trainfabric 0.1.45 → 0.1.47

package/dist/index.cjs

@@ -7648,6 +7648,7 @@ var runCreateSchema = external_exports.object({
   runtimeManifestId: external_exports.string().min(1).optional(),
   profileSessionId: external_exports.string().min(1).optional(),
   pricingQuoteId: external_exports.string().min(1).optional(),
+  maxCostUsd: external_exports.number().positive().max(1e5).optional(),
   experimentTracking: experimentTrackingInputSchema.optional()
 });
 var jobProfileSchema = runCreateSchema;
@@ -7693,6 +7694,7 @@ var runQuoteSchema = external_exports.object({
   hyperparameters: runHyperparametersSchema.optional(),
   runtimeVersion: external_exports.string().min(1).optional(),
   mode: external_exports.enum(trainingModes).optional(),
+  maxCostUsd: external_exports.number().positive().max(1e5).optional(),
   reservationClass: external_exports.enum(reservationClasses).optional(),
   slaClass: external_exports.enum(slaClasses).optional(),
   runtimeManifestId: external_exports.string().min(1).optional(),
@@ -7762,11 +7764,20 @@ var serviceAccountCreateSchema = external_exports.object({
   scopes: external_exports.array(external_exports.enum(apiKeyScopes)).min(1)
 });
 var billingCheckoutSessionCreateSchema = external_exports.object({
-  amountUsd: external_exports.number().positive().max(1e5).optional(),
+  amountUsd: external_exports.number().min(5).max(1e5).optional(),
   mode: external_exports.enum(["subscription", "credits"]).default("subscription"),
   planId: external_exports.enum(["hobby", "pro"]).optional(),
   priceId: external_exports.string().min(1).optional()
 });
+var couponRedeemSchema = external_exports.object({
+  code: external_exports.string().trim().min(1).max(64)
+});
+var creditAdjustmentCreateSchema = external_exports.object({
+  amountUsd: external_exports.number().min(-1e5).max(1e5).refine((value) => value !== 0, "Credit adjustment amount cannot be zero."),
+  kind: external_exports.enum(["manual_adjustment", "refund_credit", "refund_debit"]),
+  reason: external_exports.string().trim().min(1).max(500),
+  sourceId: external_exports.string().trim().min(1).max(128).optional()
+});
 var treasuryCounterpartyCreateSchema = external_exports.object({
   providerId: external_exports.enum(treasuryProviderIds).default("mercury"),
   name: external_exports.string().min(1),
@@ -8613,7 +8624,27 @@ function encodeRunId(runId) {
   }
   return encodeURIComponent(id);
 }
+function isTransientQuoteError(error) {
+  const message = error instanceof Error ? error.message : String(error);
+  return /status (502|503|504|429)\b|fetch failed|network|ECONNRESET|ETIMEDOUT/i.test(message);
+}
 var RunsClient = class extends ResourceClient {
+  async retryQuoteOperation(operation) {
+    const attempts = Math.max(1, Number(process.env.TRAINFABRIC_QUOTE_RETRY_ATTEMPTS ?? 5));
+    let lastError;
+    for (let attempt = 1; attempt <= attempts; attempt += 1) {
+      try {
+        return await operation();
+      } catch (error) {
+        lastError = error;
+        if (!isTransientQuoteError(error) || attempt === attempts) {
+          throw error;
+        }
+        await wait(Math.min(5e3, 400 * 2 ** (attempt - 1)));
+      }
+    }
+    throw lastError instanceof Error ? lastError : new Error("Quote request failed.");
+  }
   describeEmptyQuoteFailure(input) {
     const compute = input.compute && input.compute !== "auto" && input.compute !== "cpu-sim" ? input.compute : void 0;
     const requestedAccelerator = compute?.acceleratorClass;
@@ -8628,7 +8659,15 @@ var RunsClient = class extends ResourceClient {
     if (constraints.length === 0) {
       return "No quote options were returned for the requested run.";
     }
-    return `No quote options were returned for the requested run under the current hardware constraints (${constraints.join(", ")}). Omit the accelerator constraint or request a larger GPU memory floor.`;
+    const suggestions = [];
+    if (requestedAccelerator) {
+      suggestions.push("omit the accelerator constraint");
+    }
+    if (requestedMemory) {
+      suggestions.push("lower --min-memory if it excludes available GPUs");
+    }
+    suggestions.push("try a different mode or retry when capacity changes");
+    return `No quote options were returned for the requested run under the current hardware constraints (${constraints.join(", ")}). ${suggestions.join(", ")}.`;
   }
   profile(input) {
     const payload = jobProfileSchema.parse({
@@ -8664,26 +8703,29 @@ var RunsClient = class extends ResourceClient {
     if (profileSessionId) {
       return this.waitForProfile(profileSessionId);
     }
-    const session = await this.profile(input);
-    return this.waitForProfile(session.id);
+    const session = await this.retryQuoteOperation(() => this.profile(input));
+    return this.retryQuoteOperation(() => this.waitForProfile(session.id));
   }
   async quote(input) {
-    const detail = await this.ensureReadyProfile(input);
+    const detail = await this.retryQuoteOperation(() => this.ensureReadyProfile(input));
     const payload = runQuotesSchema.parse({
       ...input,
       projectId: input.projectId ?? this.parent.projectId,
       profileSessionId: detail.session.id
     });
-    return this.requestPost("/v1/runs/quotes", payload);
+    return this.requestQuoteWithRetry("/v1/runs/quotes", payload);
   }
   async quoteLegacy(input) {
-    const detail = await this.ensureReadyProfile(input);
+    const detail = await this.retryQuoteOperation(() => this.ensureReadyProfile(input));
     const payload = runQuoteSchema.parse({
       ...input,
       projectId: input.projectId ?? this.parent.projectId,
       profileSessionId: detail.session.id
     });
-    return this.requestPost("/v1/runs/quote", payload);
+    return this.requestQuoteWithRetry("/v1/runs/quote", payload);
+  }
+  async requestQuoteWithRetry(pathname, payload) {
+    return this.retryQuoteOperation(() => this.requestPost(pathname, payload));
   }
   async create(input) {
     const projectId = input.projectId ?? this.parent.projectId;
@@ -9269,6 +9311,16 @@ function normalizeLearningRate(lr) {
   }
   return parsed;
 }
+function normalizeMaxCost(maxCost) {
+  if (maxCost === void 0) {
+    return void 0;
+  }
+  const parsed = Number(maxCost);
+  if (!Number.isFinite(parsed) || parsed <= 0 || parsed > 1e5) {
+    throw new Error("Max cost must be a positive USD amount up to 100000.");
+  }
+  return Number(parsed.toFixed(2));
+}
 function buildComputeSpec(options) {
   const gpuCount = normalizePositiveInteger(options.gpus, "GPU count", 1, 64);
   const nodeCount = normalizePositiveInteger(options.nodes, "Node count", 1, 16);
@@ -9310,7 +9362,7 @@ function buildComputeSpec(options) {
 
 // src/index.ts
 var DEFAULT_TRAINFABRIC_API_URL2 = "https://api.trainfabric.com";
-var CLI_VERSION = "0.1.45";
+var CLI_VERSION = "0.1.47";
 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");
@@ -9749,6 +9801,7 @@ function buildRunInput(options, config = loadConfig()) {
     pricingQuoteId: quoteId,
     evalDatasetId: normalizeOptionalId(options.eval, "dataset", "Eval dataset ID"),
     ...sourceOptions,
+    maxCostUsd: normalizeMaxCost(options.maxCost),
     mode: normalizeTrainingMode(options.mode)
   };
   if (projectId) {
@@ -9787,10 +9840,18 @@ function assertQuoteOptions(bundle, options) {
   if (constraints.length === 0) {
     throw new Error("No quote options were returned for the requested run.");
   }
+  const suggestions = [];
+  if (options.accelerator) {
+    suggestions.push("omit --accelerator");
+  }
+  if (options.minMemory) {
+    suggestions.push("adjust --min-memory if it excludes available GPUs");
+  }
+  suggestions.push("try a different --mode or retry when capacity changes");
   throw new Error(
     `No quote options were returned for the requested run under the current hardware constraints (${constraints.join(
       ", "
-    )}). Try a different --mode, omit --accelerator, adjust --min-memory if it excludes available GPUs, or retry when capacity changes.`
+    )}). ${suggestions.join(", ")}.`
   );
 }
 function filterQuoteBundleForRequestedMode(bundle, options) {
@@ -9843,7 +9904,8 @@ async function watchRun(runId, json = false, timeoutMs, pollMs) {
     });
     handle.on("log", (event) => {
       const payload = event.payload;
-      console.log(`[${payload.phase ?? "run"}:${payload.level ?? "info"}] ${payload.message ?? ""}`);
+      const eventLabel = formatRunEventCode(payload.eventCode);
+      console.log(`[${eventLabel ?? payload.phase ?? "run"}:${payload.level ?? "info"}] ${payload.message ?? ""}`);
     });
     handle.on("metric", (event) => {
       const payload = event.payload;
@@ -9874,6 +9936,24 @@ async function watchRun(runId, json = false, timeoutMs, pollMs) {
     );
   }
 }
+function formatRunEventCode(eventCode) {
+  switch (eventCode) {
+    case "provider_waiting_for_ssh":
+      return "provider-waiting";
+    case "relay_confirmation_timeout_nonfatal":
+      return "relay-monitoring";
+    case "dependency_installing":
+      return "dependency-installing";
+    case "runtime_bootstrap_fallback":
+      return "runtime-bootstrap";
+    case "training_started":
+      return "training-started";
+    case "artifact_syncing":
+      return "artifact-syncing";
+    default:
+      return void 0;
+  }
+}
 function parseDurationMs(value) {
   if (!value) {
     return void 0;
@@ -10082,7 +10162,7 @@ program2.command("datasets:upload").argument("<file>").option("--project <projec
   });
   printJson(dataset);
 });
-program2.command("runs:create").option("--project <projectId>").option("--dataset <datasetId>").option("--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").option("--quote <pricingQuoteId>", "launch the exact accepted quote id from runs:quote; use only with --yes").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. When using --quote, do not pass project, dataset, model, compute, mode, source, or hyperparameter flags; launch with `trainfabric runs:create --quote <quoteId> --yes`.").action(async (options, command) => {
+program2.command("runs:create").option("--project <projectId>").option("--dataset <datasetId>").option("--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").option("--max-cost <usd>", "maximum accepted quoted customer charge in USD").option("--quote <pricingQuoteId>", "launch the exact accepted quote id from runs:quote; use only with --yes").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. When using --quote, do not pass project, dataset, model, compute, mode, source, or hyperparameter flags; launch with `trainfabric runs:create --quote <quoteId> --yes`.").action(async (options, command) => {
   const config = loadConfig();
   assertQuoteCreateHasNoLaunchOverrides(options, command);
   const runInput = buildRunInput(options, config);
@@ -10129,7 +10209,7 @@ program2.command("runtime:build").option("--project <projectId>").requiredOption
     )
   );
 });
-program2.command("runs:quote").option("--project <projectId>").requiredOption("--dataset <datasetId>").requiredOption("--model <baseModel>").option("--eval <evalDatasetId>").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").option("--repo <path>").option("--git <url>").option("--branch <branch>").option("--json", "print raw JSON output; this is the default unless --summary is used").option("--summary", "print a human-readable price summary instead of raw JSON").description("Preview Efficient, Balanced, and Power launch options").action(async (options) => {
+program2.command("runs:quote").option("--project <projectId>").requiredOption("--dataset <datasetId>").requiredOption("--model <baseModel>").option("--eval <evalDatasetId>").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").option("--max-cost <usd>", "maximum accepted quoted customer charge in USD").option("--repo <path>").option("--git <url>").option("--branch <branch>").option("--json", "print raw JSON output; this is the default unless --summary is used").option("--summary", "print a human-readable price summary instead of raw JSON").description("Preview Efficient, Balanced, and Power launch options").action(async (options) => {
   const config = loadConfig();
   const bundle = filterQuoteBundleForRequestedMode(
     await createClient(config).runs.quote(buildRunInput(options, config)),

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "trainfabric",
-  "version": "0.1.45",
+  "version": "0.1.47",
   "description": "Trainfabric CLI for launching GPU training jobs on the hosted Trainfabric backend.",
   "type": "module",
   "main": "dist/index.cjs",