Expand each <|image|> placeholder id into boi + image*softCount + eoi, matching optiq's _build_image_inputs. Returns the spliced ids (the diffusion engine's encoder prefills the merged vision features at the contiguous image run).
<|image|>
boi + image*softCount + eoi
_build_image_inputs
Expand each
<|image|>placeholder id intoboi + image*softCount + eoi, matching optiq's_build_image_inputs. Returns the spliced ids (the diffusion engine's encoder prefills the merged vision features at the contiguous image run).