pub fn pixel_bounds(
vcfg: &VisionConfig,
operator_max: Option<usize>,
) -> (u64, u64)Expand description
2026-09-25: (min_pixels, max_pixels) for smart_resize, as pixel areas.
GLM states its budget in merged tokens (min_image_tokens,
max_image_tokens), converted as tokens × factor²: one merged token
covers one factor × factor pixel block. max_pixels is the least of:
- the encoder’s capacity, from
derive_max_patches, the functionGlmVitsizes its buffers with; - the declared token budget, when nonzero;
operator_max, when given;
and never below one merged token. With no bound the encoder holds 6400
patches (FALLBACK_MAX_PATCHES), while GLM’s default budget of 8000 merged
tokens is 32000 patches, so the encoder term binds.