MCPcopy
hub / github.com/NVIDIA/TensorRT-LLM / forward

Method forward

tensorrt_llm/tools/multimodal_builder.py:795–801  ·  view source on GitHub ↗
(self, images)

Source from the content-addressed store, hash-verified

793 self.connector = connector
794
795 def forward(self, images):
796 vision_x = self.encoder(pixel_values=images,
797 output_hidden_states=True)
798 vision_x = vision_x.hidden_states[-2]
799 vision_x = vision_x[:, 1:]
800 vision_x = self.connector(vision_x)
801 return vision_x
802
803 vision_path = vision_config["from_pretrained"]
804 joined_path = os.path.join(os.path.dirname(args.model_path),

Callers

nothing calls this directly

Calls 1

viewMethod · 0.45

Tested by

no test coverage detected