mirror of
https://github.com/huggingface/text-generation-inference.git
synced 2025-04-23 07:52:06 +00:00
c00add9c03
* feat: add neuron backend * feat(neuron): add server standalone installation * feat(neuron): add server and integration tests * fix(neuron): increase ulimit when building image The base image used to compile the rust components seems to have a low ulimit for opened files, which leads to errors during compilation. * test(neuron): merge integration tests and fixtures * test: add --neuron option * review: do not use latest tag * review: remove ureq pinned version * review: --privileged should be the exception * feat: add neuron case to build ci * fix(neuron): export models from container in test fixtures The neuron tests require models to have been previously exported and cached on the hub. This is done automatically by the neuron.model fixture the first time the tests are ran for a specific version. This fixture used to export the models using optimum-neuron directly, but this package is not necessarily present on the system. Instead, it is now done through the neuron TGI itself, since it contains all the tools required to export the models. Note that since the CI runs docker in docker (dind) it does not seem possible to share a volume between the CI container and the container used to export the model. For that reason, a specific image with a modified entrypoint is built on-the-fly when a model export is required. * refactor: remove sagemaker entry-point The SageMaker image is built differently anyway. * fix(neuron): avoid using Levenshtein * test(neuron): use smaller llama model * feat(neuron): avoid installing CUDA in image * test(neuron): no error anymore when requesting too many tokens * ci: doing a precompilation step (with a different token). * test(neuron): avoid using image sha when exporting models We now manually evaluate the apparent hash of the neuron backend by combining the hash of the neuron backend directory and Dockerfile. This new hash is used to identify exported neuron models instead of the image sha. This has two benefits: - it changes less frequently (only hwen the neuron backend changes), which means less neuron models being pushed to the hub, - it can be evaluated locally, meaning that running the tests once locally will export the models before the CI uses them. * test(neuron): added a small script to prune test models --------- Co-authored-by: drbh <david.richard.holtz@gmail.com> Co-authored-by: Nicolas Patry <patry.nicolas@protonmail.com>
94 lines
3.5 KiB
Python
94 lines
3.5 KiB
Python
import pytest
|
|
|
|
|
|
@pytest.fixture
|
|
async def tgi_service(neuron_launcher, neuron_model_config):
|
|
model_name_or_path = neuron_model_config["neuron_model_path"]
|
|
service_name = neuron_model_config["name"]
|
|
with neuron_launcher(service_name, model_name_or_path) as tgi_service:
|
|
await tgi_service.health(600)
|
|
yield tgi_service
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_model_single_request(tgi_service):
|
|
service_name = tgi_service.client.service_name
|
|
prompt = "What is Deep Learning?"
|
|
# Greedy bounded without input
|
|
response = await tgi_service.client.text_generation(
|
|
prompt, max_new_tokens=17, details=True, decoder_input_details=True
|
|
)
|
|
assert response.details.generated_tokens == 17
|
|
greedy_expectations = {
|
|
"gpt2": "\n\nDeep learning is a new field of research that has been around for a while",
|
|
"llama": " and How Does it Work?\nDeep learning is a subset of machine learning that uses artificial",
|
|
"mistral": "\nWhat is Deep Learning?\nDeep Learning is a type of machine learning that",
|
|
"qwen2": " - Part 1\n\nDeep Learning is a subset of Machine Learning that is based on",
|
|
"granite": "\n\nDeep Learning is a subset of Machine Learning, which is a branch of Art",
|
|
}
|
|
assert response.generated_text == greedy_expectations[service_name]
|
|
|
|
# Greedy bounded with input
|
|
response = await tgi_service.client.text_generation(
|
|
"What is Deep Learning?", max_new_tokens=17, return_full_text=True, details=True, decoder_input_details=True
|
|
)
|
|
assert response.details.generated_tokens == 17
|
|
assert response.generated_text == prompt + greedy_expectations[service_name]
|
|
|
|
# Sampling
|
|
response = await tgi_service.client.text_generation(
|
|
"What is Deep Learning?",
|
|
do_sample=True,
|
|
top_k=50,
|
|
top_p=0.9,
|
|
repetition_penalty=1.2,
|
|
max_new_tokens=128,
|
|
seed=42,
|
|
)
|
|
sample_expectations = {
|
|
"gpt2": "Deep Learning",
|
|
"llama": "Deep Learning",
|
|
"mistral": "Deep learning",
|
|
"qwen2": "Deep Learning",
|
|
"granite": "Deep learning",
|
|
}
|
|
assert sample_expectations[service_name] in response
|
|
|
|
# Sampling with stop sequence
|
|
stop_sequence = sample_expectations[service_name][-5:]
|
|
response = await tgi_service.client.text_generation(
|
|
"What is Deep Learning?",
|
|
do_sample=True,
|
|
top_k=50,
|
|
top_p=0.9,
|
|
repetition_penalty=1.2,
|
|
max_new_tokens=128,
|
|
seed=42,
|
|
stop_sequences=[stop_sequence],
|
|
)
|
|
assert response.endswith(stop_sequence)
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_model_multiple_requests(tgi_service, neuron_generate_load):
|
|
num_requests = 4
|
|
responses = await neuron_generate_load(
|
|
tgi_service.client,
|
|
"What is Deep Learning?",
|
|
max_new_tokens=17,
|
|
n=num_requests,
|
|
)
|
|
|
|
assert len(responses) == 4
|
|
expectations = {
|
|
"gpt2": "Deep learning is a new field of research that has been around for a while",
|
|
"llama": "Deep learning is a subset of machine learning that uses artificial",
|
|
"mistral": "Deep Learning is a type of machine learning that",
|
|
"qwen2": "Deep Learning is a subset of Machine Learning that is based on",
|
|
"granite": "Deep Learning is a subset of Machine Learning, which is a branch of Art",
|
|
}
|
|
expected = expectations[tgi_service.client.service_name]
|
|
for r in responses:
|
|
assert r.details.generated_tokens == 17
|
|
assert expected in r.generated_text
|