[V0 Deprecation] Remove LLMEngine (#25033)

Signed-off-by: Woosuk Kwon <woosuk@thinkingmachines.ai>
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon
2025-09-20 17:56:30 -07:00
committed by GitHub
parent 367a480bd3
commit 52c2a8d4ad
29 changed files with 65 additions and 2763 deletions
-8
View File
@@ -1508,14 +1508,6 @@ class EngineArgs:
recommend_to_remove=True)
return False
if self.kv_cache_dtype != "auto":
supported = current_platform.is_kv_cache_dtype_supported(
self.kv_cache_dtype, model_config)
if not supported:
_raise_or_fallback(feature_name="--kv-cache-dtype",
recommend_to_remove=False)
return False
# No Mamba or Encoder-Decoder so far.
if not model_config.is_v1_compatible:
_raise_or_fallback(feature_name=model_config.architectures,
+2 -1831
View File
File diff suppressed because it is too large Load Diff
+2 -9
View File
@@ -11,7 +11,6 @@ from pydantic import ValidationError
from tqdm.auto import tqdm
from typing_extensions import TypeVar
import vllm.envs as envs
from vllm.beam_search import (BeamSearchInstance, BeamSearchOutput,
BeamSearchSequence,
create_sort_beams_key_function)
@@ -19,7 +18,6 @@ from vllm.config import (CompilationConfig, ModelDType,
StructuredOutputsConfig, TokenizerMode, is_init_field)
from vllm.engine.arg_utils import (ConvertOption, EngineArgs, HfOverrides,
PoolerConfig, RunnerOption)
from vllm.engine.llm_engine import LLMEngine
from vllm.entrypoints.chat_utils import (ChatCompletionMessageParam,
ChatTemplateContentFormatOption,
apply_hf_chat_template,
@@ -54,6 +52,7 @@ from vllm.transformers_utils.tokenizer import (AnyTokenizer, MistralTokenizer,
get_cached_tokenizer)
from vllm.usage.usage_lib import UsageContext
from vllm.utils import Counter, Device, as_iter, is_list_of
from vllm.v1.engine.llm_engine import LLMEngine
from vllm.v1.sample.logits_processor import LogitsProcessor
if TYPE_CHECKING:
@@ -309,11 +308,7 @@ class LLM:
self.request_counter = Counter()
self.default_sampling_params: Union[dict[str, Any], None] = None
if envs.VLLM_USE_V1:
supported_tasks = self.llm_engine \
.get_supported_tasks() # type: ignore
else:
supported_tasks = self.llm_engine.model_config.supported_tasks
supported_tasks = self.llm_engine.get_supported_tasks() # type: ignore
logger.info("Supported_tasks: %s", supported_tasks)
@@ -1473,8 +1468,6 @@ class LLM:
Note:
This method is only available with the V1 LLM engine.
"""
from vllm.v1.engine.llm_engine import LLMEngine as V1LLMEngine
assert isinstance(self.llm_engine, V1LLMEngine)
return self.llm_engine.get_metrics()
def _validate_and_add_requests(
+8 -14
View File
@@ -672,21 +672,15 @@ def tensorize_vllm_model(engine_args: "EngineArgs",
) as stream:
stream.write(encryption_params.key)
from vllm import LLMEngine
from vllm.v1.engine.llm_engine import LLMEngine as V1LLMEngine
assert envs.VLLM_USE_V1
if not envs.VLLM_USE_V1:
engine = LLMEngine.from_engine_args(engine_args)
engine.model_executor.collective_rpc(
"save_tensorized_model",
kwargs={"tensorizer_config": tensorizer_config.to_serializable()},
)
else:
engine = V1LLMEngine.from_vllm_config(engine_config)
engine.collective_rpc(
"save_tensorized_model",
kwargs={"tensorizer_config": tensorizer_config.to_serializable()},
)
from vllm.v1.engine.llm_engine import LLMEngine
engine = LLMEngine.from_vllm_config(engine_config)
engine.collective_rpc(
"save_tensorized_model",
kwargs={"tensorizer_config": tensorizer_config.to_serializable()},
)
def tensorize_lora_adapter(lora_path: str,