Downloads · 30 days
16
28% of all-time downloads
alexhegit/so101-simstudio-lab01-pnp-smolvla
so101-simstudio-lab01-pnp-smolvla is a robotics model from alexhegit. Use it for the robotics task on the model card, and read the license before you ship it in a product. It is set up for lerobot. The card lists the license as apache-2.0.
SmolVLA policy fine-tuned from lerobot/smolvlabase on expert demonstrations collected and validated with SO-101 SimStudio (MuJoCo sim2sim, leader-arm teleop).
Downloads · 30 days
16
28% of all-time downloads
All-time downloads
57
Public
Parameters
450M
907 MB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors907 MB · 100%
How the weights are stored.
BF16447M · 99%
From the Hugging Face model README
SmolVLA policy fine-tuned from lerobot/smolvla_base on expert demonstrations collected and validated with SO-101 SimStudio (MuJoCo sim2sim, leader-arm teleop).
Training data: alexhegit/so101-simstudio-lab01-pnp
This Hub revision is the MI300X run: batch 64, 50 000 steps (~3.2M sample updates), final train loss 0.018, ~26 GB of 192 GB HBM. Wall time ~7h 45m on AMD Instinct MI300X (DORobot).
A shorter Strix Halo / 8060S iGPU schedule (batch_size=4, 7500 steps) is documented in Lab 01; it is not this checkpoint.
| Resource | Link |
|---|---|
| SimStudio repo | rocPAI-Forge/so101-simstudio |
| Lab 01 walkthrough (record → train → eval) | labs/lab01_pnp/lab01_pnp.md |
Camera keys in the dataset (camera_top / camera_front / camera_wrist) map to SmolVLA camera1 / camera2 / camera3. See Lab 01 §5 for rename_map and the MI300X training recipe (batch_size=64, steps=50000).
from lerobot.policies.smolvla.modeling_smolvla import SmolVLAPolicy
policy = SmolVLAPolicy.from_pretrained("alexhegit/so101-simstudio-lab01-pnp-smolvla")
Sim2sim eval in MuJoCo: follow §6 Policy eval in lab01_pnp.md (SmolVLA configs under labs/lab01_pnp/configs/rollout_smolvla*.yaml).