[6/N][warmup][DSv4] Migrate sampling, and DFlash JIT kernels (#56323)
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com> Signed-off-by: LopezCastroRoberto <rocastro@redhat.com> Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com> Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
co-authored by
Lucas Wilkinson
OpenAI Codex
parent
e19a3e172e
commit
aed894c190
@@ -79,6 +79,10 @@ def test_rocm_aiter_sampler_defers_import_when_generators_force_native(
|
||||
def is_cuda():
|
||||
return False
|
||||
|
||||
@staticmethod
|
||||
def is_cuda_alike():
|
||||
return True
|
||||
|
||||
@staticmethod
|
||||
def is_cpu():
|
||||
return False
|
||||
|
||||
@@ -11,6 +11,7 @@ from tests.v1.shutdown.utils import (
|
||||
SHUTDOWN_TEST_TIMEOUT_SEC,
|
||||
)
|
||||
from vllm import LLM, SamplingParams
|
||||
from vllm.config import KernelConfig
|
||||
from vllm.engine.arg_utils import AsyncEngineArgs
|
||||
from vllm.platforms import current_platform
|
||||
from vllm.sampling_params import RequestOutputKind
|
||||
@@ -39,7 +40,10 @@ async def test_async_llm_delete(
|
||||
pytest.skip(reason="Not enough CUDA devices")
|
||||
|
||||
engine_args = AsyncEngineArgs(
|
||||
model=model, enforce_eager=True, tensor_parallel_size=tensor_parallel_size
|
||||
model=model,
|
||||
enforce_eager=True,
|
||||
tensor_parallel_size=tensor_parallel_size,
|
||||
kernel_config=KernelConfig(enable_jit_warmup=False),
|
||||
)
|
||||
|
||||
# Instantiate AsyncLLM; make request to complete any deferred
|
||||
@@ -94,7 +98,10 @@ def test_llm_delete(
|
||||
# Instantiate LLM; make request to complete any deferred
|
||||
# initialization; then delete instance
|
||||
llm = LLM(
|
||||
model=model, enforce_eager=True, tensor_parallel_size=tensor_parallel_size
|
||||
model=model,
|
||||
enforce_eager=True,
|
||||
tensor_parallel_size=tensor_parallel_size,
|
||||
kernel_config=KernelConfig(enable_jit_warmup=False),
|
||||
)
|
||||
if send_one_request:
|
||||
llm.generate(
|
||||
@@ -122,7 +129,10 @@ def test_llm_delete_inprocess(
|
||||
with monkeypatch.context() as m:
|
||||
m.setenv("VLLM_ENABLE_V1_MULTIPROCESSING", "0")
|
||||
|
||||
with VllmRunner(model) as vllm_model:
|
||||
with VllmRunner(
|
||||
model,
|
||||
kernel_config=KernelConfig(enable_jit_warmup=False),
|
||||
) as vllm_model:
|
||||
if send_one_request:
|
||||
vllm_model.generate(
|
||||
["Hello my name is"],
|
||||
|
||||
@@ -43,7 +43,7 @@ def _run_prepare(
|
||||
num_scheduled_tokens=np.array([4], dtype=np.int32),
|
||||
positions=torch.tensor(target_positions, dtype=torch.int64, device=device),
|
||||
query_start_loc=torch.tensor([0, 4], dtype=torch.int32, device=device),
|
||||
idx_mapping=torch.tensor([2], dtype=torch.int32, device=device),
|
||||
idx_mapping=torch.tensor([2], dtype=torch.int64, device=device),
|
||||
)
|
||||
query_slot_mapping = torch.full(
|
||||
(max_num_tokens,), -2, dtype=torch.int64, device=device
|
||||
|
||||
@@ -157,6 +157,7 @@ def test_mm_support_configured_after_model_load(monkeypatch):
|
||||
speculator.device = device
|
||||
speculator.max_num_tokens = 4
|
||||
speculator.max_num_reqs = 2
|
||||
speculator.num_speculative_steps = 1
|
||||
speculator.hidden_size = 3
|
||||
speculator.dtype = torch.float32
|
||||
speculator.draft_model_config = draft_model_config
|
||||
|
||||
Reference in New Issue
Block a user