Skip to content
RobothecaBeta
Log in

BitVLA

  • AI model
  • Humanoids and robot learning

BitVLA is a 1-bit vision-language-action model for robotic manipulation, with its weights quantized to 1.58 bits to cut memory use. It is evaluated on LIBERO tasks and VQA benchmarks.

Source on GitHub: github.com/ustcwhy/BitVLA (opens in a new tab). Facts read at commit 75ff1b0 (opens in a new tab).

Licenses

Code
MIT (opens in a new tab)

Read by Robotheca from LICENSE (opens in a new tab) (MIT) at commit 75ff1b0 (opens in a new tab), on 16 Oct 2026.

Commercial use
Yes

What it does

  • Vision-language-action model whose weights are quantized to 1.58-bit precision, with 8-bit activations.
  • Pre-trained with next-action prediction on a subset of the Open X-Embodiment dataset.
  • Fine-tuned on LIBERO task suites (Spatial, Object, Goal, Long) using the OpenVLA-OFT training recipe.
  • Also includes evaluation code for visual question answering benchmarks through LMMS-Eval.
  • Pre-trained checkpoints and fine-tuned LIBERO models are published on Hugging Face; pre-training code is not yet released.

Written by Robotheca from the project's README (opens in a new tab) at commit 75ff1b0.

Getting started

Difficulty
Advanced
Hardware it needs, as the project states it
Pre-training took about 14 days on 16 NVIDIA H800 (80GB) GPUs. Evaluation is run in an nvidia/pytorch 24.07 Docker container with GPUs; no other hardware requirements are stated.

Maintenance

Status
Active: a commit in the last 90 days
Newest commit
6 Sept 2026 on main
Latest release
None published on GitHub

Checked on GitHub on 16 Oct 2026.

Links

Robots that can run it

No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.

Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.