load_glm5_next_vision

spark_model::weight_loader::glm5_next_vision

Function load_glm5_next_vision 

Source 
pub fn load_glm5_next_vision(
    store: &WeightStore,
    config: &ModelConfig,
    gpu: &dyn GpuBackend,
) -> Result<Option<VisionTower>>
Expand description

Glm5NextWeightLoader::load_vision_encoder.

Ok(None) for a text-only checkpoint (glm5_next_text, whose config has no vision_config) and for a multimodal one whose tower is absent from the shards — the second case warns, because a declared-but-missing tower means image requests will be rejected later and the operator should hear it at load rather than at the first request.