
-  (res.system_fingerprint === void 0 || typeof res.system_fingerprint === "string") && typeof res?.usage === "object";
+  const isValidOutput = typeof res === "object" && Array.isArray(res?.choices) && typeof res?.created === "number" && typeof res?.id === "string" && typeof res?.model === "string" && /// Together.ai and Nebius do not output a system_fingerprint
+  (res.system_fingerprint === void 0 || res.system_fingerprint === null || typeof res.system_fingerprint === "string") && typeof res?.usage === "object";
   if (!isValidOutput) {
     throw new InferenceOutputError("Expected ChatCompletionOutput");
   }
@@ -1235,6 +1249,7 @@ var HfInferenceEndpoint = class {
 var INFERENCE_PROVIDERS = [
   "fal-ai",
   "fireworks-ai",
+  "nebius",
   "hf-inference",
   "replicate",
   "sambanova",

package/dist/src/lib/makeRequestOptions.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"makeRequestOptions.d.ts","sourceRoot":"","sources":["../../../src/lib/makeRequestOptions.ts"],"names":[],"mappings":"AAOA,OAAO,KAAK,EAAE,aAAa,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,UAAU,CAAC;AAapE;;GAEG;AACH,wBAAsB,kBAAkB,CACvC,IAAI,EAAE,WAAW,GAAG;IACnB,IAAI,CAAC,EAAE,IAAI,GAAG,WAAW,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,CAAC;CACjB,EACD,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,yFAAyF;IACzF,SAAS,CAAC,EAAE,MAAM,GAAG,aAAa,CAAC;IACnC,sCAAsC;IACtC,QAAQ,CAAC,EAAE,aAAa,CAAC;IACzB,cAAc,CAAC,EAAE,OAAO,CAAC;CACzB,GACC,OAAO,CAAC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CAAC,CAqH7C"}
+{"version":3,"file":"makeRequestOptions.d.ts","sourceRoot":"","sources":["../../../src/lib/makeRequestOptions.ts"],"names":[],"mappings":"AAQA,OAAO,KAAK,EAAE,aAAa,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,UAAU,CAAC;AAapE;;GAEG;AACH,wBAAsB,kBAAkB,CACvC,IAAI,EAAE,WAAW,GAAG;IACnB,IAAI,CAAC,EAAE,IAAI,GAAG,WAAW,CAAC;IAC1B,MAAM,CAAC,EAAE,OAAO,CAAC;CACjB,EACD,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,yFAAyF;IACzF,SAAS,CAAC,EAAE,MAAM,GAAG,aAAa,CAAC;IACnC,sCAAsC;IACtC,QAAQ,CAAC,EAAE,aAAa,CAAC;IACzB,cAAc,CAAC,EAAE,OAAO,CAAC;CACzB,GACC,OAAO,CAAC;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,WAAW,CAAA;CAAE,CAAC,CAwG7C"}

package/dist/src/providers/consts.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"consts.d.ts","sourceRoot":"","sources":["../../../src/providers/consts.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,UAAU,CAAC;AAClD,OAAO,EAAE,KAAK,OAAO,EAAE,MAAM,UAAU,CAAC;AAExC,KAAK,UAAU,GAAG,MAAM,CAAC;AACzB;;;;;;GAMG;AACH,eAAO,MAAM,0BAA0B,EAAE,MAAM,CAAC,iBAAiB,EAAE,MAAM,CAAC,OAAO,EAAE,UAAU,CAAC,CAa7F,CAAC"}
+{"version":3,"file":"consts.d.ts","sourceRoot":"","sources":["../../../src/providers/consts.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,UAAU,CAAC;AAClD,OAAO,EAAE,KAAK,OAAO,EAAE,MAAM,UAAU,CAAC;AAExC,KAAK,UAAU,GAAG,MAAM,CAAC;AACzB;;;;;;GAMG;AACH,eAAO,MAAM,0BAA0B,EAAE,MAAM,CAAC,iBAAiB,EAAE,MAAM,CAAC,OAAO,EAAE,UAAU,CAAC,CAc7F,CAAC"}

package/dist/src/providers/nebius.d.ts

@@ -0,0 +1,18 @@
+export declare const NEBIUS_API_BASE_URL = "https://api.studio.nebius.ai";
+/**
+ * See the registered mapping of HF model ID => Nebius model ID here:
+ *
+ * https://huggingface.co/api/partners/nebius/models
+ *
+ * This is a publicly available mapping.
+ *
+ * If you want to try to run inference for a new model locally before it's registered on huggingface.co,
+ * you can add it to the dictionary "HARDCODED_MODEL_ID_MAPPING" in consts.ts, for dev purposes.
+ *
+ * - If you work at Nebius and want to update this mapping, please use the model mapping API we provide on huggingface.co
+ * - If you're a community member and want to add a new supported HF model to Nebius, please open an issue on the present repo
+ * and we will tag Nebius team members.
+ *
+ * Thanks!
+ */
+//# sourceMappingURL=nebius.d.ts.map

package/dist/src/providers/nebius.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"nebius.d.ts","sourceRoot":"","sources":["../../../src/providers/nebius.ts"],"names":[],"mappings":"AAAA,eAAO,MAAM,mBAAmB,iCAAiC,CAAC;AAElE;;;;;;;;;;;;;;;GAeG"}

package/dist/src/tasks/custom/request.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"request.d.ts","sourceRoot":"","sources":["../../../../src/tasks/custom/request.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAGvE;;GAEG;AACH,wBAAsB,OAAO,CAAC,CAAC,EAC9B,IAAI,EAAE,WAAW,EACjB,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,yFAAyF;IACzF,IAAI,CAAC,EAAE,MAAM,GAAG,aAAa,CAAC;IAC9B,sCAAsC;IACtC,QAAQ,CAAC,EAAE,aAAa,CAAC;IACzB,oCAAoC;IACpC,cAAc,CAAC,EAAE,OAAO,CAAC;CACzB,GACC,OAAO,CAAC,CAAC,CAAC,CAmCZ"}
+{"version":3,"file":"request.d.ts","sourceRoot":"","sources":["../../../../src/tasks/custom/request.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAGvE;;GAEG;AACH,wBAAsB,OAAO,CAAC,CAAC,EAC9B,IAAI,EAAE,WAAW,EACjB,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,yFAAyF;IACzF,IAAI,CAAC,EAAE,MAAM,GAAG,aAAa,CAAC;IAC9B,sCAAsC;IACtC,QAAQ,CAAC,EAAE,aAAa,CAAC;IACzB,oCAAoC;IACpC,cAAc,CAAC,EAAE,OAAO,CAAC;CACzB,GACC,OAAO,CAAC,CAAC,CAAC,CAgCZ"}

package/dist/src/tasks/custom/streamingRequest.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"streamingRequest.d.ts","sourceRoot":"","sources":["../../../../src/tasks/custom/streamingRequest.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAKvE;;GAEG;AACH,wBAAuB,gBAAgB,CAAC,CAAC,EACxC,IAAI,EAAE,WAAW,EACjB,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,yFAAyF;IACzF,IAAI,CAAC,EAAE,MAAM,GAAG,aAAa,CAAC;IAC9B,sCAAsC;IACtC,QAAQ,CAAC,EAAE,aAAa,CAAC;IACzB,oCAAoC;IACpC,cAAc,CAAC,EAAE,OAAO,CAAC;CACzB,GACC,cAAc,CAAC,CAAC,CAAC,CAsFnB"}
+{"version":3,"file":"streamingRequest.d.ts","sourceRoot":"","sources":["../../../../src/tasks/custom/streamingRequest.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,OAAO,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAKvE;;GAEG;AACH,wBAAuB,gBAAgB,CAAC,CAAC,EACxC,IAAI,EAAE,WAAW,EACjB,OAAO,CAAC,EAAE,OAAO,GAAG;IACnB,yFAAyF;IACzF,IAAI,CAAC,EAAE,MAAM,GAAG,aAAa,CAAC;IAC9B,sCAAsC;IACtC,QAAQ,CAAC,EAAE,aAAa,CAAC;IACzB,oCAAoC;IACpC,cAAc,CAAC,EAAE,OAAO,CAAC;CACzB,GACC,cAAc,CAAC,CAAC,CAAC,CAmFnB"}

package/dist/src/tasks/cv/textToImage.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"textToImage.d.ts","sourceRoot":"","sources":["../../../../src/tasks/cv/textToImage.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAqB,MAAM,oBAAoB,CAAC;AAE9E,OAAO,KAAK,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,aAAa,CAAC;AAIrD,MAAM,MAAM,eAAe,GAAG,QAAQ,GAAG,gBAAgB,CAAC;AAW1D;;;GAGG;AACH,wBAAsB,WAAW,CAAC,IAAI,EAAE,eAAe,EAAE,OAAO,CAAC,EAAE,OAAO,GAAG,OAAO,CAAC,IAAI,CAAC,CAoCzF"}
+{"version":3,"file":"textToImage.d.ts","sourceRoot":"","sources":["../../../../src/tasks/cv/textToImage.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAqB,MAAM,oBAAoB,CAAC;AAE9E,OAAO,KAAK,EAAE,QAAQ,EAAqB,OAAO,EAAE,MAAM,aAAa,CAAC;AAIxE,MAAM,MAAM,eAAe,GAAG,QAAQ,GAAG,gBAAgB,CAAC;AA0B1D;;;GAGG;AACH,wBAAsB,WAAW,CAAC,IAAI,EAAE,eAAe,EAAE,OAAO,CAAC,EAAE,OAAO,GAAG,OAAO,CAAC,IAAI,CAAC,CAqCzF"}

package/dist/src/tasks/nlp/chatCompletion.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"chatCompletion.d.ts","sourceRoot":"","sources":["../../../../src/tasks/nlp/chatCompletion.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,aAAa,CAAC;AAErD,OAAO,KAAK,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAAM,oBAAoB,CAAC;AAEpF;;GAEG;AACH,wBAAsB,cAAc,CACnC,IAAI,EAAE,QAAQ,GAAG,mBAAmB,EACpC,OAAO,CAAC,EAAE,OAAO,GACf,OAAO,CAAC,oBAAoB,CAAC,CAoB/B"}
+{"version":3,"file":"chatCompletion.d.ts","sourceRoot":"","sources":["../../../../src/tasks/nlp/chatCompletion.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,aAAa,CAAC;AAErD,OAAO,KAAK,EAAE,mBAAmB,EAAE,oBAAoB,EAAE,MAAM,oBAAoB,CAAC;AAEpF;;GAEG;AACH,wBAAsB,cAAc,CACnC,IAAI,EAAE,QAAQ,GAAG,mBAAmB,EACpC,OAAO,CAAC,EAAE,OAAO,GACf,OAAO,CAAC,oBAAoB,CAAC,CAuB/B"}

package/dist/src/types.d.ts

@@ -5,25 +5,9 @@ import type { ChatCompletionInput, PipelineType } from "@huggingface/tasks";
 export type ModelId = string;
 export interface Options {
     /**
-     * (Default: true) Boolean. If a request 503s and wait_for_model is set to false, the request will be retried with the same parameters but with wait_for_model set to true.
+     * (Default: true) Boolean. If a request 503s, the request will be retried with the same parameters.
      */
     retry_on_error?: boolean;
-    /**
-     * (Default: true). Boolean. There is a cache layer on Inference API (serverless) to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
-     */
-    use_cache?: boolean;
-    /**
-     * (Default: false). Boolean. Do not load the model if it's not already available.
-     */
-    dont_load_model?: boolean;
-    /**
-     * (Default: false). Boolean to use GPU instead of CPU for inference (requires Startup plan at least).
-     */
-    use_gpu?: boolean;
-    /**
-     * (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.
-     */
-    wait_for_model?: boolean;
     /**
      * Custom fetch function to use instead of the default one, for example to use a proxy or edit headers.
      */
@@ -38,7 +22,7 @@ export interface Options {
     includeCredentials?: string | boolean;
 }
 export type InferenceTask = Exclude<PipelineType, "other">;
-export declare const INFERENCE_PROVIDERS: readonly ["fal-ai", "fireworks-ai", "hf-inference", "replicate", "sambanova", "together"];
+export declare const INFERENCE_PROVIDERS: readonly ["fal-ai", "fireworks-ai", "nebius", "hf-inference", "replicate", "sambanova", "together"];
 export type InferenceProvider = (typeof INFERENCE_PROVIDERS)[number];
 export interface BaseArgs {
     /**

package/dist/src/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,mBAAmB,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAE5E;;GAEG;AACH,MAAM,MAAM,OAAO,GAAG,MAAM,CAAC;AAE7B,MAAM,WAAW,OAAO;IACvB;;OAEG;IACH,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB;;OAEG;IACH,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB;;OAEG;IACH,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B;;OAEG;IACH,OAAO,CAAC,EAAE,OAAO,CAAC;IAElB;;OAEG;IACH,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB;;OAEG;IACH,KAAK,CAAC,EAAE,OAAO,KAAK,CAAC;IACrB;;OAEG;IACH,MAAM,CAAC,EAAE,WAAW,CAAC;IAErB;;OAEG;IACH,kBAAkB,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC;CACtC;AAED,MAAM,MAAM,aAAa,GAAG,OAAO,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;AAE3D,eAAO,MAAM,mBAAmB,2FAOtB,CAAC;AACX,MAAM,MAAM,iBAAiB,GAAG,CAAC,OAAO,mBAAmB,CAAC,CAAC,MAAM,CAAC,CAAC;AAErE,MAAM,WAAW,QAAQ;IACxB;;;;;;OAMG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB;;;;;;;OAOG;IACH,KAAK,CAAC,EAAE,OAAO,CAAC;IAEhB;;;;OAIG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB;;;;OAIG;IACH,QAAQ,CAAC,EAAE,iBAAiB,CAAC;CAC7B;AAED,MAAM,MAAM,WAAW,GAAG,QAAQ,GACjC,CACG;IAAE,IAAI,EAAE,IAAI,GAAG,WAAW,CAAA;CAAE,GAC5B;IAAE,MAAM,EAAE,OAAO,CAAA;CAAE,GACnB;IAAE,MAAM,EAAE,MAAM,CAAA;CAAE,GAClB;IAAE,IAAI,EAAE,MAAM,CAAA;CAAE,GAChB;IAAE,SAAS,EAAE,MAAM,CAAA;CAAE,GACrB,mBAAmB,CACrB,GAAG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACrC,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB,CAAC"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../src/types.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,mBAAmB,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAE5E;;GAEG;AACH,MAAM,MAAM,OAAO,GAAG,MAAM,CAAC;AAE7B,MAAM,WAAW,OAAO;IACvB;;OAEG;IACH,cAAc,CAAC,EAAE,OAAO,CAAC;IAEzB;;OAEG;IACH,KAAK,CAAC,EAAE,OAAO,KAAK,CAAC;IACrB;;OAEG;IACH,MAAM,CAAC,EAAE,WAAW,CAAC;IAErB;;OAEG;IACH,kBAAkB,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC;CACtC;AAED,MAAM,MAAM,aAAa,GAAG,OAAO,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;AAE3D,eAAO,MAAM,mBAAmB,qGAQtB,CAAC;AACX,MAAM,MAAM,iBAAiB,GAAG,CAAC,OAAO,mBAAmB,CAAC,CAAC,MAAM,CAAC,CAAC;AAErE,MAAM,WAAW,QAAQ;IACxB;;;;;;OAMG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB;;;;;;;OAOG;IACH,KAAK,CAAC,EAAE,OAAO,CAAC;IAEhB;;;;OAIG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IAErB;;;;OAIG;IACH,QAAQ,CAAC,EAAE,iBAAiB,CAAC;CAC7B;AAED,MAAM,MAAM,WAAW,GAAG,QAAQ,GACjC,CACG;IAAE,IAAI,EAAE,IAAI,GAAG,WAAW,CAAA;CAAE,GAC5B;IAAE,MAAM,EAAE,OAAO,CAAA;CAAE,GACnB;IAAE,MAAM,EAAE,MAAM,CAAA;CAAE,GAClB;IAAE,IAAI,EAAE,MAAM,CAAA;CAAE,GAChB;IAAE,SAAS,EAAE,MAAM,CAAA;CAAE,GACrB,mBAAmB,CACrB,GAAG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACrC,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@huggingface/inference",
-  "version": "3.3.0",
+  "version": "3.3.1",
   "packageManager": "pnpm@8.10.5",
   "license": "MIT",
   "author": "Tim Mikeladze <tim.mikeladze@gmail.com>",

package/src/lib/makeRequestOptions.ts

@@ -1,5 +1,6 @@
 import { HF_HUB_URL, HF_ROUTER_URL } from "../config";
 import { FAL_AI_API_BASE_URL } from "../providers/fal-ai";
+import { NEBIUS_API_BASE_URL } from "../providers/nebius";
 import { REPLICATE_API_BASE_URL } from "../providers/replicate";
 import { SAMBANOVA_API_BASE_URL } from "../providers/sambanova";
 import { TOGETHER_API_BASE_URL } from "../providers/together";
@@ -38,8 +39,7 @@ export async function makeRequestOptions(
 	let otherArgs = remainingArgs;
 	const provider = maybeProvider ?? "hf-inference";
 
-	const { forceTask, includeCredentials, taskHint, wait_for_model, use_cache, dont_load_model, chatCompletion } =
-		options ?? {};
+	const { forceTask, includeCredentials, taskHint, chatCompletion } = options ?? {};
 
 	if (endpointUrl && provider !== "hf-inference") {
 		throw new Error(`Cannot use endpointUrl with a third-party provider.`);
@@ -101,18 +101,6 @@ export async function makeRequestOptions(
 		headers["Content-Type"] = "application/json";
 	}
 
-	if (provider === "hf-inference") {
-		if (wait_for_model) {
-			headers["X-Wait-For-Model"] = "true";
-		}
-		if (use_cache === false) {
-			headers["X-Use-Cache"] = "false";
-		}
-		if (dont_load_model) {
-			headers["X-Load-Model"] = "0";
-		}
-	}
-
 	if (provider === "replicate") {
 		headers["Prefer"] = "wait";
 	}
@@ -143,7 +131,7 @@ export async function makeRequestOptions(
 			? args.data
 			: JSON.stringify({
 					...otherArgs,
-					...(chatCompletion || provider === "together" ? { model } : undefined),
+					...(chatCompletion || provider === "together" || provider === "nebius" ? { model } : undefined),
 			  }),
 		...(credentials ? { credentials } : undefined),
 		signal: options?.signal,
@@ -172,6 +160,22 @@ function makeUrl(params: {
 				: FAL_AI_API_BASE_URL;
 			return `${baseUrl}/${params.model}`;
 		}
+		case "nebius": {
+			const baseUrl = shouldProxy
+				? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider)
+				: NEBIUS_API_BASE_URL;
+
+			if (params.taskHint === "text-to-image") {
+				return `${baseUrl}/v1/images/generations`;
+			}
+			if (params.taskHint === "text-generation") {
+				if (params.chatCompletion) {
+					return `${baseUrl}/v1/chat/completions`;
+				}
+				return `${baseUrl}/v1/completions`;
+			}
+			return baseUrl;
+		}
 		case "replicate": {
 			const baseUrl = shouldProxy
 				? HF_HUB_INFERENCE_PROXY_TEMPLATE.replace("{{PROVIDER}}", params.provider)

package/src/providers/consts.ts

@@ -19,6 +19,7 @@ export const HARDCODED_MODEL_ID_MAPPING: Record<InferenceProvider, Record<ModelI
 	"fal-ai": {},
 	"fireworks-ai": {},
 	"hf-inference": {},
+	nebius: {},
 	replicate: {},
 	sambanova: {},
 	together: {},

package/src/providers/nebius.ts

@@ -0,0 +1,18 @@
+export const NEBIUS_API_BASE_URL = "https://api.studio.nebius.ai";
+
+/**
+ * See the registered mapping of HF model ID => Nebius model ID here:
+ *
+ * https://huggingface.co/api/partners/nebius/models
+ *
+ * This is a publicly available mapping.
+ *
+ * If you want to try to run inference for a new model locally before it's registered on huggingface.co,
+ * you can add it to the dictionary "HARDCODED_MODEL_ID_MAPPING" in consts.ts, for dev purposes.
+ *
+ * - If you work at Nebius and want to update this mapping, please use the model mapping API we provide on huggingface.co
+ * - If you're a community member and want to add a new supported HF model to Nebius, please open an issue on the present repo
+ * and we will tag Nebius team members.
+ *
+ * Thanks!
+ */

package/src/tasks/custom/request.ts

@@ -18,11 +18,8 @@ export async function request<T>(
 	const { url, info } = await makeRequestOptions(args, options);
 	const response = await (options?.fetch ?? fetch)(url, info);
 
-	if (options?.retry_on_error !== false && response.status === 503 && !options?.wait_for_model) {
-		return request(args, {
-			...options,
-			wait_for_model: true,
-		});
+	if (options?.retry_on_error !== false && response.status === 503) {
+		return request(args, options);
 	}
 
 	if (!response.ok) {

package/src/tasks/custom/streamingRequest.ts

@@ -20,11 +20,8 @@ export async function* streamingRequest<T>(
 	const { url, info } = await makeRequestOptions({ ...args, stream: true }, options);
 	const response = await (options?.fetch ?? fetch)(url, info);
 
-	if (options?.retry_on_error !== false && response.status === 503 && !options?.wait_for_model) {
-		return yield* streamingRequest(args, {
-			...options,
-			wait_for_model: true,
-		});
+	if (options?.retry_on_error !== false && response.status === 503) {
+		return yield* streamingRequest(args, options);
 	}
 	if (!response.ok) {
 		if (response.headers.get("Content-Type")?.startsWith("application/json")) {

package/src/tasks/cv/textToImage.ts

@@ -1,6 +1,6 @@
 import type { TextToImageInput, TextToImageOutput } from "@huggingface/tasks";
 import { InferenceOutputError } from "../../lib/InferenceOutputError";
-import type { BaseArgs, Options } from "../../types";
+import type { BaseArgs, InferenceProvider, Options } from "../../types";
 import { omit } from "../../utils/omit";
 import { request } from "../custom/request";
 
@@ -15,24 +15,40 @@ interface OutputUrlImageGeneration {
 	output: string[];
 }
 
+function getResponseFormatArg(provider: InferenceProvider) {
+	switch (provider) {
+		case "fal-ai":
+			return { sync_mode: true };
+		case "nebius":
+			return { response_format: "b64_json" };
+		case "replicate":
+			return undefined;
+		case "together":
+			return { response_format: "base64" };
+		default:
+			return undefined;
+	}
+}
+
 /**
  * This task reads some text input and outputs an image.
  * Recommended model: stabilityai/stable-diffusion-2
  */
 export async function textToImage(args: TextToImageArgs, options?: Options): Promise<Blob> {
 	const payload =
-		args.provider === "together" || args.provider === "fal-ai" || args.provider === "replicate"
-			? {
+		!args.provider || args.provider === "hf-inference" || args.provider === "sambanova"
+			? args
+			: {
 					...omit(args, ["inputs", "parameters"]),
 					...args.parameters,
-					...(args.provider !== "replicate" ? { response_format: "base64" } : undefined),
+					...getResponseFormatArg(args.provider),
 					prompt: args.inputs,
-			  }
-			: args;
+			  };
 	const res = await request<TextToImageOutput | Base64ImageGeneration | OutputUrlImageGeneration>(payload, {
 		...options,
 		taskHint: "text-to-image",
 	});
+
 	if (res && typeof res === "object") {
 		if (args.provider === "fal-ai" && "images" in res && Array.isArray(res.images) && res.images[0].url) {
 			const image = await fetch(res.images[0].url);

package/src/tasks/nlp/chatCompletion.ts

@@ -15,14 +15,17 @@ export async function chatCompletion(
 		taskHint: "text-generation",
 		chatCompletion: true,
 	});
+
 	const isValidOutput =
 		typeof res === "object" &&
 		Array.isArray(res?.choices) &&
 		typeof res?.created === "number" &&
 		typeof res?.id === "string" &&
 		typeof res?.model === "string" &&
-		/// Together.ai does not output a system_fingerprint
-		(res.system_fingerprint === undefined || typeof res.system_fingerprint === "string") &&
+		/// Together.ai and Nebius do not output a system_fingerprint
+		(res.system_fingerprint === undefined ||
+			res.system_fingerprint === null ||
+			typeof res.system_fingerprint === "string") &&
 		typeof res?.usage === "object";
 
 	if (!isValidOutput) {

package/src/types.ts

@@ -7,26 +7,10 @@ export type ModelId = string;
 
 export interface Options {
 	/**
-	 * (Default: true) Boolean. If a request 503s and wait_for_model is set to false, the request will be retried with the same parameters but with wait_for_model set to true.
+	 * (Default: true) Boolean. If a request 503s, the request will be retried with the same parameters.
 	 */
 	retry_on_error?: boolean;
-	/**
-	 * (Default: true). Boolean. There is a cache layer on Inference API (serverless) to speedup requests we have already seen. Most models can use those results as is as models are deterministic (meaning the results will be the same anyway). However if you use a non deterministic model, you can set this parameter to prevent the caching mechanism from being used resulting in a real new query.
-	 */
-	use_cache?: boolean;
-	/**
-	 * (Default: false). Boolean. Do not load the model if it's not already available.
-	 */
-	dont_load_model?: boolean;
-	/**
-	 * (Default: false). Boolean to use GPU instead of CPU for inference (requires Startup plan at least).
-	 */
-	use_gpu?: boolean;
 
-	/**
-	 * (Default: false) Boolean. If the model is not ready, wait for it instead of receiving 503. It limits the number of requests required to get your inference done. It is advised to only set this flag to true after receiving a 503 error as it will limit hanging in your application to known places.
-	 */
-	wait_for_model?: boolean;
 	/**
 	 * Custom fetch function to use instead of the default one, for example to use a proxy or edit headers.
 	 */
@@ -47,6 +31,7 @@ export type InferenceTask = Exclude<PipelineType, "other">;
 export const INFERENCE_PROVIDERS = [
 	"fal-ai",
 	"fireworks-ai",
+	"nebius",
 	"hf-inference",
 	"replicate",
 	"sambanova",