MLL-Lab/LAGEN-datasets

LAGEN datasets

Project resources: LAGEN collection.

Training data, episode-level evaluations and paper experiment manifests.

Experiment	Entry
Sim2Real calibration	30-case held-out calibration
Visual history	Visual history
Latency in prompt	Latency in prompt
Latency transfer	Latency transfer
Task transfer	Task transfer
VLA fine-tuning scope	VLA fine-tuning scope
Mean vs. profile training	Mean vs. profile training
Observation stride	Observation stride
Context window	Context window

Resource	Repository
benchmark-models	MLL-Lab/LAGEN-models
profiles	MLL-Lab/LAGEN-profiles

Benchmark release inventory

Artifact retention

HAIC and Extreme Parkour releases are retired. Model bundles retain the published evaluation checkpoint, or the latest checkpoint when no evaluation selection exists. Optimizer and trainer recovery state are not release assets. Identical dataset copies use the canonical task paths. Original configurations and experiment evidence remain source records.

Figure 2 latency degradation

Latency-degradation catalogue preserves all 15 curves, exact evaluation records, fixed checkpoint sources and provenance gaps. Six MIKASA checkpoints are reused, and all six confirmed game checkpoints are published in benchmark-models with unchanged SHA256 and size.

Consolidated historical sources

Source archive index records the original dataset trees, exact fixed references, model metadata and coverage verification. These archives retain unresolved source bindings. Hardware records are in profiles.

Root archive/ was retired on 2026-10-08. Current task datasets, formal evaluations and nested experiment archives retain their existing locations. Historical source inventories describe the earlier consolidation, not current download availability.