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 3, 17:45 UTC
Remove verifier model type check in speculative config (#42536)
Signed-off-by: Fynn Schmitt-Ulms <fschmitt@redhat.com>
Fynn Schmitt-Ulms committed
May 13, 2026 at 14:14 UTC
ab1ad0d7a97fbcc175cf160daf64146a6fe8bd41
1 file changed
-29
vllm/config/speculative.py
-29
@@ -1001,35 +1001,6 @@ class SpeculativeConfig:
1001
self.draft_parallel_config
1002
)
1003
1004
- aux_hidden_states_supported = [
1005
- "llama",
1006
- "qwen",
1007
- "minicpm",
1008
- "gpt_oss",
1009
- "hunyuan_vl",
1010
- "hunyuan_v1_dense",
1011
- "afmoe",
1012
- "nemotron_h",
1013
- "deepseek_v2",
1014
- "deepseek_v3",
1015
- "kimi_k2",
1016
- "kimi_k25",
1017
- "minimax_m2",
1018
- "gemma4",
1019
- "laguna",
1020
- ]
1021
- if (
1022
- self.method in ("eagle3", "extract_hidden_states", "dflash")
1023
- and self.target_model_config
1024
- and not any(
1025
- supported_model in self.target_model_config.hf_text_config.model_type
1026
- for supported_model in aux_hidden_states_supported
1027
- )
1028
- ):
1029
- raise ValueError(
1030
- f"{self.method} is only supported for {aux_hidden_states_supported}"
1031
- f" models. Got {self.target_model_config.hf_text_config.model_type=}"
1032
- )
1004
self.verify_equal_vocab_size_if_draft_model()
1005
return self
1006