diff --git a/integration-tests/fixtures/neuron/export_models.py b/integration-tests/fixtures/neuron/export_models.py index d4d0f01c..beee2ba7 100644 --- a/integration-tests/fixtures/neuron/export_models.py +++ b/integration-tests/fixtures/neuron/export_models.py @@ -46,6 +46,15 @@ MODEL_CONFIGURATIONS = { "auto_cast_type": "fp16", }, }, + "qwen3": { + "model_id": "Qwen/Qwen3-1.7B", + "export_kwargs": { + "batch_size": 4, + "sequence_length": 4096, + "num_cores": 2, + "auto_cast_type": "bf16", + }, + }, "granite": { "model_id": "ibm-granite/granite-3.1-2b-instruct", "export_kwargs": { @@ -55,6 +64,15 @@ MODEL_CONFIGURATIONS = { "auto_cast_type": "bf16", }, }, + "phi3": { + "model_id": "microsoft/Phi-3-mini-4k-instruct", + "export_kwargs": { + "batch_size": 4, + "sequence_length": 4096, + "num_cores": 2, + "auto_cast_type": "bf16", + }, + }, } diff --git a/integration-tests/neuron/test_generate.py b/integration-tests/neuron/test_generate.py index 0b42e3ca..d96aa36e 100644 --- a/integration-tests/neuron/test_generate.py +++ b/integration-tests/neuron/test_generate.py @@ -23,6 +23,8 @@ async def test_model_single_request(tgi_service): "llama": " and how does it work?\nDeep learning is a subset of machine learning that uses artificial", "qwen2": " - Deep Learning is a subset of Machine Learning that involves the use of artificial neural networks", "granite": "\n\nDeep learning is a subset of machine learning techniques based on artificial neural networks", + "qwen3": " A Deep Learning is a subset of machine learning that uses neural networks with multiple layers to", + "phi3": "\n\nDeep learning is a subfield of machine learning that focuses on creating", } assert response.generated_text == greedy_expectations[service_name] @@ -80,6 +82,8 @@ async def test_model_multiple_requests(tgi_service, neuron_generate_load): "llama": "Deep learning is a subset of machine learning that uses artificial", "qwen2": "Deep Learning is a subset of Machine Learning that involves", "granite": "Deep learning is a subset of machine learning techniques", + "qwen3": "Deep Learning is a subset of machine learning that uses neural networks", + "phi3": "Deep learning is a subfield of machine learning that focuses on creating", } expected = expectations[tgi_service.client.service_name] for r in responses: