badgr-cli 1.1.5 → 1.1.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "badgr-cli",
-  "version": "1.1.5",
+  "version": "1.1.6",
   "description": "Badgr — run or serve GPU workloads from one command",
   "type": "module",
   "bin": {

package/src/commands/serve.js

@@ -151,7 +151,12 @@ function _resolveHealthPath({ healthPath, isLlamaCpp, customImage, task }) {
  * directly. The backend does the real health_path probing (see
  * DeploymentService.check_endpoint_readiness); a pod reporting RUNNING only
  * means infrastructure is up, not that the app inside is serving.
- * Returns { ready: boolean, timedOut: boolean, depFailed: boolean, failReason?: string }
+ * Returns { ready: boolean, timedOut: boolean, depFailed: boolean, failReason?: string,
+ *           dep? }. `dep` (only present when ready) is the last polled
+ * /deployments/{id} response — carries supports_completions/
+ * supports_chat_completions, probed server-side once the endpoint first
+ * reports ready, so the caller's final usage example can match what the
+ * model actually serves instead of guessing.
  */
 // vLLM cold start (model download + load) often exceeds 5 min on first boot.
 const VLLM_SERVE_WAIT_MS = 15 * 60 * 1000;
@@ -183,7 +188,7 @@ async function waitForEndpoint(deploymentId, config, timeoutMs = VLLM_SERVE_WAIT
       }
       if (dep.endpoint_ready) {
         if (blockLines > 0) { _clearBlock(blockLines); blockLines = 0; }
-        return { ready: true, timedOut: false, depFailed: false };
+        return { ready: true, timedOut: false, depFailed: false, dep };
       }
 
       const now     = Date.now();
@@ -793,6 +798,11 @@ export async function serveCommand(config, args, chalk) {
     if (endpointReady) {
       console.log(chalk.dim(_stage(healthStageN, STAGE_TOTAL, 'Checking endpoint health...')));
       console.log(chalk.green(_stage(readyStageN, STAGE_TOTAL, 'Ready')));
+      if (healthResult.dep) {
+        dep.supports_completions = healthResult.dep.supports_completions;
+        dep.supports_chat_completions = healthResult.dep.supports_chat_completions;
+        dep.capability_source = healthResult.dep.capability_source;
+      }
     } else {
       updateReceipt(rcptId, { status: 'health_check_timeout' });
     }
@@ -820,7 +830,14 @@ export async function serveCommand(config, args, chalk) {
     console.log();
   }
 
-  console.log(`  ${chalk.bold('Base URL:')}     ${chalk.cyan(endpointUrl)}`);
+  // dep.service_url is the Badgr-owned https://<id>.serve.aibadgr.com URL
+  // once that proxy is actually live (see backend/service_proxy.py) --
+  // until then it's absent/equal to endpointUrl, so this falls back to the
+  // raw provider URL exactly as before. Only the customer-facing display
+  // and usage examples switch to it; polling/comfy-node validation above
+  // still talk to the real internal endpointUrl directly.
+  const displayUrl = dep.service_url || endpointUrl;
+  console.log(`  ${chalk.bold('Base URL:')}     ${chalk.cyan(displayUrl)}`);
   if (isLlamaCpp) {
     console.log(`  ${chalk.bold('HF Repo:')}      ${flags.hfRepo}`);
     console.log(`  ${chalk.bold('HF File:')}      ${flags.hfFile}`);
@@ -843,7 +860,7 @@ export async function serveCommand(config, args, chalk) {
 
   if (endpointReady && isOllama) {
     console.log(`  ${chalk.bold('Test with curl:')}`);
-    console.log(chalk.dim(`    curl ${endpointUrl}/api/generate \\`));
+    console.log(chalk.dim(`    curl ${displayUrl}/api/generate \\`));
     console.log(chalk.dim(`      -H "Content-Type: application/json" \\`));
     console.log(chalk.dim(`      -d '{"model":"${dep.model || effectiveModel}","prompt":"Hello","stream":false}'`));
     console.log();
@@ -860,13 +877,45 @@ export async function serveCommand(config, args, chalk) {
       console.log(`  ${chalk.bold('API key:')}      ${chalk.yellow(dep.endpoint_api_key)}`);
       console.log(chalk.dim('    Shown once — copy it now. This key is scoped to this endpoint only.'));
     }
+    // dep.supports_chat_completions is probed server-side (see backend
+    // model_capabilities.py): a live call against this exact endpoint is
+    // authoritative when it's reachable (dep.capability_source ===
+    // "runtime"); otherwise it falls back to a best-effort hint from the
+    // model's public HF chat-template config (capability_source ===
+    // "hf_hint"). `true`/`false` here always mean one of those two actually
+    // produced a real signal -- a model without a chat template (e.g.
+    // facebook/opt-125m) answers /v1/completions but 400s on
+    // /v1/chat/completions, so a generic chat example would be wrong.
+    // `undefined`/`null` means NEITHER produced a signal (auth-protected
+    // endpoint + HF lookup failed, or a non-vLLM/task image) -- unlike this
+    // command's old behavior, that must never silently default to showing
+    // a chat example that might not work; show neither example instead.
+    const noTask = !flags.task;
+    const chatKnown = dep.supports_chat_completions === true || dep.supports_chat_completions === false;
+    const showChatExample = noTask && chatKnown && dep.supports_chat_completions === true;
+    const showCompletionsExample = noTask && chatKnown && dep.supports_chat_completions === false;
+    const capabilityUnverified = noTask && !chatKnown;
+    const unconfirmedNote = dep.capability_source === 'hf_hint'
+      ? chalk.dim('    (based on the model\'s public Hugging Face config — not confirmed against this live endpoint)')
+      : null;
     console.log(`  ${chalk.bold('Test with curl:')}`);
-    console.log(chalk.dim(`    curl ${endpointUrl}/chat/completions \\`));
-    console.log(chalk.dim(`      -H "Authorization: Bearer ${authKey}" -H "Content-Type: application/json" \\`));
-    console.log(chalk.dim(`      -d '{"model":"${sdkModel}","messages":[{"role":"user","content":"Hello"}]}'`));
+    if (showChatExample) {
+      console.log(chalk.dim(`    curl ${displayUrl}/chat/completions \\`));
+      console.log(chalk.dim(`      -H "Authorization: Bearer ${authKey}" -H "Content-Type: application/json" \\`));
+      console.log(chalk.dim(`      -d '{"model":"${sdkModel}","messages":[{"role":"user","content":"Hello"}]}'`));
+      if (unconfirmedNote) console.log(unconfirmedNote);
+    } else if (showCompletionsExample) {
+      console.log(chalk.dim(`    curl ${displayUrl}/completions \\`));
+      console.log(chalk.dim(`      -H "Authorization: Bearer ${authKey}" -H "Content-Type: application/json" \\`));
+      console.log(chalk.dim(`      -d '{"model":"${sdkModel}","prompt":"Hello","max_tokens":32}'`));
+      if (unconfirmedNote) console.log(unconfirmedNote);
+    } else if (capabilityUnverified) {
+      console.log(chalk.yellow('    Capability not verified yet — check available models first:'));
+      console.log(chalk.dim(`    curl ${displayUrl}/models -H "Authorization: Bearer ${authKey}"`));
+    }
     console.log(`  ${chalk.bold('Use with OpenAI SDK:')}`);
     console.log(chalk.dim(`    from openai import OpenAI`));
-    console.log(chalk.dim(`    client = OpenAI(base_url="${endpointUrl}", api_key="${authKey}")`));
+    console.log(chalk.dim(`    client = OpenAI(base_url="${displayUrl}", api_key="${authKey}")`));
     if (flags.task === 'transcribe') {
       console.log(chalk.dim(`    with open("audio.mp3", "rb") as f:`));
       console.log(chalk.dim(`        t = client.audio.transcriptions.create(model="${sdkModel}", file=f, response_format="text")`));
@@ -875,8 +924,13 @@ export async function serveCommand(config, args, chalk) {
       console.log(chalk.dim(`    # resp.data[0].b64_json contains the base64-encoded PNG`));
     } else if (flags.task === 'embed') {
       console.log(chalk.dim(`    resp = client.embeddings.create(model="${sdkModel}", input=["hello world"])`));
-    } else {
+    } else if (showChatExample) {
       console.log(chalk.dim(`    resp = client.chat.completions.create(model="${sdkModel}", messages=[{"role": "user", "content": "Hello"}])`));
+    } else if (showCompletionsExample) {
+      console.log(chalk.dim(`    resp = client.completions.create(model="${sdkModel}", prompt="Hello", max_tokens=32)`));
+      console.log(chalk.dim(`    # this model has no chat template — /v1/chat/completions returns 400 for it`));
+    } else {
+      console.log(chalk.dim(`    resp = client.models.list()  # check capability_unverified — chat vs completions support isn't confirmed yet`));
     }
     console.log();
     if (flags.idleTimeout) {