Downloads · 30 days
5
18% of all-time downloads
starvla_qwenpi_pretrain_qwen3_4B
FsqZ/starvla_qwenpi_pretrain_qwen3_4B
starvla_qwenpi_pretrain_qwen3_4B is a robotics model from FsqZ. Use it for the robotics task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
StarVLA QwenPI stage-1 pretrain on OXE subset (Bridge + RT-1 + LIBERO ×4).
All-time downloads
28
Public
Repo size
16.4 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt16.4 GB · 100%
At a glance
- Task
- Robotics
- License
- apache-2.0
- Access
- Public
- Created
- May 19, 2026
- Updated
- May 19, 2026
- SHA
- 25c1f75d
Model card
From the Hugging Face model README
starvla_qwenpi_pretrain_qwen3_4B
StarVLA QwenPI stage-1 pretrain on OXE subset (Bridge + RT-1 + LIBERO ×4).
- Base VLM: Qwen3-VL-4B-Instruct-Action
- Action head: DiT-B, action_horizon=8, action_dim=7
- Pretrain steps: 10k–80k checkpoints provided
- Final loss: ~0.18 (action_dit_loss)
- Recommended ckpt for downstream finetune:
steps_60000_pytorch_model.pt
Usage
--trainer.pretrained_checkpoint <local-path>/steps_60000_pytorch_model.pt
Trained for the 创智夏季营 starVLA 实训 (Calvin ABC→D benchmark).
- Task
- Robotics
- License
- apache-2.0
- Created
- May 19, 2026
- Updated
- May 19, 2026