Generate genai metadata only if ORT generate() API is used (#1332)

## Describe your changes
Generate genai metadata only if ORT generate() API is used

## Checklist before requesting a review
- [ ] Add unit tests for this change.
- [ ] Make sure all tests can pass.
- [ ] Update documents if necessary.
- [ ] Lint and apply fixes to your code by running `lintrunner -a`
- [ ] Is this a user-facing change? If yes, give a description of this
change to be included in the release notes.
- [ ] Is this PR including examples changes? If yes, please remember to
update [example
documentation](https://github.com/microsoft/Olive/blob/main/docs/source/examples.md)
in a follow-up PR.

## (Optional) Issue link

---------

Co-authored-by: Jambay Kinley <jambaykinley@microsoft.com>

examples/llama2/llama2_multilora.ipynb

@@ -98,7 +98,7 @@
     "    --train_split \"train[:4096]\" --eval_split \"train[4096:4224]\" \\\n",
     "    --text_template \"### Language: {programming_language} \\n### Question: {prompt} \\n### Answer: {response}\" \\\n",
     "    --per_device_train_batch_size 16 --per_device_eval_batch_size 16 --max_steps 150 --logging_steps 50 \\\n",
-    "    -o models/tiny-codes"
+    "    -o models/tiny-codes --use_ort_genai"
    ]
   },
   {

olive/cli/finetune.py

@@ -41,7 +41,7 @@ class FineTuneCommand(BaseOliveCLICommand):
             help="The precision of the optimized model and adapters.",
         )
 
-        # model options
+        # Model options
         model_group = sub_parser.add_argument_group("model options")
         model_group.add_argument(
             "-m",
@@ -63,7 +63,7 @@ class FineTuneCommand(BaseOliveCLICommand):
             choices=["bfloat16", "float16", "float32"],
             help="The torch dtype to use for training.",
         )
-        # dataset options
+        # Dataset options
         dataset_group = sub_parser.add_argument_group("dataset options")
         dataset_group.add_argument(
             "-d",
@@ -100,7 +100,7 @@ class FineTuneCommand(BaseOliveCLICommand):
             default=1024,
             help="Maximum sequence length for the data.",
         )
-        # lora options
+        # LoRA options
         lora_group = sub_parser.add_argument_group("lora options")
         lora_group.add_argument(
             "--method",
@@ -134,7 +134,7 @@ class FineTuneCommand(BaseOliveCLICommand):
         # TODO(jambayk): what about checkpoint_dir and resume from checkpoint support? clean checkpoint dir?
         sub_parser.add_argument("--clean", action="store_true", help="Run in a clean cache directory")
 
-        # remote options
+        # AzureML options
         remote_group = sub_parser.add_argument_group("remote options")
         remote_group.add_argument(
             "--azureml_config",
@@ -158,6 +158,9 @@ class FineTuneCommand(BaseOliveCLICommand):
             ),
         )
         # Cloud cache doesn't support azureml resources yet, only hf-id
+        model_group.add_argument(
+            "--use_ort_genai", action="store_true", help="Use OnnxRuntie generate() API to run the model"
+        )
 
         sub_parser.set_defaults(func=FineTuneCommand)
 
@@ -179,7 +182,7 @@ class FineTuneCommand(BaseOliveCLICommand):
             # need to improve the output structure of olive run
             output_path = Path(self.args.output_path)
             output_path.mkdir(parents=True, exist_ok=True)
-            hardlink_copy_dir(Path(tempdir) / "f-c-o-e-m" / "gpu-cuda_model", output_path)
+            hardlink_copy_dir(Path(tempdir) / "-".join(run_config["passes"].keys()) / "gpu-cuda_model", output_path)
 
             logger.info("Model and adapters saved to %s", output_path.resolve())
 
@@ -242,6 +245,9 @@ class FineTuneCommand(BaseOliveCLICommand):
             config["data_configs"].append(eval_data_config)
             config["passes"]["f"]["eval_data_config"] = "eval_data"
 
+        if not self.args.use_ort_genai:
+            del config["passes"]["m"]
+
         if self.args.azureml_config:
             assert self.args.azureml_cluster, "AzureML cluster must be provided if using azureml_config."
 

requirements-dev.txt

@@ -1,6 +1,6 @@
 -r requirements.txt
 black
-editorconfig-checker
+editorconfig-checker==2.7.3
 flake8
 isort
 lintrunner

test/unit_test/cli/test_cli.py

@@ -111,7 +111,7 @@ def test_finetune_command(mock_tempdir, mock_run, tmp_path):
 
     # setup
     mock_tempdir.return_value = tmpdir.resolve()
-    workflow_output_dir = tmpdir / "f-c-o-e-m" / "gpu-cuda_model"
+    workflow_output_dir = tmpdir / "f-c-o-e" / "gpu-cuda_model"
     workflow_output_dir.mkdir(parents=True)
     dummy_output = workflow_output_dir / "dummy_output"
     with open(dummy_output, "w") as f: