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

[Hardware][CPU] using current_platform.is_cpu (#9536)

Browse Source
This commit is contained in:
wangshuai09
2024-10-22 15:50:43 +08:00
committed by GitHub
parent 0d02747f2e
commit 3ddbe25502
17 changed files with 60 additions and 64 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
tests/worker/test_encoder_decoder_model_runner.py
View File

@ -5,8 +5,9 @@ import pytest
import torch
from vllm.engine.arg_utils import EngineArgs
from vllm.platforms import current_platform
from vllm.sequence import SamplingParams, SequenceData, SequenceGroupMetadata
from vllm.utils import is_cpu, make_tensor_with_pad
from vllm.utils import make_tensor_with_pad
from vllm.worker.enc_dec_model_runner import EncoderDecoderModelRunner
from vllm.worker.model_runner import _get_graph_batch_size
@ -31,7 +32,7 @@ def _create_model_runner(model: str, *args,
return model_runner
@pytest.mark.skipif(condition=is_cpu(),
@pytest.mark.skipif(condition=current_platform.is_cpu(),
reason="CPU backend is currently "
"unsupported for encoder/ "
"decoder models")
@ -74,7 +75,7 @@ def test_empty_seq_group():
assert return_seq_lens is None
@pytest.mark.skipif(condition=is_cpu(),
@pytest.mark.skipif(condition=current_platform.is_cpu(),
reason="CPU backend is currently "
"unsupported for encoder/ "
"decoder models")
@ -264,7 +265,7 @@ def test_prepare_prompt(batch_size):
assert torch.equal(actual, expected)
@pytest.mark.skipif(condition=is_cpu(),
@pytest.mark.skipif(condition=current_platform.is_cpu(),
reason="CPU backend is currently "
"unsupported for encoder/ "
"decoder models")
@ -490,7 +491,7 @@ def test_prepare_decode(batch_size, multiple_seqs_per_seq_group):
def test_prepare_decode_cuda_graph(batch_size, multiple_seqs_per_seq_group):
"""
Tests that for encoder-decoder models with CUDA Graph capture and replay
enabled, the tensors used during the decode phase are correctly padded
enabled, the tensors used during the decode phase are correctly padded
for varying input batch sizes.
"""
model_runner = _create_model_runner(