test(deploy): share chat and embedding API coverage

.github/actions/dynamo-deploy-test/action.yml

@@ -41,6 +41,10 @@ inputs:
     description: 'Full container image reference for the framework runtime'
     required: false
     default: ''
+  frontend_image:
+    description: 'Full container image reference for the standalone frontend'
+    required: false
+    default: ''
   platform_arch:
     description: 'Platform architecture (amd64, arm64)'
     required: false
@@ -58,7 +62,7 @@ inputs:
     required: false
     default: ''
   extra_pytest_args:
-    description: 'Additional pytest arguments (e.g., --frontend-image=...)'
+    description: 'Additional pytest arguments (e.g., -m framework_only)'
     required: false
     default: ''
   cleanup_deployment:
@@ -105,6 +109,7 @@ runs:
         FRAMEWORK: ${{ inputs.framework }}
         PROFILE: ${{ inputs.profile }}
         IMAGE: ${{ inputs.image }}
+        FRONTEND_IMAGE: ${{ inputs.frontend_image }}
         DYN_TEST_OUTPUT_PATH: ${{ github.workspace }}/test-output
         TEST_NAME: ${{ inputs.test_name }}
         TEST_FILE: ${{ inputs.test_file }}
@@ -127,6 +132,9 @@ runs:
         if [ -n "${IMAGE}" ]; then
           PYTEST_ARGS+=" --image=${IMAGE}"
         fi
+        if [ -n "${FRONTEND_IMAGE}" ]; then
+          PYTEST_ARGS+=" --frontend-image=${FRONTEND_IMAGE}"
+        fi
         if [ -n "${MODEL_CACHE_PVC}" ]; then
           PYTEST_ARGS+=" --model-cache-pvc=${MODEL_CACHE_PVC}"
         fi

.github/workflows/nightly-ci.yml

@@ -536,6 +536,18 @@ jobs:
       copy_timeout_minutes: 20
     secrets: inherit
 
+  frontend-copy-to-acr:
+    name: frontend-nightly
+    needs: [compute-release-mode, frontend-build]
+    if: ${{ needs.compute-release-mode.outputs.run_tests == 'true' }}
+    uses: $/.github/workflows/shared-copy.yml
+    with:
+      target_tag_plain: ${{ format('{0}-nightly', needs.frontend-build.outputs.target_tag_plain) }}
+      cuda_version: '[""]'
+      override_arch: amd64
+      copy_timeout_minutes: 10
+    secrets: inherit
+
   deepseek-v4-pro-agg-perf:
     name: Recipe tests
     needs: [compute-release-mode, resolve-source-sha, vllm-build, vllm-copy-to-acr]
@@ -632,41 +644,44 @@ jobs:
 
   deploy-test-vllm:
     name: vLLM Nightly Deploy Test
-    needs: [compute-release-mode, deploy-operator, vllm-build, vllm-copy-to-acr]
+    needs: [compute-release-mode, deploy-operator, vllm-build, vllm-copy-to-acr, frontend-copy-to-acr]
     if: ${{ needs.compute-release-mode.outputs.run_tests == 'true' }}
     uses: $/.github/workflows/shared-deploy-test.yml
     with:
       framework: vllm
       profiles: '["agg", "agg_router_kv_approx"]'
       image_tag: ${{ needs.vllm-copy-to-acr.outputs.image_tag }}
+      frontend_tag: ${{ needs.frontend-copy-to-acr.outputs.image_tag }}
       namespace: ${{ needs.deploy-operator.outputs.namespace }}
       vcluster_name: ${{ needs.deploy-operator.outputs.vcluster_name }}
       operator_tag: ${{ needs.deploy-operator.outputs.operator_tag }}
     secrets: inherit
 
   deploy-test-sglang:
     name: sglang Nightly Deploy Test
-    needs: [compute-release-mode, deploy-operator, sglang-build, sglang-copy-to-acr]
+    needs: [compute-release-mode, deploy-operator, sglang-build, sglang-copy-to-acr, frontend-copy-to-acr]
     if: ${{ needs.compute-release-mode.outputs.run_tests == 'true' }}
     uses: $/.github/workflows/shared-deploy-test.yml
     with:
       framework: sglang
-      profiles: '["agg", "agg_logging"]'
+      profiles: '["agg", "agg_logging", "agg_embed"]'
       image_tag: ${{ needs.sglang-copy-to-acr.outputs.image_tag }}
+      frontend_tag: ${{ needs.frontend-copy-to-acr.outputs.image_tag }}
       namespace: ${{ needs.deploy-operator.outputs.namespace }}
       vcluster_name: ${{ needs.deploy-operator.outputs.vcluster_name }}
       operator_tag: ${{ needs.deploy-operator.outputs.operator_tag }}
     secrets: inherit
 
   deploy-test-trtllm:
     name: TensorRT-LLM Nightly Deploy Test
-    needs: [compute-release-mode, deploy-operator, trtllm-build, trtllm-copy-to-acr]
+    needs: [compute-release-mode, deploy-operator, trtllm-build, trtllm-copy-to-acr, frontend-copy-to-acr]
     if: ${{ needs.compute-release-mode.outputs.run_tests == 'true' }}
     uses: $/.github/workflows/shared-deploy-test.yml
     with:
       framework: trtllm
       profiles: '["agg"]'
       image_tag: ${{ needs.trtllm-copy-to-acr.outputs.image_tag }}
+      frontend_tag: ${{ needs.frontend-copy-to-acr.outputs.image_tag }}
       namespace: ${{ needs.deploy-operator.outputs.namespace }}
       vcluster_name: ${{ needs.deploy-operator.outputs.vcluster_name }}
       operator_tag: ${{ needs.deploy-operator.outputs.operator_tag }}

.github/workflows/pr.yaml

@@ -757,6 +757,9 @@ jobs:
     name: frontend
     needs: [changed-files]
     if: |
+      needs.changed-files.outputs.vllm == 'true' ||
+      needs.changed-files.outputs.sglang == 'true' ||
+      needs.changed-files.outputs.trtllm == 'true' ||
       needs.changed-files.outputs.frontend == 'true' ||
       needs.changed-files.outputs.operator == 'true' ||
       needs.changed-files.outputs.snapshot == 'true' ||
@@ -1191,7 +1194,13 @@ jobs:
     needs: [changed-files, frontend-build]
     if: |
       always() && !cancelled() &&
-      (needs.changed-files.outputs.snapshot == 'true' ||
+      (needs.changed-files.outputs.frontend == 'true' ||
+       needs.changed-files.outputs.vllm == 'true' ||
+       needs.changed-files.outputs.sglang == 'true' ||
+       needs.changed-files.outputs.trtllm == 'true' ||
+       needs.changed-files.outputs.deploy == 'true' ||
+       needs.changed-files.outputs.operator == 'true' ||
+       needs.changed-files.outputs.snapshot == 'true' ||
       needs.changed-files.outputs.snapshot_vllm == 'true' ||
       needs.changed-files.outputs.snapshot_sglang == 'true' ||
       needs.changed-files.outputs.snapshot_trtllm == 'true') &&
@@ -1311,7 +1320,7 @@ jobs:
 
   deploy-test-vllm:
     name: vllm Deploy Test
-    needs: [changed-files, deploy-operator, vllm-copy-to-acr]
+    needs: [changed-files, deploy-operator, vllm-copy-to-acr, frontend-copy-to-acr]
     if: |
       needs.changed-files.outputs.run_deploy_tests == 'true' &&
       !cancelled() && !failure() &&
@@ -1325,14 +1334,15 @@ jobs:
       framework: vllm
       profiles: '["agg", "agg_router", "disagg", "disagg_router"]'
       image_tag: ${{ needs.vllm-copy-to-acr.outputs.image_tag }}
+      frontend_tag: ${{ needs.frontend-copy-to-acr.outputs.image_tag }}
       namespace: ${{ needs.deploy-operator.outputs.namespace }}
       vcluster_name: ${{ needs.deploy-operator.outputs.vcluster_name }}
       operator_tag: ${{ needs.deploy-operator.outputs.operator_tag }}
     secrets: inherit
 
   deploy-test-sglang:
     name: sglang Deploy Test
-    needs: [changed-files, deploy-operator, sglang-copy-to-acr]
+    needs: [changed-files, deploy-operator, sglang-copy-to-acr, frontend-copy-to-acr]
     if: |
       needs.changed-files.outputs.run_deploy_tests == 'true' &&
       !cancelled() && !failure() &&
@@ -1344,16 +1354,17 @@ jobs:
     uses: $/.github/workflows/shared-deploy-test.yml
     with:
       framework: sglang
-      profiles: '["agg", "agg_router"]'
+      profiles: '["agg", "agg_router", "agg_embed"]'
       image_tag: ${{ needs.sglang-copy-to-acr.outputs.image_tag }}
+      frontend_tag: ${{ needs.frontend-copy-to-acr.outputs.image_tag }}
       namespace: ${{ needs.deploy-operator.outputs.namespace }}
       vcluster_name: ${{ needs.deploy-operator.outputs.vcluster_name }}
       operator_tag: ${{ needs.deploy-operator.outputs.operator_tag }}
     secrets: inherit
 
   deploy-test-trtllm:
     name: trtllm Deploy Test
-    needs: [changed-files, deploy-operator, trtllm-copy-to-acr]
+    needs: [changed-files, deploy-operator, trtllm-copy-to-acr, frontend-copy-to-acr]
     if: |
       needs.changed-files.outputs.run_deploy_tests == 'true' &&
       !cancelled() && !failure() &&
@@ -1367,6 +1378,7 @@ jobs:
       framework: trtllm
       profiles: '["agg", "agg_router"]'
       image_tag: ${{ needs.trtllm-copy-to-acr.outputs.image_tag }}
+      frontend_tag: ${{ needs.frontend-copy-to-acr.outputs.image_tag }}
       namespace: ${{ needs.deploy-operator.outputs.namespace }}
       vcluster_name: ${{ needs.deploy-operator.outputs.vcluster_name }}
       operator_tag: ${{ needs.deploy-operator.outputs.operator_tag }}

.github/workflows/shared-deploy-test.yml

@@ -18,6 +18,10 @@ on:
         description: 'Runtime image tag'
         type: string
         required: true
+      frontend_tag:
+        description: 'Standalone frontend image tag'
+        type: string
+        default: ''
       namespace:
         description: 'Host namespace where the vCluster lives'
         type: string
@@ -94,6 +98,7 @@ jobs:
           profile: ${{ matrix.profile }}
           image: ${{ secrets.AZURE_ACR_HOSTNAME }}/${{ vars.ACR_REPOSITORY }}:${{ inputs.image_tag }}
           platform_arch: amd64
+          frontend_image: ${{ inputs.frontend_tag != '' && format('{0}/{1}:{2}', secrets.AZURE_ACR_HOSTNAME, vars.ACR_REPOSITORY, inputs.frontend_tag) || '' }}
           extra_pytest_args: -m framework_only
           # Mount the shared model cache only when both endpoint vars are set
           # (matches the PV/PVC creation gate); otherwise workers download from HF.

components/src/dynamo/sglang/tests/test_sglang_embedding_inputs.py

@@ -13,6 +13,7 @@
 
 from sglang.srt.managers.io_struct import EmbeddingReqInput  # noqa: E402
 
+from dynamo.sglang.health_check import SglangEmbeddingHealthCheckPayload  # noqa: E402
 from dynamo.sglang.request_handlers.embedding import (  # noqa: E402
     embedding_handler as eh,
 )
@@ -24,6 +25,7 @@
     pytest.mark.gpu_0,
     pytest.mark.profiled_vram_gib(0),
     pytest.mark.pre_merge,
+    pytest.mark.timeout(30),
 ]
 
 
@@ -61,6 +63,29 @@ def _handler(*, enable_trace: bool = True) -> eh.EmbeddingWorkerHandler:
     return handler
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("use_text_input", [True, False])
+async def test_health_check_runs_embedding_inference(monkeypatch, use_text_input):
+    monkeypatch.delenv("DYN_HEALTH_CHECK_PAYLOAD", raising=False)
+    handler = _handler()
+    payload = SglangEmbeddingHealthCheckPayload(
+        "embedding-model", use_text_input=use_text_input
+    ).to_dict()
+
+    outputs = [output async for output in handler.generate(payload, _Context())]
+
+    assert len(outputs) == 1
+    assert outputs[0]["model"] == "embedding-model"
+    assert len(outputs[0]["data"]) == 1
+    if use_text_input:
+        assert handler.engine.async_encode_calls[0]["prompt"] == "Test"
+        assert handler.engine.tokenizer_manager.requests == []
+    else:
+        [(request, _)] = handler.engine.tokenizer_manager.requests
+        assert request.input_ids == [1]
+        assert handler.engine.async_encode_calls == []
+
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize(
     ("embedding_input", "expected_request_id"),

components/src/dynamo/sglang/tests/test_sglang_health_check.py

@@ -19,12 +19,15 @@
 from dynamo.health_check import HEALTH_CHECK_KEY
 from dynamo.sglang.health_check import (
     SglangDisaggHealthCheckPayload,
+    SglangEmbeddingHealthCheckPayload,
     SglangHealthCheckPayload,
 )
+from dynamo.sglang.protocol import EmbeddingRequest
 
 pytestmark = [
     pytest.mark.unit,
     pytest.mark.sglang,
+    pytest.mark.fault_tolerance,
     pytest.mark.gpu_0,
     pytest.mark.pre_merge,
 ]
@@ -39,6 +42,30 @@ def test_decode_payload_has_no_marker():
     assert HEALTH_CHECK_KEY not in SglangHealthCheckPayload().to_dict()
 
 
+def test_embedding_payload_uses_engine_bos_token(monkeypatch):
+    monkeypatch.delenv("DYN_HEALTH_CHECK_PAYLOAD", raising=False)
+    engine = SimpleNamespace(
+        tokenizer_manager=SimpleNamespace(tokenizer=SimpleNamespace(bos_token_id=42))
+    )
+    payload = SglangEmbeddingHealthCheckPayload(
+        "embedding-model", engine, use_text_input=False
+    ).to_dict()
+
+    request = EmbeddingRequest(**payload)
+    assert request.model == "embedding-model"
+    assert request.input == [42]
+
+
+def test_embedding_env_override(monkeypatch):
+    override = {"model": "embedding-model", "input": "custom probe"}
+    monkeypatch.setenv("DYN_HEALTH_CHECK_PAYLOAD", json.dumps(override))
+
+    payload = SglangEmbeddingHealthCheckPayload("embedding-model").to_dict()
+
+    assert payload == override
+    assert EmbeddingRequest(**payload).input == "custom probe"
+
+
 def test_disagg_env_override_preserves_marker(monkeypatch):
     """DYN_HEALTH_CHECK_PAYLOAD must not drop the canary marker."""
     monkeypatch.setenv(

docs/fern/pages/recipes/kubernetes-templates/dgd/sglang.mdx

@@ -22,6 +22,19 @@ kubectl apply -f agg.yaml
 <Accordion title="agg.yaml · Baseline aggregated serving">
 <Code src="../../../../../../examples/backends/sglang/deploy/agg.yaml" title="agg.yaml" language="yaml" maxLines={0} />
 </Accordion>
+<Accordion title="agg_embed.yaml · Aggregated embedding serving">
+This Kubernetes template uses `Qwen/Qwen3-Embedding-0.6B` for the single-GPU CI profile; the
+[local CLI example](../../cli-templates/sglang.mdx) defaults to `Qwen/Qwen3-Embedding-4B`.
+
+<Code src="../../../../../../examples/backends/sglang/deploy/agg_embed.yaml" title="agg_embed.yaml" language="yaml" maxLines={0} />
+
+After forwarding the Frontend service to `localhost:8000`, check the `/v1/embeddings` endpoint:
+
+```bash
+curl http://localhost:8000/v1/embeddings -H 'Content-Type: application/json' \
+  -d '{"model": "Qwen/Qwen3-Embedding-0.6B", "input": "Hello world"}'
+```
+</Accordion>
 <Accordion title="agg_router.yaml · Aggregated with KV-aware routing">
 <Code src="../../../../../../examples/backends/sglang/deploy/agg_router.yaml" title="agg_router.yaml" language="yaml" maxLines={0} />
 </Accordion>

examples/backends/sglang/deploy/agg_embed.yaml

@@ -0,0 +1,51 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+apiVersion: nvidia.com/v1beta1
+kind: DynamoGraphDeployment
+metadata:
+  name: sglang-agg-embed
+spec:
+  components:
+  - name: Frontend
+    podTemplate:
+      spec:
+        containers:
+        - image: my-registry/sglang-runtime:my-tag
+          name: main
+    replicas: 1
+    type: frontend
+  - name: decode
+    podTemplate:
+      spec:
+        containers:
+        - args:
+          - --embedding-worker
+          - --use-sglang-tokenizer
+          - --model-path
+          - Qwen/Qwen3-Embedding-0.6B
+          - --served-model-name
+          - Qwen/Qwen3-Embedding-0.6B
+          - --page-size
+          - "16"
+          - --tp
+          - "1"
+          - --trust-remote-code
+          command:
+          - python3
+          - -m
+          - dynamo.sglang
+          envFrom:
+          - secretRef:
+              name: hf-token-secret
+          image: my-registry/sglang-runtime:my-tag
+          name: main
+          resources:
+            limits:
+              nvidia.com/gpu: "1"
+            requests:
+              # Increase this value for larger models.
+              ephemeral-storage: 2Gi
+          workingDir: /workspace/examples/backends/sglang
+    replicas: 1
+    type: worker