[V1][Spec Decoding] Use model_loader.get_model() to load models (#18273)

Signed-off-by: Mark McLoughlin <markmc@redhat.com>
This commit is contained in:
Mark McLoughlin
2025-05-23 02:05:44 +00:00
committed by GitHub
parent 04eb88dc80
commit c6b636f9fb
16 changed files with 59 additions and 135 deletions
@@ -92,9 +92,9 @@ class GGUFModelLoader(BaseModelLoader):
def download_model(self, model_config: ModelConfig) -> None:
self._prepare_weights(model_config.model)
def load_model(self, vllm_config: VllmConfig) -> nn.Module:
def load_model(self, vllm_config: VllmConfig,
model_config: ModelConfig) -> nn.Module:
device_config = vllm_config.device_config
model_config = vllm_config.model_config
local_model_path = self._prepare_weights(model_config.model)
gguf_weights_map = self._get_gguf_weights_map(model_config)
# we can only know if tie word embeddings after mapping weights