trainfabric 0.1.10 → 0.1.12

package/README.md

@@ -29,7 +29,7 @@ trainfabric projects:list
 trainfabric datasets:upload ./train.jsonl --project <projectId>
 trainfabric runs:quote --summary --project <projectId> --dataset <datasetId> --model llama-3-8b
 trainfabric runs:create --yes --project <projectId> --dataset <datasetId> --model llama-3-8b
-trainfabric runs:watch <runId>
+trainfabric runs:watch <runId> --timeout 30m
 trainfabric runs:cost-breakdown <runId> --summary
 ```
 

package/dist/index.cjs

@@ -8204,6 +8204,9 @@ var DatasetsClient = class extends ResourceClient {
 
 // ../sdk/dist/client/deployments.js
 var DeploymentsClient = class extends ResourceClient {
+  list() {
+    return this.requestGet("/v1/deployments");
+  }
   create(input) {
     return this.requestPost("/v1/deployments", input);
   }
@@ -8385,8 +8388,11 @@ var RunHandle = class {
       this.listeners.get(event)?.delete(listener);
     };
   }
-  async wait() {
+  async wait(options) {
     this.ensureStreaming();
+    const startedAt = Date.now();
+    const pollMs = options?.pollMs ?? 1e3;
+    const timeoutMs = options?.timeoutMs;
     for (; ; ) {
       const detail = await this.client.runs.get(this.snapshot.id);
       this.snapshot = detail.run;
@@ -8403,7 +8409,11 @@ var RunHandle = class {
         }
         return detail;
       }
-      await new Promise((resolve) => setTimeout(resolve, 1e3));
+      if (timeoutMs !== void 0 && Date.now() - startedAt > timeoutMs) {
+        this.close();
+        throw new Error(`Timed out waiting for run ${this.snapshot.id}.`);
+      }
+      await new Promise((resolve) => setTimeout(resolve, pollMs));
     }
   }
   close() {
@@ -8621,9 +8631,9 @@ var RunsClient = class extends ResourceClient {
   explanation(runId) {
     return this.requestGet(`/v1/runs/${runId}/explanation`);
   }
-  async wait(runId) {
+  async wait(runId, options) {
     const handle = await this.watch(runId);
-    return handle.wait();
+    return handle.wait(options);
   }
 };
 
@@ -8911,7 +8921,7 @@ function buildComputeSpec(options) {
 
 // src/index.ts
 var DEFAULT_TRAINFABRIC_API_URL2 = "https://api.trainfabric.com";
-var CLI_VERSION = "0.1.10";
+var CLI_VERSION = "0.1.12";
 var CONFIG_DIR = import_node_path3.default.join(import_node_os2.default.homedir(), ".trainfabric");
 var CONFIG_PATH = import_node_path3.default.join(CONFIG_DIR, "config.json");
 var FALLBACK_SECRET_PATH = import_node_path3.default.join(CONFIG_DIR, "session.enc");
@@ -9287,7 +9297,7 @@ function assertQuoteOptions(bundle, options) {
     )}). Omit --accelerator or increase --min-memory.`
   );
 }
-async function watchRun(runId, json = false) {
+async function watchRun(runId, json = false, timeoutMs) {
   const handle = await createClient(loadConfig()).runs.watch(runId);
   if (json) {
     const emit = (type, payload) => printJson({ type, payload });
@@ -9324,9 +9334,26 @@ async function watchRun(runId, json = false) {
       console.log(`[failed] ${payload.reason ?? "Run failed."}`);
     });
   }
-  const detail = await handle.wait();
+  const detail = await handle.wait({ timeoutMs });
   printJson(detail);
 }
+function parseDurationMs(value) {
+  if (!value) {
+    return void 0;
+  }
+  const trimmed = value.trim().toLowerCase();
+  const match = trimmed.match(/^(\d+(?:\.\d+)?)(ms|s|m|h)?$/);
+  if (!match) {
+    throw new Error(`Invalid duration "${value}". Use values like 30s, 10m, 1h, or 5000ms.`);
+  }
+  const amount = Number(match[1]);
+  const unit = match[2] ?? "ms";
+  const multiplier = unit === "h" ? 36e5 : unit === "m" ? 6e4 : unit === "s" ? 1e3 : 1;
+  return Math.max(1, Math.round(amount * multiplier));
+}
+function parsePositiveMs(value, fallback) {
+  return parseDurationMs(value) ?? fallback;
+}
 function createClient(config) {
   return new TrainingClient({
     accessToken: config.accessToken,
@@ -9495,11 +9522,16 @@ program2.command("runs:status").argument("<runId>").description("Fetch full run
 program2.command("runs:logs").argument("<runId>").description("Fetch run logs").action(async (runId) => {
   printJson(await createClient(loadConfig()).runs.logs(String(runId)));
 });
-program2.command("runs:wait").argument("<runId>").description("Wait for a run to reach a terminal state").action(async (runId) => {
-  printJson(await createClient(loadConfig()).runs.wait(String(runId)));
+program2.command("runs:wait").argument("<runId>").option("--timeout <duration>", "maximum wait duration, for example 30s, 10m, or 1h").option("--poll <duration>", "poll interval, for example 1s or 500ms").description("Wait for a run to reach a terminal state").action(async (runId, options) => {
+  printJson(
+    await createClient(loadConfig()).runs.wait(String(runId), {
+      timeoutMs: parseDurationMs(options.timeout),
+      pollMs: parsePositiveMs(options.poll, 1e3)
+    })
+  );
 });
-program2.command("runs:watch").argument("<runId>").option("--json", "emit event output as JSON").description("Stream a run until it completes").action(async (runId, options) => {
-  await watchRun(String(runId), Boolean(options.json));
+program2.command("runs:watch").argument("<runId>").option("--json", "emit event output as JSON").option("--timeout <duration>", "maximum watch duration, for example 30s, 10m, or 1h").description("Stream a run until it completes").action(async (runId, options) => {
+  await watchRun(String(runId), Boolean(options.json), parseDurationMs(options.timeout));
 });
 program2.command("runs:usage").argument("<runId>").option("--summary", "print a human-readable cost summary instead of raw JSON").description("Fetch run usage summary").action(async (runId, options) => {
   const usage = await createClient(loadConfig()).runs.usage(String(runId));
@@ -9547,7 +9579,10 @@ program2.command("models:list").description("List exported models").action(async
 program2.command("models:export").argument("<modelId>").description("Fetch the export package path for a model").action(async (modelId) => {
   printJson(await createClient(loadConfig()).models.export(String(modelId)));
 });
-program2.command("deployments:create").requiredOption("--model <modelId>").description("Create a deployment for a model").action(async (options) => {
+program2.command("deployments:list").description("List deployments").action(async () => {
+  printJson(await createClient(loadConfig()).deployments.list());
+});
+program2.command("deployments:create").requiredOption("--model <modelId>").option("--project <projectId>", "accepted for consistency; deployment project is inferred from the model").description("Create a deployment for a model").action(async (options) => {
   printJson(await createClient(loadConfig()).deployments.create({ modelId: options.model }));
 });
 program2.command("deployments:get").argument("<deploymentId>").description("Fetch a deployment").action(async (deploymentId) => {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "trainfabric",
-  "version": "0.1.10",
+  "version": "0.1.12",
   "description": "Trainfabric CLI for launching GPU training jobs on the hosted Trainfabric backend.",
   "type": "module",
   "main": "dist/index.cjs",