biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Model] Support Qwen2 embeddings and use tags to select model tests (#10184)

Browse Source
This commit is contained in:
Cyrus Leung
2024-11-15 12:23:09 +08:00
committed by GitHub
parent 2885ba0e24
commit b40cf6402e
19 changed files with 252 additions and 178 deletions
Download Patch File Download Diff File Expand all files Collapse all files

71
tests/models/decoder_only/language/test_models.py
View File

@@ -4,37 +4,52 @@ Run `pytest tests/models/test_models.py`.
"""
import pytest
from vllm.platforms import current_platform
from ...utils import check_logprobs_close
MODELS = [
"facebook/opt-125m", # opt
"openai-community/gpt2", # gpt2
# "Milos/slovak-gpt-j-405M", # gptj
# "bigcode/tiny_starcoder_py", # gpt_bigcode
# "EleutherAI/pythia-70m", # gpt_neox
"bigscience/bloom-560m", # bloom - testing alibi slopes
"microsoft/phi-2", # phi
# "stabilityai/stablelm-3b-4e1t", # stablelm
# "bigcode/starcoder2-3b", # starcoder2
"google/gemma-1.1-2b-it", # gemma
"Qwen/Qwen2.5-0.5B-Instruct", # qwen2
"meta-llama/Llama-3.2-1B-Instruct", # llama
]
if not current_platform.is_cpu():
MODELS += [
# fused_moe which not supported on CPU
"openbmb/MiniCPM3-4B",
]
target_dtype = "half"
@pytest.mark.core_model
@pytest.mark.parametrize("model", MODELS)
@pytest.mark.parametrize("dtype", [target_dtype])
@pytest.mark.parametrize(
"model",
[
pytest.param(
"bigscience/bloom-560m", # bloom - testing alibi slopes
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
),
pytest.param(
"openai-community/gpt2", # gpt2
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
),
pytest.param("Milos/slovak-gpt-j-405M"), # gptj
pytest.param("bigcode/tiny_starcoder_py"), # gpt_bigcode
pytest.param("EleutherAI/pythia-70m"), # gpt_neox
pytest.param(
"google/gemma-1.1-2b-it", # gemma
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
),
pytest.param(
"meta-llama/Llama-3.2-1B-Instruct", # llama
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
),
pytest.param(
"openbmb/MiniCPM3-4B",
# fused_moe not supported on CPU
marks=[pytest.mark.core_model],
),
pytest.param(
"facebook/opt-125m", # opt
marks=[pytest.mark.core_model, pytest.mark.cpu_model],
),
pytest.param(
"microsoft/phi-2", # phi
marks=[pytest.mark.core_model],
),
pytest.param(
"Qwen/Qwen2.5-0.5B-Instruct", # qwen2
marks=[pytest.mark.core_model],
),
pytest.param("stabilityai/stablelm-3b-4e1t"), # stablelm
pytest.param("bigcode/starcoder2-3b"), # starcoder2
])
@pytest.mark.parametrize("dtype", ["half"])
@pytest.mark.parametrize("max_tokens", [32])
@pytest.mark.parametrize("num_logprobs", [5])
def test_models(