@gpuai/sdk 0.3.2 → 0.3.3

package/README.md

@@ -1,4 +1,4 @@
-# @gpuai/sdk@0.3.2
+# @gpuai/sdk@0.3.3
 
 A TypeScript SDK client for the api.gpu.ai API.
 
@@ -196,7 +196,7 @@ and is automatically generated by the
 [OpenAPI Generator](https://openapi-generator.tech) project:
 
 - API version: `1.0.0`
-- Package version: `0.3.2`
+- Package version: `0.3.3`
 - Generator version: `7.24.0`
 - Build package: `org.openapitools.codegen.languages.TypeScriptFetchClientCodegen`
 

package/dist/esm/models/CreateInstanceRequest.d.ts

@@ -64,7 +64,7 @@ export interface CreateInstanceRequest {
      */
     viewedPricePerHour?: number;
     /**
-     * Deploy a curated application template (e.g. comfyui, vllm). When set, gpu_type may be omitted — the cheapest eligible GPU is auto-selected. For templates serving a Hugging Face model via env.MODEL (vllm, sglang), the model's published weights size is checked against the GPU's VRAM at submit time: an oversized model on an explicit gpu_type returns 422 `model_too_large` (auto-select instead raises its VRAM floor to the estimate, and 422s only when no available GPU fits). Gated/private/unknown repos are never blocked — the deploy proceeds and the 202 carries a `warnings` entry.
+     * Deploy a curated application template (e.g. comfyui, vllm). When set, gpu_type may be omitted — the cheapest eligible GPU is auto-selected. For templates serving a Hugging Face model via env.MODEL (vllm, sglang), the model's published weights size is checked against the GPU's VRAM at submit time: an oversized model on an explicit gpu_type returns 422 `model_too_large` (auto-select instead raises its VRAM floor to the estimate, and 422s only when no available GPU fits). Gated/private/unknown repos are never blocked — the deploy proceeds and the 202 carries a `warnings` entry. Separately, an explicit gpu_type whose VRAM is below the template's own `min_vram_gb` returns 422 `gpu_vram_too_small`, naming the floor, the type's VRAM and the compatible GPU types. Auto-select is unaffected — it already filters by the floor.
      * @type {string}
      * @memberof CreateInstanceRequest
      */

package/dist/esm/models/Problem.d.ts

@@ -67,6 +67,7 @@ export declare const ProblemCodeEnum: {
     readonly DiskUnavailable: "disk_unavailable";
     readonly DiskExceedsOffering: "disk_exceeds_offering";
     readonly ModelTooLarge: "model_too_large";
+    readonly GpuVramTooSmall: "gpu_vram_too_small";
     readonly PriceChanged: "price_changed";
     readonly InsufficientBalance: "insufficient_balance";
     readonly QuotaExceeded: "quota_exceeded";

package/dist/esm/models/Problem.js

@@ -26,6 +26,7 @@ export const ProblemCodeEnum = {
     DiskUnavailable: 'disk_unavailable',
     DiskExceedsOffering: 'disk_exceeds_offering',
     ModelTooLarge: 'model_too_large',
+    GpuVramTooSmall: 'gpu_vram_too_small',
     PriceChanged: 'price_changed',
     InsufficientBalance: 'insufficient_balance',
     QuotaExceeded: 'quota_exceeded',

package/dist/models/CreateInstanceRequest.d.ts

@@ -64,7 +64,7 @@ export interface CreateInstanceRequest {
      */
     viewedPricePerHour?: number;
     /**
-     * Deploy a curated application template (e.g. comfyui, vllm). When set, gpu_type may be omitted — the cheapest eligible GPU is auto-selected. For templates serving a Hugging Face model via env.MODEL (vllm, sglang), the model's published weights size is checked against the GPU's VRAM at submit time: an oversized model on an explicit gpu_type returns 422 `model_too_large` (auto-select instead raises its VRAM floor to the estimate, and 422s only when no available GPU fits). Gated/private/unknown repos are never blocked — the deploy proceeds and the 202 carries a `warnings` entry.
+     * Deploy a curated application template (e.g. comfyui, vllm). When set, gpu_type may be omitted — the cheapest eligible GPU is auto-selected. For templates serving a Hugging Face model via env.MODEL (vllm, sglang), the model's published weights size is checked against the GPU's VRAM at submit time: an oversized model on an explicit gpu_type returns 422 `model_too_large` (auto-select instead raises its VRAM floor to the estimate, and 422s only when no available GPU fits). Gated/private/unknown repos are never blocked — the deploy proceeds and the 202 carries a `warnings` entry. Separately, an explicit gpu_type whose VRAM is below the template's own `min_vram_gb` returns 422 `gpu_vram_too_small`, naming the floor, the type's VRAM and the compatible GPU types. Auto-select is unaffected — it already filters by the floor.
      * @type {string}
      * @memberof CreateInstanceRequest
      */

package/dist/models/Problem.d.ts

@@ -67,6 +67,7 @@ export declare const ProblemCodeEnum: {
     readonly DiskUnavailable: "disk_unavailable";
     readonly DiskExceedsOffering: "disk_exceeds_offering";
     readonly ModelTooLarge: "model_too_large";
+    readonly GpuVramTooSmall: "gpu_vram_too_small";
     readonly PriceChanged: "price_changed";
     readonly InsufficientBalance: "insufficient_balance";
     readonly QuotaExceeded: "quota_exceeded";

package/dist/models/Problem.js

@@ -34,6 +34,7 @@ exports.ProblemCodeEnum = {
     DiskUnavailable: 'disk_unavailable',
     DiskExceedsOffering: 'disk_exceeds_offering',
     ModelTooLarge: 'model_too_large',
+    GpuVramTooSmall: 'gpu_vram_too_small',
     PriceChanged: 'price_changed',
     InsufficientBalance: 'insufficient_balance',
     QuotaExceeded: 'quota_exceeded',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gpuai/sdk",
-  "version": "0.3.2",
+  "version": "0.3.3",
   "description": "TypeScript SDK for the GPU.ai Public Developer API",
   "author": "GPU.ai Support <support@gpu.ai>",
   "repository": {

package/src/models/CreateInstanceRequest.ts

@@ -68,7 +68,7 @@ export interface CreateInstanceRequest {
      */
     viewedPricePerHour?: number;
     /**
-     * Deploy a curated application template (e.g. comfyui, vllm). When set, gpu_type may be omitted — the cheapest eligible GPU is auto-selected. For templates serving a Hugging Face model via env.MODEL (vllm, sglang), the model's published weights size is checked against the GPU's VRAM at submit time: an oversized model on an explicit gpu_type returns 422 `model_too_large` (auto-select instead raises its VRAM floor to the estimate, and 422s only when no available GPU fits). Gated/private/unknown repos are never blocked — the deploy proceeds and the 202 carries a `warnings` entry.
+     * Deploy a curated application template (e.g. comfyui, vllm). When set, gpu_type may be omitted — the cheapest eligible GPU is auto-selected. For templates serving a Hugging Face model via env.MODEL (vllm, sglang), the model's published weights size is checked against the GPU's VRAM at submit time: an oversized model on an explicit gpu_type returns 422 `model_too_large` (auto-select instead raises its VRAM floor to the estimate, and 422s only when no available GPU fits). Gated/private/unknown repos are never blocked — the deploy proceeds and the 202 carries a `warnings` entry. Separately, an explicit gpu_type whose VRAM is below the template's own `min_vram_gb` returns 422 `gpu_vram_too_small`, naming the floor, the type's VRAM and the compatible GPU types. Auto-select is unaffected — it already filters by the floor.
      * @type {string}
      * @memberof CreateInstanceRequest
      */

package/src/models/Problem.ts

@@ -73,6 +73,7 @@ export const ProblemCodeEnum = {
     DiskUnavailable: 'disk_unavailable',
     DiskExceedsOffering: 'disk_exceeds_offering',
     ModelTooLarge: 'model_too_large',
+    GpuVramTooSmall: 'gpu_vram_too_small',
     PriceChanged: 'price_changed',
     InsufficientBalance: 'insufficient_balance',
     QuotaExceeded: 'quota_exceeded',