@gpuai/sdk 0.3.5 → 0.3.7

package/README.md

@@ -1,4 +1,4 @@
-# @gpuai/sdk@0.3.5
+# @gpuai/sdk@0.3.7
 
 A TypeScript SDK client for the api.gpu.ai API.
 
@@ -203,7 +203,7 @@ and is automatically generated by the
 [OpenAPI Generator](https://openapi-generator.tech) project:
 
 - API version: `1.0.0`
-- Package version: `0.3.5`
+- Package version: `0.3.7`
 - Generator version: `7.24.0`
 - Build package: `org.openapitools.codegen.languages.TypeScriptFetchClientCodegen`
 

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

@@ -97,7 +97,7 @@ export interface CreateInstanceRequest {
         [key: string]: string;
     };
     /**
-     * Launch from your own container image, e.g. `pytorch/pytorch:2.4.1-cuda12.4-cudnn9-devel` or `ghcr.io/acme/trainer:v1`. The image must be publicly pullable (private registries are not yet supported) and provide linux/amd64 — both are verified synchronously before anything is provisioned: an unknown image returns 422 `image_unavailable` and a registry outage 502 `registry_unreachable`, with no operation created and nothing billed. The launch is pinned to the exact manifest digest verified, so a tag re-push cannot change what boots. After SSH/tunnel setup the image's own ENTRYPOINT/CMD is started (its output is in /var/log/gpuai-entrypoint.log on the instance); an image with neither is a plain SSH environment. Placement narrows to container-native capacity — a gpu_type only available on VM capacity returns 422 `image_unavailable`. Mutually exclusive with `template_id` and `environment` (422). A malformed reference (bad charset, localhost/private-IP registry) returns 400.
+     * Launch from your own container image, e.g. `pytorch/pytorch:2.4.1-cuda12.4-cudnn9-devel` or `ghcr.io/acme/trainer:v1`. The image must be publicly pullable (private registries are not yet supported) and provide linux/amd64 — both are verified synchronously before anything is provisioned: an unknown image returns 422 `image_unavailable` and a registry outage 502 `registry_unreachable`, with no operation created and nothing billed. The launch is pinned to the exact manifest digest verified, so a tag re-push cannot change what boots. After SSH/tunnel setup the image's own ENTRYPOINT/CMD is started (its output is in /var/log/gpuai-entrypoint.log on the instance); an image with neither is a plain SSH environment. Placement prefers container-native capacity (the image runs natively; SSH lands inside it) and falls back to VM capacity (the image runs as a container beside the host's SSH — inspect with `docker logs gpuai-workload`); private images stay container-native. A gpu_type with no capable capacity returns 422 `image_unavailable`. Mutually exclusive with `template_id` and `environment` (422). A malformed reference (bad charset, localhost/private-IP registry) returns 400. On-demand only for now: `tier: spot` with `image` is refused (422 `validation_failed`) because a preemption re-route cannot yet carry a custom image.
      * @type {string}
      * @memberof CreateInstanceRequest
      */

package/dist/models/CreateInstanceRequest.d.ts

@@ -97,7 +97,7 @@ export interface CreateInstanceRequest {
         [key: string]: string;
     };
     /**
-     * Launch from your own container image, e.g. `pytorch/pytorch:2.4.1-cuda12.4-cudnn9-devel` or `ghcr.io/acme/trainer:v1`. The image must be publicly pullable (private registries are not yet supported) and provide linux/amd64 — both are verified synchronously before anything is provisioned: an unknown image returns 422 `image_unavailable` and a registry outage 502 `registry_unreachable`, with no operation created and nothing billed. The launch is pinned to the exact manifest digest verified, so a tag re-push cannot change what boots. After SSH/tunnel setup the image's own ENTRYPOINT/CMD is started (its output is in /var/log/gpuai-entrypoint.log on the instance); an image with neither is a plain SSH environment. Placement narrows to container-native capacity — a gpu_type only available on VM capacity returns 422 `image_unavailable`. Mutually exclusive with `template_id` and `environment` (422). A malformed reference (bad charset, localhost/private-IP registry) returns 400.
+     * Launch from your own container image, e.g. `pytorch/pytorch:2.4.1-cuda12.4-cudnn9-devel` or `ghcr.io/acme/trainer:v1`. The image must be publicly pullable (private registries are not yet supported) and provide linux/amd64 — both are verified synchronously before anything is provisioned: an unknown image returns 422 `image_unavailable` and a registry outage 502 `registry_unreachable`, with no operation created and nothing billed. The launch is pinned to the exact manifest digest verified, so a tag re-push cannot change what boots. After SSH/tunnel setup the image's own ENTRYPOINT/CMD is started (its output is in /var/log/gpuai-entrypoint.log on the instance); an image with neither is a plain SSH environment. Placement prefers container-native capacity (the image runs natively; SSH lands inside it) and falls back to VM capacity (the image runs as a container beside the host's SSH — inspect with `docker logs gpuai-workload`); private images stay container-native. A gpu_type with no capable capacity returns 422 `image_unavailable`. Mutually exclusive with `template_id` and `environment` (422). A malformed reference (bad charset, localhost/private-IP registry) returns 400. On-demand only for now: `tier: spot` with `image` is refused (422 `validation_failed`) because a preemption re-route cannot yet carry a custom image.
      * @type {string}
      * @memberof CreateInstanceRequest
      */

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gpuai/sdk",
-  "version": "0.3.5",
+  "version": "0.3.7",
   "description": "TypeScript SDK for the GPU.ai Public Developer API",
   "author": "GPU.ai Support <support@gpu.ai>",
   "repository": {

package/src/models/CreateInstanceRequest.ts

@@ -106,7 +106,7 @@ export interface CreateInstanceRequest {
      */
     env?: { [key: string]: string; };
     /**
-     * Launch from your own container image, e.g. `pytorch/pytorch:2.4.1-cuda12.4-cudnn9-devel` or `ghcr.io/acme/trainer:v1`. The image must be publicly pullable (private registries are not yet supported) and provide linux/amd64 — both are verified synchronously before anything is provisioned: an unknown image returns 422 `image_unavailable` and a registry outage 502 `registry_unreachable`, with no operation created and nothing billed. The launch is pinned to the exact manifest digest verified, so a tag re-push cannot change what boots. After SSH/tunnel setup the image's own ENTRYPOINT/CMD is started (its output is in /var/log/gpuai-entrypoint.log on the instance); an image with neither is a plain SSH environment. Placement narrows to container-native capacity — a gpu_type only available on VM capacity returns 422 `image_unavailable`. Mutually exclusive with `template_id` and `environment` (422). A malformed reference (bad charset, localhost/private-IP registry) returns 400.
+     * Launch from your own container image, e.g. `pytorch/pytorch:2.4.1-cuda12.4-cudnn9-devel` or `ghcr.io/acme/trainer:v1`. The image must be publicly pullable (private registries are not yet supported) and provide linux/amd64 — both are verified synchronously before anything is provisioned: an unknown image returns 422 `image_unavailable` and a registry outage 502 `registry_unreachable`, with no operation created and nothing billed. The launch is pinned to the exact manifest digest verified, so a tag re-push cannot change what boots. After SSH/tunnel setup the image's own ENTRYPOINT/CMD is started (its output is in /var/log/gpuai-entrypoint.log on the instance); an image with neither is a plain SSH environment. Placement prefers container-native capacity (the image runs natively; SSH lands inside it) and falls back to VM capacity (the image runs as a container beside the host's SSH — inspect with `docker logs gpuai-workload`); private images stay container-native. A gpu_type with no capable capacity returns 422 `image_unavailable`. Mutually exclusive with `template_id` and `environment` (422). A malformed reference (bad charset, localhost/private-IP registry) returns 400. On-demand only for now: `tier: spot` with `image` is refused (422 `validation_failed`) because a preemption re-route cannot yet carry a custom image.
      * @type {string}
      * @memberof CreateInstanceRequest
      */