xiaoye-123/alphabrain-rlt-qwenoft-5traj-alltasks-libero-goal
robotics model by xiaoye-123
Older snapshot: check the original source for current information.
From the publisher
Model card & documentation
Source previewRead the publisher’s intended use, setup instructions, evaluations and limitations. The original model card is the source of truth.
Lightweight RL-Token (RLT) checkpoint trained on top of a frozen QwenOFT VLA. This is not the official full-scale AlphaBrain RLT release. It is a single-GPU, reduced-budget reproduction: about 1/24 of the official interaction volume. On LIBERO-Goal it reaches 68.5% success, which is below both the frozen QwenOFT base (89.1%) and the official RLT checkpoint (92%). RL-Token keeps a pretrained Vision-Language-Action (VLA) model frozen and trains a small Encoder + Actor + Critic stack with off-policy RL. The base VLA was itself supervised-finetuned from Qwen2.5-VL-3B on 5 human trajectories per task. That SFT data is not reused during this RL run; the RLT heads only see environment rollouts. Use this repo if you want a low-data / single-GPU reference point. Do not treat it as a drop-in replacement for the official 400-iter weights. The VLA is frozen. A transformer bottleneck encoder compresses VLA hidden features into a compact RL state. A…
Read the full model card ↗ · Preview checked 2026-11-19T08:19:22.938Z
Inside the original model card — Document outline
- AlphaBrain RLT (low-data, LIBERO-Goal, all 10 tasks)
- Model overview
- How this differs from the official RLT
- Architecture
- Training details
- Stage context
- RL budget (this run)
- What this budget implies
- Evaluation
- Per-task results
- Comparison
- Files
Headings are captured from the source. Links open the publisher’s document, not a locally hosted copy.
Documentation belongs to its respective authors. Reported project/model license: mit. A listing is not a grant of reuse or training rights. Confirm the document’s own terms at the source.
Model overview
xiaoye-123/alphabrain-rlt-qwenoft-5traj-alltasks-libero-goal is a robotics model repository published by xiaoye-123 on Hugging Face. The source reports the pytorch library.
This page summarizes Hub metadata. For intended use, training data, evaluation results and limitations, consult the original model card.
Model facts
- Task
- robotics
- Library
- pytorch
- Recent downloads (30 days)
- 0
- Cumulative likes
- 0
- Hugging Face trending score
- 0
- Reported safetensors parameters
- Not reported
- Architecture
- Not reported
- License
- mit
- Access
- Not gated by Hugging Face
- Created
- 2026-10-06T06:04:51.000Z
- Last modified
- 2026-10-06T06:24:22.000Z
Compatibility and lineage
Reported languages: en
Reported base models: AlphaBrainGroup/qwenoft-5traj-libero-goal
Parameter count is not a RAM/VRAM requirement. Check precision, quantization, context length and runtime compatibility in the model card; no hardware or API-cost claim is inferred here.
Use and evaluate
Check license terms and access requirements first. Review the model card’s documented loading instructions, evaluations and safety limitations. Benchmark scores are not imported by this directory, and popularity is not an accuracy ranking.
No model weights are downloaded or executed by AltAPIs. Never enable remote model code without reviewing it.
Source and freshness
Source: Hugging Face Hub. Metadata observed 2026-10-06T06:39:56.505Z. Daily imports are snapshots, not real-time monitoring.
Popularity and source listings do not establish security, suitability, licensing rights or benchmark performance.
Related models
1977kantomoyuki/act_so101_test_20260920_143445
abdulrahimmirani/pi05-piperx-pick-cube-ft01-simdata