A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 01:00 UTC
[Misc] Remove deprecated args in v0.10 (#21349)
Signed-off-by: Kebe <mail@kebe7jun.com>
Kebe committed
Jul 22, 2025 at 20:26 UTC
bc8a8ce5ec374dd18e86f59be7cb0057a4b21992
4 files changed
-25
examples/offline_inference/neuron_speculation.py
-1
@@ -37,7 +37,6 @@ def initialize_llm():
37
max_num_seqs=4,
38
max_model_len=2048,
39
block_size=2048,
40
- use_v2_block_manager=True,
40
device="neuron",
41
tensor_parallel_size=32,
42
)
tests/neuron/2_core/test_mistral.py
-1
@@ -9,7 +9,6 @@ def test_mistral():
9
tensor_parallel_size=2,
10
max_num_seqs=4,
11
max_model_len=128,
12
- use_v2_block_manager=True,
12
override_neuron_config={
13
"sequence_parallel_enabled": False,
14
"skip_warmup": True
tests/neuron/2_core/test_multi_lora.py
-2
@@ -14,7 +14,6 @@ def test_llama_single_lora():
14
tensor_parallel_size=2,
15
max_num_seqs=4,
16
max_model_len=512,
17
- use_v2_block_manager=True,
17
override_neuron_config={
18
"sequence_parallel_enabled": False,
19
"skip_warmup": True,
@@ -57,7 +56,6 @@ def test_llama_multiple_lora():
56
tensor_parallel_size=2,
57
max_num_seqs=4,
58
max_model_len=512,
60
- use_v2_block_manager=True,
59
override_neuron_config={
60
"sequence_parallel_enabled":
61
False,
vllm/engine/arg_utils.py
-21
@@ -313,7 +313,6 @@ class EngineArgs:
313
CacheConfig.prefix_caching_hash_algo
314
disable_sliding_window: bool = ModelConfig.disable_sliding_window
315
disable_cascade_attn: bool = ModelConfig.disable_cascade_attn
316
- use_v2_block_manager: bool = True
316
swap_space: float = CacheConfig.swap_space
317
cpu_offload_gb: float = CacheConfig.cpu_offload_gb
318
gpu_memory_utilization: float = CacheConfig.gpu_memory_utilization
@@ -364,7 +363,6 @@ class EngineArgs:
363
max_prompt_adapter_token: int = \
364
PromptAdapterConfig.max_prompt_adapter_token
365
367
- device: Device = DeviceConfig.device
366
num_scheduler_steps: int = SchedulerConfig.num_scheduler_steps
367
multi_step_stream_outputs: bool = SchedulerConfig.multi_step_stream_outputs
368
ray_workers_use_nsight: bool = ParallelConfig.ray_workers_use_nsight
@@ -745,16 +743,6 @@ class EngineArgs:
743
"--max-prompt-adapter-token",
744
**prompt_adapter_kwargs["max_prompt_adapter_token"])
745
748
- # Device arguments
749
- device_kwargs = get_kwargs(DeviceConfig)
750
- device_group = parser.add_argument_group(
751
- title="DeviceConfig",
752
- description=DeviceConfig.__doc__,
753
- )
754
- device_group.add_argument("--device",
755
- **device_kwargs["device"],
756
- deprecated=True)
757
-
746
# Speculative arguments
747
speculative_group = parser.add_argument_group(
748
title="SpeculativeConfig",
@@ -856,15 +844,6 @@ class EngineArgs:
844
**vllm_kwargs["additional_config"])
845
846
# Other arguments
859
- parser.add_argument('--use-v2-block-manager',
860
- action='store_true',
861
- default=True,
862
- deprecated=True,
863
- help='[DEPRECATED] block manager v1 has been '
864
- 'removed and SelfAttnBlockSpaceManager (i.e. '
865
- 'block manager v2) is now the default. '
866
- 'Setting this flag to True or False'
867
- ' has no effect on vLLM behavior.')
847
parser.add_argument('--disable-log-stats',
848
action='store_true',
849
help='Disable logging statistics.')