conftest.py 541 Bytes
Newer Older
1
import pytest
2
import os
3
from text_generation_server.pb import generate_pb2
4

5
6
7
os.environ["USE_PREFIX_CACHING"] = "1"
os.environ["ATTENTION"] = "flashinfer"

8
9
10

@pytest.fixture
def default_pb_parameters():
OlivierDehaene's avatar
OlivierDehaene committed
11
    return generate_pb2.NextTokenChooserParameters(
12
        temperature=1.0,
13
        repetition_penalty=1.0,
14
15
        top_k=0,
        top_p=1.0,
16
        typical_p=1.0,
17
18
19
20
        do_sample=False,
    )


21
22
23
@pytest.fixture
def default_pb_stop_parameters():
    return generate_pb2.StoppingCriteriaParameters(stop_sequences=[], max_new_tokens=10)