[QAIRT] Implement QAIRT ORT->Genie workflow (microsoft#2358)

## Describe your changes

Implements a complete QAIRT workflow for converting ONNX Runtime models
to Genie-compatible format through three new passes:

**QairtPreparation**: Executes external preparation scripts to quantize
and prepare HuggingFace models for QAIRT, with configurable caching and
script parameters.

**QairtGenAIBuilder**: Converts prepared models using QAIRT GenAIBuilder
API with support for:
- CPU and HTP backend targets
- Device-specific optimizations (VTCM size, HVX threads, extended UDMA)
- Model configurations (sequence lengths, multi-graph, model splits)

**QairtEncapsulation**: Wraps QAIRT DLC models in ONNX protobuf format
with EPContext nodes, generating genai_config.json for onnxruntime-genai
compatibility.

This enables end-to-end optimization of generative AI models for
Qualcomm hardware accelerators.

## Checklist before requesting a review
- [x] Add unit tests for this change.
- [x] Make sure all tests can pass.
- [x] Update documents if necessary.
- [x] Lint and apply fixes to your code by running `lintrunner -a`
- [x] Is this a user-facing change? If yes, give a description of this
change to be included in the release notes.

## (Optional) Issue link

Author: qti-kromero

olive/constants.py

@@ -15,6 +15,7 @@ class Framework(StrEnumBase):
 
     ONNX = "ONNX"
     PYTORCH = "PyTorch"
+    QAIRT = "QAIRT"
     QNN = "QNN"
     TENSORFLOW = "TensorFlow"
     OPENVINO = "OpenVINO"
@@ -30,6 +31,8 @@ class ModelFileFormat(StrEnumBase):
     PYTORCH_SLICE_GPT_MODEL = "PyTorch.SliceGPT"
     TENSORFLOW_PROTOBUF = "TensorFlow.Protobuf"
     TENSORFLOW_SAVED_MODEL = "TensorFlow.SavedModel"
+    QAIRT = "QAIRT"
+    QAIRT_PREPARED = "QAIRT.Prepared"
     QNN_CPP = "QNN.CPP"
     QNN_LIB = "QNN.LIB"
     QNN_SERIALIZED_BIN = "QNN.SERIALIZED.BIN"

olive/model/handler/__init__.py

@@ -9,6 +9,7 @@
 from olive.model.handler.onnx import DistributedOnnxModelHandler, ONNXModelHandler
 from olive.model.handler.openvino import OpenVINOModelHandler
 from olive.model.handler.pytorch import PyTorchModelHandler
+from olive.model.handler.qairt import QairtModelHandler, QairtPreparedModelHandler
 from olive.model.handler.qnn import QNNModelHandler
 from olive.model.handler.tensorflow import TensorFlowModelHandler
 
@@ -23,5 +24,7 @@
     "OpenVINOModelHandler",
     "PyTorchModelHandler",
     "QNNModelHandler",
+    "QairtModelHandler",
+    "QairtPreparedModelHandler",
     "TensorFlowModelHandler",
 ]

olive/model/handler/qairt.py

@@ -0,0 +1,105 @@
+# -------------------------------------------------------------------------
+# Copyright (c) Qualcomm Technologies, Inc. and/or its subsidiaries.
+# SPDX-License-Identifier: MIT
+# --------------------------------------------------------------------------
+
+import logging
+from typing import Any, Callable, Optional, Union
+
+from olive.constants import Framework, ModelFileFormat
+from olive.hardware.accelerator import Device
+from olive.model.config import IoConfig
+from olive.model.config.registry import model_handler_registry
+from olive.model.handler.base import OliveModelHandler
+
+logger = logging.getLogger(__name__)
+
+
+@model_handler_registry("QairtPreparedModel")
+class QairtPreparedModelHandler(OliveModelHandler):
+    json_config_keys: tuple[str, ...] = ("io_config", "model_file_format")
+
+    def __init__(
+        self,
+        model_path: str,
+        model_attributes: Optional[dict[str, Any]] = None,
+        io_config: Union[dict[str, Any], IoConfig, str, Callable] = None,
+        model_file_format: ModelFileFormat = ModelFileFormat.QAIRT_PREPARED,
+    ):
+        super().__init__(
+            framework=Framework.QAIRT,
+            model_file_format=model_file_format,
+            model_path=model_path,
+            model_attributes=model_attributes,
+            io_config=io_config,
+        )
+
+    @property
+    def size_on_disk(self) -> int:
+        """Compute size of the model on disk."""
+        return 0
+
+    def load_model(self, rank: int = None, cache_model: bool = True):
+        raise NotImplementedError("QairtPreparedModelHandler does not support load_model")
+
+    def prepare_session(
+        self,
+        inference_settings: Union[dict[str, Any], None] = None,
+        device: Device = Device.CPU,
+        execution_providers: Union[str, list[str]] = None,
+        rank: Union[int, None] = None,
+    ):
+        raise NotImplementedError("QairtPreparedModelHandler does not support prepare_session")
+
+    def run_session(
+        self,
+        session: Any = None,
+        inputs: Union[dict[str, Any], list[Any], tuple[Any, ...]] = None,
+        **kwargs: dict[str, Any],
+    ) -> Any:
+        raise NotImplementedError("QairtPreparedModelHandler does not support prepare_session")
+
+
+@model_handler_registry("QairtModel")
+class QairtModelHandler(OliveModelHandler):
+    json_config_keys: tuple[str, ...] = ("io_config", "model_file_format")
+
+    def __init__(
+        self,
+        model_path: str,
+        model_attributes: Optional[dict[str, Any]] = None,
+        io_config: Union[dict[str, Any], IoConfig, str, Callable] = None,
+        model_file_format: ModelFileFormat = ModelFileFormat.QAIRT,
+    ):
+        super().__init__(
+            framework=Framework.QAIRT,
+            model_file_format=model_file_format,
+            model_path=model_path,
+            model_attributes=model_attributes,
+            io_config=io_config,
+        )
+
+    @property
+    def size_on_disk(self) -> int:
+        """Compute size of the model on disk."""
+        return 0
+
+    def load_model(self, rank: int = None, cache_model: bool = True):
+        raise NotImplementedError("QairtModelHandler does not support load_model")
+
+    def prepare_session(
+        self,
+        inference_settings: Union[dict[str, Any], None] = None,
+        device: Device = Device.CPU,
+        execution_providers: Union[str, list[str]] = None,
+        rank: Union[int, None] = None,
+    ):
+        raise NotImplementedError("QairtModelHandler does not support prepare_session")
+
+    def run_session(
+        self,
+        session: Any = None,
+        inputs: Union[dict[str, Any], list[Any], tuple[Any, ...]] = None,
+        **kwargs: dict[str, Any],
+    ) -> Any:
+        raise NotImplementedError("QairtModelHandler does not support prepare_session")

olive/olive_config.json

@@ -473,13 +473,41 @@
             "supported_algorithms": [  ],
             "supported_quantization_encodings": [  ]
         },
+        "QairtEncapsulation": {
+            "module_path": "olive.passes.qairt.encapsulation.QairtEncapsulation",
+            "supported_providers": [ "QNNExecutionProvider" ],
+            "supported_accelerators": [ "npu" ],
+            "supported_precisions": [ "*" ],
+            "supported_algorithms": [  ],
+            "supported_quantization_encodings": [  ],
+            "extra_dependencies": [ "qairt-dev" ]
+        },
+        "QairtGenAIBuilder": {
+            "module_path": "olive.passes.qairt.gen_ai_builder.QairtGenAIBuilder",
+            "supported_providers": [ "QNNExecutionProvider" ],
+            "supported_accelerators": [ "npu" ],
+            "supported_precisions": [ "*" ],
+            "supported_algorithms": [  ],
+            "supported_quantization_encodings": [  ],
+            "extra_dependencies": [ "qairt-dev" ]
+        },
         "QairtMHA2SHA": {
             "module_path": "olive.passes.onnx.qairt.mha2sha.QairtMHA2SHA",
             "supported_providers": [ "QNNExecutionProvider" ],
             "supported_accelerators": [ "npu" ],
             "supported_precisions": [ "*" ],
             "supported_algorithms": [  ],
-            "supported_quantization_encodings": [  ]
+            "supported_quantization_encodings": [  ],
+            "extra_dependencies": [ "qairt-dev" ]
+        },
+        "QairtPreparation": {
+            "module_path": "olive.passes.qairt.preparation.QairtPreparation",
+            "supported_providers": [ "QNNExecutionProvider" ],
+            "supported_accelerators": [ "npu" ],
+            "supported_precisions": [ "*" ],
+            "supported_algorithms": [  ],
+            "supported_quantization_encodings": [  ],
+            "extra_dependencies": [ "qairt-dev" ]
         },
         "QLoRA": {
             "module_path": "olive.passes.pytorch.lora.QLoRA",
@@ -661,6 +689,7 @@
             "onnxruntime-openvino"
         ],
         "optimum": [ "optimum" ],
+        "qairt": [ "qairt-dev[onnx]" ],
         "qnn": [ "onnxruntime-qnn" ],
         "tf": [ "tensorflow==1.15.0" ],
         "torch-tensorrt": [ "torch-tensorrt" ],

olive/passes/qairt/__init__.py

@@ -0,0 +1,4 @@
+# -------------------------------------------------------------------------
+# Copyright (c) Qualcomm Technologies, Inc. and/or its subsidiaries.
+# SPDX-License-Identifier: MIT
+# --------------------------------------------------------------------------