WorldFoundry
English中文

SpatialLadder

SpatialLadder: multimodal-reasoning and spatial-reasoning; Runner parity pending.

Edit this pageGitHubMarkdown

SpatialLadder is a 3B multimodal model specialized in spatial reasoning — answering questions about spatial relations in images and videos. The catalog records it for multimodal/spatial reasoning and image/video QA with model_response artifacts; weights live at hongxingli/SpatialLadder-3B.

No official teaser or method figure is checked into this docs tree.

Code: hongxingli/SpatialLadder-3B.

Build your run

Pipeline
SpatialLadderPipelineworldfoundry.pipelines.video_official.pipeline_official_video
Task
Environmentworldfoundry-unified-cu128
DeviceCUDA 12.8
Runnerworldfoundry.pipeline
Loadingfrom_pretrained
Invocationnative_or_standard_invocation
worldfoundry-eval run \
  spatial-ladder \
  --pipeline.task-profile multimodal-reasoning \
  --json

What it is for

Use spatial-ladder for multimodal reasoning inference on this card.

A 3B-parameter vision-language model; the staged-training design implied by its name (progressive spatial reasoning) is described upstream, and the publishing institution and internals are not recorded in this repository.

Catalog tasks include multimodal-reasoning, spatial-reasoning, image-question-answering, and video-question-answering. Catalog aliases include spatial_ladder and hongxingli/SpatialLadder-3B.

Runtime contract

Optional prompt, image, video, seed, and frames; artifacts spatial-ladder.json.

CLI task profiles: multimodal-reasoning, spatial-reasoning, image-question-answering, video-question-answering, and video_generation. WorldFoundry binds SpatialLadderPipeline.

Family status: Runner parity pending.

Use it as a lightweight spatial-QA judge for generated scenes; the 3B scale runs on modest GPUs.

WorldFoundry route is complete; missing checkpoint/assets are reported at runtime.

Running in WorldFoundry

The catalog launch is worldfoundry-eval run spatial-ladder --pipeline.task-profile multimodal-reasoning. The command builder above switches --pipeline.task-profile among recorded tasks and still emits worldfoundry-eval run.

It runs in the unified environment worldfoundry-unified-cu128 (Python 3.11, CUDA 12.8).

Recorded weight repositories include hongxingli/SpatialLadder-3B. Prepare the environment and check local assets with bash scripts/inference/prepare_model_infer.sh spatial-ladder. Add --download to fetch missing public checkpoints.

Status: Runner parity pending.