trainfabric 0.1.11 → 0.1.13

package/README.md

@@ -28,8 +28,8 @@ Secrets are stored in macOS Keychain when available. Other platforms use an encr
 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:create --yes --quote <quoteId> --project <projectId> --dataset <datasetId> --model llama-3-8b
+trainfabric runs:watch <runId> --timeout 30m
 trainfabric runs:cost-breakdown <runId> --summary
 ```
 

package/dist/index.cjs

@@ -7975,12 +7975,6 @@ function validateChatJsonl(text) {
     }
     tokenEstimate += estimateTokens(result.data.messages.map((message) => `${message.role}:${message.content}`).join("\n"));
   }
-  if (tokenEstimate < 100) {
-    warnings.push({
-      code: "small_dataset",
-      message: "Dataset is valid but very small for a meaningful training run."
-    });
-  }
   const invalidWarnings = /* @__PURE__ */ new Set([
     "empty_file",
     "invalid_jsonl",
@@ -7988,8 +7982,15 @@ function validateChatJsonl(text) {
     "missing_assistant",
     "empty_assistant"
   ]);
+  const isValid2 = !warnings.some((warning) => invalidWarnings.has(warning.code));
+  if (tokenEstimate < 100) {
+    warnings.push({
+      code: "small_dataset",
+      message: isValid2 ? "Dataset is valid but very small for a meaningful training run." : "Dataset is very small; fix schema errors before using it for a training run."
+    });
+  }
   return {
-    valid: !warnings.some((warning) => invalidWarnings.has(warning.code)),
+    valid: isValid2,
     format: "chat_jsonl",
     rowCount,
     duplicateCount,
@@ -8388,8 +8389,11 @@ 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;
@@ -8406,7 +8410,11 @@ 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 ${this.snapshot.id}.`);
+      }
+      await new Promise((resolve) => setTimeout(resolve, pollMs));
     }
   }
   close() {
@@ -8624,9 +8632,9 @@ var RunsClient = class extends ResourceClient {
   explanation(runId) {
     return this.requestGet(`/v1/runs/${runId}/explanation`);
   }
-  async wait(runId) {
+  async wait(runId, options) {
     const handle = await this.watch(runId);
-    return handle.wait();
+    return handle.wait(options);
   }
 };
 
@@ -8914,7 +8922,7 @@ function buildComputeSpec(options) {
 
 // src/index.ts
 var DEFAULT_TRAINFABRIC_API_URL2 = "https://api.trainfabric.com";
-var CLI_VERSION = "0.1.11";
+var CLI_VERSION = "0.1.13";
 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");
@@ -9208,6 +9216,7 @@ function printQuoteSummary(bundle) {
     console.log(`  data path: ${quote.dataPath ?? "auto"}`);
     console.log(`  GPUs: ${gpuCount}`);
     console.log(`  pool: ${quote.selectedPoolId}`);
+    console.log(`  launch with: --quote ${quote.id}`);
   }
   console.log("");
   console.log("Launch only after the customer accepts the quote. Streaming/data-plane cost is included in the estimate when the selected data path is stream, and realized usage may differ.");
@@ -9260,6 +9269,7 @@ function buildRunInput(options) {
     baseModel: normalizeBaseModel(options.model),
     datasetId: options.dataset,
     evalDatasetId: options.eval,
+    pricingQuoteId: options.quote,
     ...sourceOptions,
     mode: options.mode,
     compute: buildComputeSpec(options),
@@ -9290,7 +9300,7 @@ 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 });
@@ -9327,9 +9337,26 @@ 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,
@@ -9429,12 +9456,14 @@ program2.command("projects:create").requiredOption("--name <name>").option("--or
   printJson(await createClient(loadConfig()).projects.create({ name: options.name, organizationId: options.org }));
 });
 program2.command("datasets:validate").argument("<file>").description("Validate a local dataset file").action(async (file) => {
-  printJson(
-    await createClient(loadConfig()).datasets.validate({
-      path: file,
-      format: "chat_jsonl"
-    })
-  );
+  const validation = await createClient(loadConfig()).datasets.validate({
+    path: file,
+    format: "chat_jsonl"
+  });
+  printJson(validation);
+  if (!validation.valid) {
+    process.exitCode = 1;
+  }
 });
 program2.command("datasets:list").option("--project <projectId>").description("List datasets").action(async (options) => {
   printJson(await createClient(loadConfig()).datasets.list(options.project));
@@ -9452,7 +9481,7 @@ program2.command("datasets:upload").argument("<file>").option("--project <projec
   });
   printJson(dataset);
 });
-program2.command("runs:create").requiredOption("--project <projectId>").requiredOption("--dataset <datasetId>").requiredOption("--model <baseModel>").option("--eval <evalDatasetId>").option("--epochs <epochs>", "number of epochs", "3").option("--lr <lr>", "learning rate", "0.0002").option("--gpus <gpuCount>", "gpu count", "1").option("--nodes <nodeCount>", "node count", "1").option("--accelerator <acceleratorClass>", "optional hard accelerator constraint (for example: a10g, a100, h100)").option("--min-memory <gigabytes>", "minimum GPU memory in GB").option("--precision <precision>", "fp16_bf16, fp8, or fp32").option("--interconnect <interconnect>", "pcie or nvlink").option("--mode <mode>", "efficient, balanced, or power", "balanced").option("--repo <path>", "local repo path for runtime autodetect").option("--git <url>", "git repo URL for runtime metadata").option("--branch <branch>", "git branch for runtime metadata").option("--yes", "confirm that you reviewed pricing with runs:quote --summary and accept fluctuating realized usage").description("Create a training run").action(async (options) => {
+program2.command("runs:create").requiredOption("--project <projectId>").requiredOption("--dataset <datasetId>").requiredOption("--model <baseModel>").option("--eval <evalDatasetId>").option("--epochs <epochs>", "number of epochs", "3").option("--lr <lr>", "learning rate", "0.0002").option("--gpus <gpuCount>", "gpu count", "1").option("--nodes <nodeCount>", "node count", "1").option("--accelerator <acceleratorClass>", "optional hard accelerator constraint (for example: a10g, a100, h100)").option("--min-memory <gigabytes>", "minimum GPU memory in GB").option("--precision <precision>", "fp16_bf16, fp8, or fp32").option("--interconnect <interconnect>", "pcie or nvlink").option("--mode <mode>", "efficient, balanced, or power", "balanced").option("--quote <pricingQuoteId>", "launch the exact accepted quote id from runs:quote").option("--repo <path>", "local repo path for runtime autodetect").option("--git <url>", "git repo URL for runtime metadata").option("--branch <branch>", "git branch for runtime metadata").option("--yes", "confirm that you reviewed pricing with runs:quote --summary and accept fluctuating realized usage").description("Create a training run").action(async (options) => {
   if (!options.yes) {
     throw new Error("Refusing to launch without explicit cost acceptance. Run `trainfabric runs:quote --summary ...` first, then rerun `runs:create` with --yes.");
   }
@@ -9498,11 +9527,16 @@ 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));
@@ -9553,7 +9587,7 @@ program2.command("models:export").argument("<modelId>").description("Fetch the e
 program2.command("deployments:list").description("List deployments").action(async () => {
   printJson(await createClient(loadConfig()).deployments.list());
 });
-program2.command("deployments:create").requiredOption("--model <modelId>").description("Create a deployment for a model").action(async (options) => {
+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": "0.1.11",
+  "version": "0.1.13",
   "description": "Trainfabric CLI for launching GPU training jobs on the hosted Trainfabric backend.",
   "type": "module",
   "main": "dist/index.cjs",