feat: hide optimizationType (#7160)

Signed-off-by: Hannah Zhang <hannahz@nvidia.com>

components/src/dynamo/profiler/utils/dgdr_v1beta1_types.py

@@ -52,11 +52,6 @@ class ProfilingPhase(str, Enum):
     Done = "Done"
 
 
-class OptimizationType(str, Enum):
-    Latency = "latency"
-    Throughput = "throughput"
-
-
 class SearchStrategy(str, Enum):
     Rapid = "rapid"
     Thorough = "thorough"
@@ -101,16 +96,11 @@ class WorkloadSpec(BaseModel):
 class SLASpec(BaseModel):
     """Service-level agreement targets.
 
-    Provide exactly one of:
+    Provide one of:
 
     - ``ttft`` + ``itl``: explicit latency targets (default: 2000 ms / 30 ms)
-    - ``e2eLatency``: end-to-end latency target
-    - ``optimizationType``: high-level objective without explicit numeric targets"""
+    - ``e2eLatency``: end-to-end latency target (mutually exclusive with ttft/itl)"""
 
-    optimizationType: Optional[OptimizationType] = Field(
-        default=None,
-        description="OptimizationType controls the profiling optimization strategy. Use when explicit SLA targets (ttft+itl or e2eLatency) are not known.",
-    )
     ttft: Optional[float] = Field(
         default=2000,
         description="TTFT is the Time To First Token target in milliseconds.",
@@ -125,20 +115,17 @@ class SLASpec(BaseModel):
 
     @model_validator(mode="after")
     def _validate_sla_options(self) -> "SLASpec":
-        """Ensure at most one SLA mode is active."""
+        """Ensure e2eLatency and ttft/itl are not both provided."""
         has_e2e = self.e2eLatency is not None
-        has_opt = self.optimizationType is not None
         ttft_itl_touched = (
             "ttft" in self.model_fields_set or "itl" in self.model_fields_set
         )
-        has_ttft_itl = (self.ttft is not None and self.itl is not None) and (
-            ttft_itl_touched or (not has_e2e and not has_opt)
-        )
-        options_count = sum([has_ttft_itl, has_e2e, has_opt])
-        if options_count > 1:
+        has_ttft_itl = (
+            self.ttft is not None or self.itl is not None
+        ) and ttft_itl_touched
+        if has_e2e and has_ttft_itl:
             raise ValueError(
-                "SLA must specify exactly one of: (ttft and itl), e2eLatency, "
-                "or optimizationType — not multiple."
+                "SLA must specify either (ttft and itl) or e2eLatency, not both."
             )
         if (self.ttft is not None) != (self.itl is not None):
             raise ValueError("ttft and itl must both be provided together.")

deploy/helm/charts/platform/components/operator/crds/nvidia.com_dynamographdeploymentrequests.yaml

@@ -9078,18 +9078,6 @@ spec:
                     itl:
                       description: ITL is the Inter-Token Latency target in milliseconds.
                       type: number
-                    optimizationType:
-                      allOf:
-                        - enum:
-                            - latency
-                            - throughput
-                        - enum:
-                            - latency
-                            - throughput
-                      description: |-
-                        OptimizationType controls the profiling optimization strategy.
-                        Use when explicit SLA targets (ttft+itl or e2eLatency) are not known.
-                      type: string
                     ttft:
                       description: TTFT is the Time To First Token target in milliseconds.
                       type: number

deploy/operator/api/scripts/generate_pydantic_from_go.py

@@ -53,10 +53,9 @@ _IMPORT_OVERRIDES: dict[str, tuple[str, str, bool]] = {
 _STRUCT_DOCSTRINGS: dict = {
     "SLASpec": (
         "Service-level agreement targets.\n\n"
-        "    Provide exactly one of:\n\n"
+        "    Provide one of:\n\n"
         "    - ``ttft`` + ``itl``: explicit latency targets (default: 2000 ms / 30 ms)\n"
-        "    - ``e2eLatency``: end-to-end latency target\n"
-        "    - ``optimizationType``: high-level objective without explicit numeric targets"
+        "    - ``e2eLatency``: end-to-end latency target (mutually exclusive with ttft/itl)"
     ),
 }
 
@@ -66,18 +65,15 @@ _STRUCT_EXTRAS: dict = {
     "SLASpec": """\
     @model_validator(mode="after")
     def _validate_sla_options(self) -> "SLASpec":
-        \"\"\"Ensure at most one SLA mode is active.\"\"\"
+        \"\"\"Ensure e2eLatency and ttft/itl are not both provided.\"\"\"
         has_e2e = self.e2eLatency is not None
-        has_opt = self.optimizationType is not None
-        ttft_itl_touched = "ttft" in self.model_fields_set or "itl" in self.model_fields_set
-        has_ttft_itl = (self.ttft is not None and self.itl is not None) and (
-            ttft_itl_touched or (not has_e2e and not has_opt)
+        ttft_itl_touched = (
+            "ttft" in self.model_fields_set or "itl" in self.model_fields_set
         )
-        options_count = sum([has_ttft_itl, has_e2e, has_opt])
-        if options_count > 1:
+        has_ttft_itl = (self.ttft is not None or self.itl is not None) and ttft_itl_touched
+        if has_e2e and has_ttft_itl:
             raise ValueError(
-                "SLA must specify exactly one of: (ttft and itl), e2eLatency, "
-                "or optimizationType \u2014 not multiple."
+                "SLA must specify either (ttft and itl) or e2eLatency, not both."
             )
         if (self.ttft is not None) != (self.itl is not None):
             raise ValueError("ttft and itl must both be provided together.")

deploy/operator/api/scripts/validate_pydantic_models.py

@@ -103,7 +103,6 @@ from dynamo.profiler.utils.dgdr_v1beta1_types import (  # noqa: E402
     FeaturesSpec,
     MockerSpec,
     ModelCacheSpec,
-    OptimizationType,
     PlannerConfig,
     PlannerPreDeploymentSweepMode,
     ProfilingPhase,
@@ -175,7 +174,6 @@ def test_sla_defaults_and_validation():
     assert sla.ttft == 2000.0
     assert sla.itl == 30.0
     assert sla.e2eLatency is None
-    assert sla.optimizationType is None
     print("✓ SLASpec defaults correct")
 
     # explicit ttft+itl mode: OK
@@ -184,13 +182,6 @@ def test_sla_defaults_and_validation():
     # e2eLatency mode: OK (null out ttft/itl)
     SLASpec(ttft=None, itl=None, e2eLatency=500.0)
 
-    # optimizationType mode: OK (null out ttft/itl)
-    SLASpec(ttft=None, itl=None, optimizationType=OptimizationType.Throughput)
-
-    # optimizationType mode: OK without explicitly nulling defaults (TC-2.5 / TC-2.6)
-    SLASpec(optimizationType=OptimizationType.Throughput)
-    SLASpec(optimizationType=OptimizationType.Latency)
-
     # e2eLatency mode: OK without explicitly nulling defaults
     SLASpec(e2eLatency=500.0)
 
@@ -230,10 +221,6 @@ def test_enums():
     assert ProfilingPhase.Initializing == "Initializing"
     assert ProfilingPhase.SweepingPrefill == "SweepingPrefill"
 
-    # OptimizationType — TitleCase from Go const names
-    assert OptimizationType.Latency == "latency"
-    assert OptimizationType.Throughput == "throughput"
-
     # SearchStrategy — TitleCase from Go const names
     assert SearchStrategy.Rapid == "rapid"
     assert SearchStrategy.Thorough == "thorough"

deploy/operator/api/v1alpha1/dynamographdeploymentrequest_conversion.go

@@ -80,7 +80,7 @@
 //
 // v1beta1-only fields with no v1alpha1 equivalent (omitted / TODO):
 //
-//	Hardware.*, Workload.{Concurrency,RequestRate}, SLA.{E2ELatency,OptimizationType},
+//	Hardware.*, Workload.{Concurrency,RequestRate}, SLA.{E2ELatency},
 //	Features.{KVRouter}, SearchStrategy
 //
 // # Status field mapping

deploy/operator/api/v1beta1/dynamographdeploymentrequest_types.go

@@ -165,15 +165,6 @@ const (
 	ProfilingReasonJobCreationFailed = "JobCreationFailed"
 )
 
-// OptimizationType specifies the profiling optimization strategy.
-// +kubebuilder:validation:Enum=latency;throughput
-type OptimizationType string
-
-const (
-	OptimizationTypeLatency    OptimizationType = "latency"
-	OptimizationTypeThroughput OptimizationType = "throughput"
-)
-
 // SearchStrategy controls the profiling search depth.
 // +kubebuilder:validation:Enum=rapid;thorough
 type SearchStrategy string
@@ -231,14 +222,7 @@ type WorkloadSpec struct {
 }
 
 // SLASpec defines the service-level agreement targets for profiling optimization.
-// Exactly one mode should be active: ttft+itl (default), e2eLatency, or optimizationType.
 type SLASpec struct {
-	// OptimizationType controls the profiling optimization strategy.
-	// Use when explicit SLA targets (ttft+itl or e2eLatency) are not known.
-	// +optional
-	// +kubebuilder:validation:Enum=latency;throughput
-	OptimizationType OptimizationType `json:"optimizationType,omitempty"`
-
 	// TTFT is the Time To First Token target in milliseconds.
 	// +optional
 	// +python-default=2000

deploy/operator/config/crd/bases/nvidia.com_dynamographdeploymentrequests.yaml

@@ -9078,18 +9078,6 @@ spec:
                     itl:
                       description: ITL is the Inter-Token Latency target in milliseconds.
                       type: number
-                    optimizationType:
-                      allOf:
-                        - enum:
-                            - latency
-                            - throughput
-                        - enum:
-                            - latency
-                            - throughput
-                      description: |-
-                        OptimizationType controls the profiling optimization strategy.
-                        Use when explicit SLA targets (ttft+itl or e2eLatency) are not known.
-                      type: string
                     ttft:
                       description: TTFT is the Time To First Token target in milliseconds.
                       type: number

docs/kubernetes/api-reference.md

@@ -1480,24 +1480,6 @@ _Appears in:_
 | `pvcMountPath` _string_ | PVCMountPath is the mount path for the PVC inside the container. | /opt/model-cache | Optional: \{\} <br /> |
 
 
-#### OptimizationType
-
-_Underlying type:_ _string_
-
-OptimizationType specifies the profiling optimization strategy.
-
-_Validation:_
-- Enum: [latency throughput]
-
-_Appears in:_
-- [SLASpec](#slaspec)
-
-| Field | Description |
-| --- | --- |
-| `latency` |  |
-| `throughput` |  |
-
-
 #### OverridesSpec
 
 
@@ -1579,7 +1561,6 @@ _Appears in:_
 
 
 SLASpec defines the service-level agreement targets for profiling optimization.
-Exactly one mode should be active: ttft+itl (default), e2eLatency, or optimizationType.
 
 
 
@@ -1588,7 +1569,6 @@ _Appears in:_
 
 | Field | Description | Default | Validation |
 | --- | --- | --- | --- |
-| `optimizationType` _[OptimizationType](#optimizationtype)_ | OptimizationType controls the profiling optimization strategy.<br />Use when explicit SLA targets (ttft+itl or e2eLatency) are not known. |  | Enum: [latency throughput] <br />Optional: \{\} <br /> |
 | `ttft` _float_ | TTFT is the Time To First Token target in milliseconds. |  | Optional: \{\} <br /> |
 | `itl` _float_ | ITL is the Inter-Token Latency target in milliseconds. |  | Optional: \{\} <br /> |
 | `e2eLatency` _float_ | E2ELatency is the target end-to-end request latency in milliseconds.<br />Alternative to specifying TTFT + ITL. |  | Optional: \{\} <br /> |