hf_text-generation-inference/integration-tests/models/test_flash_neox_sharded.py

41 lines
1.2 KiB
Python
Raw Normal View History

import pytest
@pytest.fixture(scope="module")
def flash_neox_sharded_handle(launcher):
with launcher("OpenAssistant/oasst-sft-1-pythia-12b", num_shard=2) as handle:
yield handle
@pytest.fixture(scope="module")
async def flash_neox_sharded(flash_neox_sharded_handle):
2023-05-31 02:55:59 -06:00
await flash_neox_sharded_handle.health(300)
return flash_neox_sharded_handle.client
@pytest.mark.asyncio
async def test_flash_neox(flash_neox_sharded, response_snapshot):
response = await flash_neox_sharded.generate(
"<|prompter|>What is a meme, and what's the history behind this word?<|endoftext|><|assistant|>",
max_new_tokens=10,
decoder_input_details=True,
)
assert response.details.generated_tokens == 10
assert response == response_snapshot
@pytest.mark.asyncio
async def test_flash_neox_load(flash_neox_sharded, generate_load, response_snapshot):
responses = await generate_load(
flash_neox_sharded,
"<|prompter|>What is a meme, and what's the history behind this word?<|endoftext|><|assistant|>",
max_new_tokens=10,
n=4,
)
assert len(responses) == 4
assert all([r.generated_text == responses[0].generated_text for r in responses])
assert responses == response_snapshot