test_bloom_560m_sharded.py 1.19 KB
Newer Older
1
2
import pytest

3
4
5
6
7

@pytest.fixture(scope="module")
def bloom_560m_sharded_handle(launcher):
    with launcher("bigscience/bloom-560m", num_shard=2) as handle:
        yield handle
8
9
10


@pytest.fixture(scope="module")
11
async def bloom_560m_sharded(bloom_560m_sharded_handle):
OlivierDehaene's avatar
OlivierDehaene committed
12
    await bloom_560m_sharded_handle.health(240)
13
    return bloom_560m_sharded_handle.client
14
15
16


@pytest.mark.asyncio
17
async def test_bloom_560m_sharded(bloom_560m_sharded, response_snapshot):
18
19
20
21
    response = await bloom_560m_sharded.generate(
        "Pour déguster un ortolan, il faut tout d'abord",
        max_new_tokens=10,
        top_p=0.9,
22
        decoder_input_details=True,
23
24
25
26
        seed=0,
    )

    assert response.details.generated_tokens == 10
27
    assert response == response_snapshot
28
29
30
31


@pytest.mark.asyncio
async def test_bloom_560m_sharded_load(
32
    bloom_560m_sharded, generate_load, response_snapshot
33
34
35
36
37
38
39
40
41
):
    responses = await generate_load(
        bloom_560m_sharded,
        "Pour déguster un ortolan, il faut tout d'abord",
        max_new_tokens=10,
        n=4,
    )

    assert len(responses) == 4
42
    assert all([r.generated_text == responses[0].generated_text for r in responses])
43

44
    assert responses == response_snapshot