8f869c8ea5
Signed-off-by: Megha Agarwal <agarwalmegha1308@gmail.com> Signed-off-by: meghaagr13 <divyanshsinghvi@gmail.com> Signed-off-by: Taichang Zhou <tzhouam@connect.ust.hk> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com> Co-authored-by: meghaagr13 <divyanshsinghvi@gmail.com> Co-authored-by: Boao Shi <aoibosh@connect.hku.hk> Co-authored-by: Taichang Zhou <tzhouam@connect.ust.hk> Co-authored-by: Cursor <cursoragent@cursor.com>
86 lines
2.3 KiB
Python
86 lines
2.3 KiB
Python
import pytest
|
|
import torch
|
|
|
|
from vllm_omni.worker.gpu_generation_model_runner import (
|
|
ExecuteModelState,
|
|
GPUGenerationModelRunner,
|
|
)
|
|
|
|
pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
|
|
|
|
|
|
class _DummyInputBatch:
|
|
def __init__(self):
|
|
self.req_ids = ["req-1"]
|
|
self.req_id_to_index = {"req-1": 0}
|
|
self.num_reqs = 1
|
|
self.vocab_size = 10
|
|
|
|
|
|
def _make_runner(multimodal_outputs):
|
|
runner = object.__new__(GPUGenerationModelRunner)
|
|
runner.execute_model_state = ExecuteModelState(
|
|
None,
|
|
None,
|
|
None,
|
|
None,
|
|
None,
|
|
None,
|
|
None,
|
|
None,
|
|
None,
|
|
None,
|
|
multimodal_outputs,
|
|
None,
|
|
)
|
|
runner.kv_connector_output = None
|
|
runner.input_batch = _DummyInputBatch()
|
|
runner.use_async_scheduling = False
|
|
runner.device = torch.device("cpu")
|
|
runner.supports_mm_inputs = False
|
|
runner.speculative_config = None
|
|
runner.routed_experts_initialized = False
|
|
return runner
|
|
|
|
|
|
def test_sample_tokens_tensor_output():
|
|
multimodal_outputs = torch.randn(1, 2, 3)
|
|
runner = _make_runner(multimodal_outputs)
|
|
|
|
output = GPUGenerationModelRunner.sample_tokens(runner)
|
|
|
|
assert len(output.multimodal_outputs) == 1
|
|
assert output.multimodal_outputs[0]["model_outputs"].shape == (2, 3)
|
|
|
|
|
|
def test_sample_tokens_list_output():
|
|
multimodal_outputs = [torch.randn(2, 1)]
|
|
runner = _make_runner(multimodal_outputs)
|
|
|
|
output = GPUGenerationModelRunner.sample_tokens(runner)
|
|
|
|
assert len(output.multimodal_outputs) == 1
|
|
assert output.multimodal_outputs[0]["model_outputs"].shape == (2, 1)
|
|
|
|
|
|
def test_sample_tokens_list_allows_none_output():
|
|
multimodal_outputs = [None]
|
|
runner = _make_runner(multimodal_outputs)
|
|
|
|
output = GPUGenerationModelRunner.sample_tokens(runner)
|
|
|
|
assert len(output.multimodal_outputs) == 1
|
|
assert output.multimodal_outputs[0]["model_outputs"] is None
|
|
|
|
|
|
def test_sample_tokens_dict_output():
|
|
multimodal_outputs = {"audio": torch.randn(1, 4), "unused": None}
|
|
runner = _make_runner(multimodal_outputs)
|
|
|
|
output = GPUGenerationModelRunner.sample_tokens(runner)
|
|
|
|
assert len(output.multimodal_outputs) == 1
|
|
assert "audio" in output.multimodal_outputs[0]
|
|
assert "unused" not in output.multimodal_outputs[0]
|
|
assert output.multimodal_outputs[0]["audio"].shape == (1, 4)
|