Deprecations for v0.23 and v0.24 (#44992)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2026-06-11 14:35:38 +00:00
committed by GitHub
parent 55911db580
commit 03878d1c22
35 changed files with 100 additions and 674 deletions
+1 -3
View File
@@ -1585,7 +1585,6 @@ def test_unquantized_bf16_flashinfer_trtllm_backend(
e: int,
topk: int,
dtype: torch.dtype,
monkeypatch,
workspace_init,
):
"""
@@ -1593,8 +1592,6 @@ def test_unquantized_bf16_flashinfer_trtllm_backend(
"""
set_random_seed(7)
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_FP16", "1")
from vllm.model_executor.layers.fused_moe.config import (
FusedMoEConfig,
FusedMoEParallelConfig,
@@ -1626,6 +1623,7 @@ def test_unquantized_bf16_flashinfer_trtllm_backend(
in_dtype=dtype,
routing_method=RoutingMethodType.Renormalize,
max_num_tokens=next_power_of_2(m),
moe_backend="flashinfer_trtllm",
)
with set_current_vllm_config(vllm_config):
+3 -6
View File
@@ -1804,12 +1804,9 @@ def test_moe_layer(
if os.environ.get("VLLM_LOGGING_LEVEL") is None:
monkeypatch.setenv("VLLM_LOGGING_LEVEL", "ERROR")
# TODO
# VLLM_FLASHINFER_MOE_BACKEND=latency
# VLLM_USE_FLASHINFER_MOE_FP16=1
# VLLM_USE_FLASHINFER_MOE_FP8
# VLLM_USE_FLASHINFER_MOE_FP4
# VLLM_USE_FLASHINFER_MOE_INT4
# TODO: cover FlashInfer MoE backends via moe_backend, e.g.
# moe_backend=flashinfer_trtllm / flashinfer_cutlass / flashinfer_cutedsl
# (BF16, FP8 and NVFP4 paths), and VLLM_USE_FLASHINFER_MOE_INT4=1.
parallel_config = ParallelConfig(
pipeline_parallel_size=1,
@@ -123,7 +123,7 @@ def test_select_rocm_aiter_backend(mock_aiter_enabled, mock_has_flashinfer):
@pytest.mark.skipif(
not current_platform.is_cuda(), reason="Only supported on NVIDIA platforms."
)
def test_select_cuda_flashinfer_trtllm_backend(mock_is_supported_trtllm, monkeypatch):
def test_select_cuda_flashinfer_trtllm_backend(mock_is_supported_trtllm):
"""Test CUDA backend selection when FlashInfer TRTLLM is available and enabled."""
with (
patch.object(current_platform, "is_cuda", return_value=True),
@@ -134,9 +134,8 @@ def test_select_cuda_flashinfer_trtllm_backend(mock_is_supported_trtllm, monkeyp
patch.object(current_platform, "is_out_of_tree", return_value=False),
patch.object(current_platform, "has_device_capability", return_value=True),
):
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_FP16", "1")
moe_config = make_dummy_moe_config()
moe_config.moe_backend = "flashinfer_trtllm"
# TRTLLM requires EP and does not support DP
moe_config.moe_parallel_config.use_ep = True
moe_config.moe_parallel_config.use_dp = False
@@ -168,7 +167,6 @@ def test_select_cuda_flashinfer_cutlass_backend(
mock_has_flashinfer,
mock_is_supported_trtllm,
mock_is_supported_cutlass,
monkeypatch,
):
"""Test CUDA backend selection when FlashInfer TRTLLM is not available
and FlashInfer CUTLASS is available."""
@@ -181,10 +179,9 @@ def test_select_cuda_flashinfer_cutlass_backend(
patch.object(current_platform, "is_out_of_tree", return_value=False),
patch.object(current_platform, "has_device_capability", return_value=True),
):
# Enable FlashInfer via env var
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_FP16", "1")
moe_config = make_dummy_moe_config()
# Select FlashInfer CUTLASS explicitly
moe_config.moe_backend = "flashinfer_cutlass"
# CUTLASS requires EP and does not support DP
moe_config.moe_parallel_config.use_ep = True
moe_config.moe_parallel_config.use_dp = False
@@ -241,37 +238,3 @@ def test_select_explicit_triton_backend(is_lora_enabled):
assert selected_backend == UnquantizedMoeBackend.TRITON
assert experts_cls is not None
@skipif_not_cuda_rocm
def test_select_explicit_triton_ignores_flashinfer_env(monkeypatch):
"""Explicit triton backend should override FlashInfer env selection."""
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_FP16", "1")
monkeypatch.setenv("VLLM_FLASHINFER_MOE_BACKEND", "throughput")
moe_config = make_dummy_moe_config()
moe_config.is_lora_enabled = False
moe_config.moe_backend = "triton"
selected_backend, experts_cls = select_unquantized_moe_backend(
moe_config=moe_config
)
assert selected_backend == UnquantizedMoeBackend.TRITON
assert experts_cls is not None
@skipif_not_cuda_rocm
def test_select_lora_ignores_flashinfer_env(monkeypatch):
"""LoRA path should still choose Triton even if FlashInfer env is on."""
monkeypatch.setenv("VLLM_USE_FLASHINFER_MOE_FP16", "1")
monkeypatch.setenv("VLLM_FLASHINFER_MOE_BACKEND", "throughput")
moe_config = make_dummy_moe_config()
moe_config.is_lora_enabled = True
selected_backend, experts_cls = select_unquantized_moe_backend(
moe_config=moe_config
)
assert selected_backend == UnquantizedMoeBackend.TRITON
assert experts_cls is not None