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

refactor hard coded device string in test files under tests/v1 and tests/lora (#37566)

Browse Source 
Signed-off-by: Liao, Wei <wei.liao@intel.com>
This commit is contained in:
wliao2
2026-04-02 20:21:47 -07:00
committed by GitHub
parent 4a06e1246e
commit 32e0c0bfa2
28 changed files with 239 additions and 146 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
tests/v1/spec_decode/test_mtp.py
View File

@@ -28,6 +28,7 @@ from vllm.v1.attention.backends.registry import AttentionBackendEnum
from vllm.v1.spec_decode.eagle import EagleProposer
mimo_7b_dir = "XiaomiMiMo/MiMo-7B-Base"
DEVICE_TYPE = current_platform.device_type
def _create_mtp_proposer(num_speculative_tokens: int) -> EagleProposer:
@@ -48,7 +49,7 @@ def _create_mtp_proposer(num_speculative_tokens: int) -> EagleProposer:
model_config=model_config,
cache_config=CacheConfig(),
speculative_config=speculative_config,
device_config=DeviceConfig(device=current_platform.device_type),
device_config=DeviceConfig(device=DEVICE_TYPE),
parallel_config=ParallelConfig(),
load_config=LoadConfig(),
scheduler_config=SchedulerConfig(
@@ -57,7 +58,7 @@ def _create_mtp_proposer(num_speculative_tokens: int) -> EagleProposer:
),
)
return EagleProposer(vllm_config=vllm_config, device=current_platform.device_type)
return EagleProposer(vllm_config=vllm_config, device=DEVICE_TYPE)
@mock.patch("vllm.v1.spec_decode.eagle.get_pp_group")
@@ -118,7 +119,7 @@ def test_mtp_load_model_unified(mock_get_model, mock_get_layers, mock_get_pp_gro
def test_mtp_propose(num_speculative_tokens, monkeypatch):
"""Test that MTP's forward method returns hidden states directly"""
device = torch.device(current_platform.device_type)
device = torch.device(DEVICE_TYPE)
batch_size = 2
seq_lens = [5, 3]
total_tokens = sum(seq_lens)