huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-09 19:34:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

test(neuron): add phi3 and qwen3 tests

Browse Source
This commit is contained in:
David Corvoysier 2025-06-27 13:14:07 +00:00
parent 83b766f9ef
commit 7eca8e584e
2 changed files with 22 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
integration-tests/fixtures/neuron/export_models.py
View File

@ -46,6 +46,15 @@ MODEL_CONFIGURATIONS = {
"auto_cast_type": "fp16", "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": { "granite": {
"model_id": "ibm-granite/granite-3.1-2b-instruct", "model_id": "ibm-granite/granite-3.1-2b-instruct",
"export_kwargs": { "export_kwargs": {
@ -55,6 +64,15 @@ MODEL_CONFIGURATIONS = {
"auto_cast_type": "bf16", "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",
},
},
} }

4
integration-tests/neuron/test_generate.py
View File

@ -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", "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", "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", "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] 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", "llama": "Deep learning is a subset of machine learning that uses artificial",
"qwen2": "Deep Learning is a subset of Machine Learning that involves", "qwen2": "Deep Learning is a subset of Machine Learning that involves",
"granite": "Deep learning is a subset of machine learning techniques", "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] expected = expectations[tgi_service.client.service_name]
for r in responses: for r in responses: