fix validations

Author: gwang111

src/sagemaker/serve/validations/optimization.py

@@ -15,7 +15,7 @@
 
 import textwrap
 import logging
-from typing import Any, Dict, Set
+from typing import Any, Dict, Set, Optional
 from enum import Enum
 from pydantic import BaseModel
 
@@ -34,10 +34,10 @@ class _OptimizationCombination(BaseModel):
     """Optimization ruleset data structure for comparing input to ruleset"""
 
     optimization_container: _OptimizationContainer = None
-    compilation: Set[bool | None]
-    speculative_decoding: Set[bool | None]
-    sharding: Set[bool | None]
-    quantization_technique: Set[str | None]
+    compilation: Set[Optional[bool]]
+    speculative_decoding: Set[Optional[bool]]
+    sharding: Set[Optional[bool]]
+    quantization_technique: Set[Optional[str]]
 
     def validate_against(self, optimization_combination, rule_set: _OptimizationContainer):
         """Validator for optimization containers"""
@@ -66,16 +66,7 @@ def validate_against(self, optimization_combination, rule_set: _OptimizationCont
             is_compiled = optimization_combination.compilation.copy().pop()
             is_quantized = optimization_combination.quantization_technique.copy().pop()
             if is_compiled and not is_quantized or is_quantized and not is_compiled:
-                raise ValueError("Compilation must be provided with Quantization")
-        else:
-            is_compiled = optimization_combination.compilation.copy().pop()
-            is_quantization_technique = optimization_combination.quantization_technique.copy().pop()
-            if (
-                is_compiled and is_quantization_technique
-            ):  # Check that the 2 techniques are not None
-                raise ValueError(
-                    f"Compilation and Quantization:{optimization_combination.quantization_technique.pop()}"
-                )
+                raise ValueError(f"Compilation must be provided with Quantization")
 
 
 TRUTHY_SET = {None, True}
@@ -95,7 +86,7 @@ def validate_against(self, optimization_combination, rule_set: _OptimizationCont
     "optimization_combination": _OptimizationCombination(
         optimization_container=_OptimizationContainer.VLLM,
         compilation=FALSY_SET,
-        quantization_technique={None},
+        quantization_technique={None, "awq", "fp8"},
         speculative_decoding=TRUTHY_SET,
         sharding=TRUTHY_SET,
     ),
@@ -200,7 +191,7 @@ def _validate_optimization_configuration(
                 )
                 print("fsdafas")
             except ValueError as vllm_compare_error:
-                if str(trt_compare_error) == "Compilation must be provided with Quantization":
+                if "Compilation must be provided with Quantization" in str(trt_compare_error):
                     joint_error_msg = f"""
                     Optimization cannot be performed for the following reasons:
                     - Optimizations that use {trt_compare_error} and vice-versa for GPU instances.