biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

feat: Enable engine-level arguments with speculators models (#25250)

Browse Source 
Signed-off-by: Rahul Tuli <rtuli@redhat.com>
Co-authored-by: Claude <noreply@anthropic.com>
This commit is contained in:
Rahul Tuli
2025-09-21 22:34:45 +05:30
committed by GitHub
parent 0ff8ebb2d7
commit c438b2951c
5 changed files with 128 additions and 85 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/config/model.py
View File

@@ -27,8 +27,7 @@ from vllm.transformers_utils.config import (
ConfigFormat, get_config, get_hf_image_processor_config,
get_hf_text_config, get_pooling_config,
get_sentence_transformer_tokenizer_config, is_encoder_decoder,
is_interleaved, maybe_override_with_speculators_target_model,
try_get_generation_config, try_get_safetensors_metadata,
is_interleaved, try_get_generation_config, try_get_safetensors_metadata,
try_get_tokenizer_config, uses_mrope)
from vllm.transformers_utils.runai_utils import (ObjectStorageModel,
is_runai_obj_uri)
@@ -416,15 +415,6 @@ class ModelConfig:
self.maybe_pull_model_tokenizer_for_runai(self.model, self.tokenizer)
if self.runner != "draft":
# If we're not running the draft model, check for speculators config
# If speculators config, set model / tokenizer to be target model
self.model, self.tokenizer = maybe_override_with_speculators_target_model( # noqa: E501
model=self.model,
tokenizer=self.tokenizer,
revision=self.revision,
trust_remote_code=self.trust_remote_code)
if (backend := envs.VLLM_ATTENTION_BACKEND
) and backend == "FLASHINFER" and find_spec("flashinfer") is None:
raise ValueError(