From 1d94f3617ad2c4a85743a74d69fbdf25af3349b4 Mon Sep 17 00:00:00 2001 From: Jash Gulabrai Date: Tue, 20 May 2025 09:43:51 -0400 Subject: [PATCH] fix: Pass model param as configuration name to NeMo Customizer --- .../providers/remote/post_training/nvidia/post_training.py | 7 ++----- tests/unit/providers/nvidia/test_supervised_fine_tuning.py | 6 +++--- 2 files changed, 5 insertions(+), 8 deletions(-) diff --git a/llama_stack/providers/remote/post_training/nvidia/post_training.py b/llama_stack/providers/remote/post_training/nvidia/post_training.py index 409818cb3..d839ffd6f 100644 --- a/llama_stack/providers/remote/post_training/nvidia/post_training.py +++ b/llama_stack/providers/remote/post_training/nvidia/post_training.py @@ -224,7 +224,7 @@ class NvidiaPostTrainingAdapter(ModelRegistryHelper): Parameters: training_config: TrainingConfig - Configuration for training - model: str - Model identifier + model: str - NeMo Customizer configuration name algorithm_config: Optional[AlgorithmConfig] - Algorithm-specific configuration checkpoint_dir: Optional[str] - Directory containing model checkpoints, ignored atm job_uuid: str - Unique identifier for the job, ignored atm @@ -299,9 +299,6 @@ class NvidiaPostTrainingAdapter(ModelRegistryHelper): User is informed about unsupported parameters via warnings. """ - # Map model to nvidia model name - # See `_MODEL_ENTRIES` for supported models - nvidia_model = self.get_provider_model_id(model) # Check for unsupported method parameters unsupported_method_params = [] @@ -347,7 +344,7 @@ class NvidiaPostTrainingAdapter(ModelRegistryHelper): # Prepare base job configuration job_config = { - "config": nvidia_model, + "config": model, "dataset": { "name": training_config["data_config"]["dataset_id"], "namespace": self.config.dataset_namespace, diff --git a/tests/unit/providers/nvidia/test_supervised_fine_tuning.py b/tests/unit/providers/nvidia/test_supervised_fine_tuning.py index 319011be3..97ca02fba 100644 --- a/tests/unit/providers/nvidia/test_supervised_fine_tuning.py +++ b/tests/unit/providers/nvidia/test_supervised_fine_tuning.py @@ -165,7 +165,7 @@ class TestNvidiaPostTraining(unittest.TestCase): training_job = self.run_async( self.adapter.supervised_fine_tune( job_uuid="1234", - model="meta-llama/Llama-3.1-8B-Instruct", + model="meta/llama-3.2-1b-instruct@v1.0.0+L40", checkpoint_dir="", algorithm_config=algorithm_config, training_config=convert_pydantic_to_json_value(training_config), @@ -184,7 +184,7 @@ class TestNvidiaPostTraining(unittest.TestCase): "POST", "/v1/customization/jobs", expected_json={ - "config": "meta/llama-3.1-8b-instruct", + "config": "meta/llama-3.2-1b-instruct@v1.0.0+L40", "dataset": {"name": "sample-basic-test", "namespace": "default"}, "hyperparameters": { "training_type": "sft", @@ -219,7 +219,7 @@ class TestNvidiaPostTraining(unittest.TestCase): self.run_async( self.adapter.supervised_fine_tune( job_uuid="1234", - model="meta-llama/Llama-3.1-8B-Instruct", + model="meta/llama-3.2-1b-instruct@v1.0.0+L40", checkpoint_dir="", algorithm_config=algorithm_config, training_config=convert_pydantic_to_json_value(training_config),