update import for quantization format from models

llama_toolchain/inference/api/config.py

@@ -8,6 +8,7 @@ from pydantic import BaseModel, Field
 from typing_extensions import Annotated
 
 from .datatypes import QuantizationConfig
+from llama_models.llama3_1.api.datatypes import CheckpointQuantizationFormat
 
 
 class ImplType(Enum):
@@ -20,17 +21,6 @@ class CheckpointType(Enum):
     huggingface = "huggingface"
 
 
-# This enum represents the format in which weights are specified
-# This does not necessarily always equal what quantization is desired
-# at runtime since there can be on-the-fly conversions done
-class CheckpointQuantizationFormat(Enum):
-    # default format
-    bf16 = "bf16"
-
-    # used for enabling fp8_rowwise inference, some weights are bf16
-    fp8_mixed = "fp8_mixed"
-
-
 class PytorchCheckpoint(BaseModel):
     checkpoint_type: Literal[CheckpointType.pytorch.value] = (
         CheckpointType.pytorch.value

llama_toolchain/models/api/endpoints.py

@@ -0,0 +1,9 @@
+from typing import Protocol 
+
+from pyopenapi import webmethod
+
+from pydantic import BaseModel
+
+
+class Models(Protocol):
+    ...