Open Source RoboticsRobots, software & learning

Spirit-v1.5

Software

VLA model implementation with RoboChallenge execution wrapper and separate base/fine-tuned checkpoints. [1] [3]

Installation & first result →Official websiteGetting started

For
Robot-learning researchers familiar with PyTorch and benchmark-specific robot interfaces [1]
Source rights
Named licenses are documented by part. Review licenses
Unknowns
operating system, compute requirements not fully established. Review gaps

Source researched · Not tested by us
Sources checked · Research details

Licenses & related hardware

Requirements · Research details and dates

On this page

Overview

Spirit-v1.5 provides a VLA implementation and a RoboChallenge execution wrapper, with separate base and task-fine-tuned checkpoints. Choosing the base model for adaptation is different from reproducing one benchmark task with its fine-tuned policy. Preserve that distinction when selecting the checkpoint and runtime configuration. [1] [3]

The named base is Spirit-AI-robotics/Spirit-v1.5. The documented task checkpoint is Spirit-AI-robotics/Spirit-v1.5-for-RoboChallenge-move-objects-into-box, paired with TASK_NAME=move_objects_into_box and scripts/run_robochallenge.sh. That task-specific route is not interchangeable with adapting the base model to another robot or dataset. [1]

Is this for you?

Known limitations
  • Base-card Apache2 and code MIT are distinct declarations; backbone, data and task-checkpoint terms remain separate.
  • Training recommendation8xA10080GB is upstream workload advice, not our requirement measurement.
  • No weights downloaded, model run, hardware command or benchmark reproduction.

Editorial scope assessment from the cited implementation/documentation; no execution or compatibility test.

[1] [2] [3] [4] [30] · Editorial assessment

What you need

Dependencies
  • Repository README Python3.10+ versus base-card recommended Python3.11+
  • Source reports inference testing on NVIDIA A10080GB, not a measured minimum
  • Separate requirements-base.txt and requirements-train.txt; flash-attn needs matching PyTorch/CUDA

[1] [2] [3] [4] [30]

Installation & first result

Open the upstream getting-started guide [1]

  • Choose inference-only or training dependencies and a named base/fine-tuned checkpoint; follow the selected runtime configuration rather than treating all checkpoints as the same robot policy.
  • Preserve exact source/dependency/model versions and evaluate the selected workflow before connecting physical hardware or reusing data.

Editorial scope assessment from the cited implementation/documentation; no execution or compatibility test.

[1] [2] [3] [4] [30] · Editorial assessment

Installation

Choose inference-only or training dependencies and a named base/fine-tuned checkpoint; follow the selected runtime configuration rather than treating all checkpoints as the same robot policy.

Editorial scope assessment from the cited implementation/documentation; no execution or compatibility test.

[1] [2] [3] [4] [30] · Editorial assessment

Hardware, models & data

Inputs and outputs

Visual/state/language policy inputs to action prediction; RoboChallenge-specific wrapper and training source.

Editorial scope assessment from the cited implementation/documentation; no execution or compatibility test.

[1] [3] · Editorial assessment

Project notes

Match the environment to the route

The repository README names Python 3.10 or later and separates inference-only requirements-base.txt from additional training requirements. The base model card recommends Python 3.11 or later and lists pinned packages including torch 2.9.1, transformers 4.57.3 and diffusers 0.36.0. Those two Python statements need reconciliation for the chosen environment; they are not proof that every listed package installs on Python 3.10. [1] [30]

There is also a package-version discrepancy: the pinned repository README lists torch 2.8.0, transformers 4.57.1 and diffusers 0.35.2. The supplied sources do not establish which combination is compatible with the selected checkpoint. Resolve the version set for the chosen code/checkpoint before installation. [1] [30]

RoboChallenge execution needs TASK_NAME, ROBOCHALLENGE_JOB_ID, USER_TOKEN and CKPT_PATH containing model.safetensors; USED_CHUNK_SIZE defaults to 60. The fine-tuning route instead needs a training dataset through DATA_ROOT and PRETRAINED_PATH containing model.safetensors and config.json, plus training dependencies. No credential, dataset or checkpoint payload was supplied or used by this review. [1]

Upstream reports inference testing on an A100 80GB GPU and recommends eight A100 80GB GPUs for the documented multi-GPU training route. These are source-reported environments, not minimum VRAM or throughput measurements made here. Flash-attention also needs a matching PyTorch/CUDA setup. Select inference or training before budgeting compute. [1]

Preserve separate declarations

The repository code is MIT, while the inspected base-checkpoint card declares Apache-2.0 and identifies Qwen3-VL-4B-Instruct as its VLM backbone with a diffusion action head. The public metadata lists the base checkpoint as ungated, but the binaries were not downloaded. A base-card declaration does not automatically settle the separate fine-tuned checkpoint, backbone or training-data terms. [2] [30]

No dependency installation, checkpoint execution, RoboChallenge result, hardware connection or physical manipulation was reproduced. Start by inspecting the documented selected-policy configuration and input/output contract rather than connecting model output directly to a robot. [1]

What we checked

Source research only. Selected implementation/model text, repository tree and license declarations were inspected. No upstream compilation, dependency installation, simulator, network/USB/device integration, data collection, benchmark or physical robot test was performed. Source files are linked rather than redistributed; external services and paid products are not endorsed.

Not yet established

  • Requirements: operating system, host computer
  • Version-specific hardware compatibility (ecosystem tags alone do not establish it)

Updates

Upstream last updated
Not established
Latest release
Not established
Profile updated
Nov 1, 2026
Sources last checked
Nov 1, 2026
Next review
Dec 1, 2026
Reviewed revision
6d67377

Source files and licenses

Distinct robot-facing model contribution with actual implementation/model content and a bounded documented next step. Review covers source artifacts, not operational performance or a complete tested deployment.

LayerAvailability / license
softwareavailable · MIT code; selected base checkpoint card Apache-2.0
Declaration for the scoped source implementation/documentation; dependencies, external services, model/data assets and inherited file-specific terms require separate checks. No blanket redistribution or operational clearance. [1] [2] [3]
model weightsavailable · apache-2.0
Public ungated checkpoint listing and model card inspected at metadata revision 9886fa7552de70b097c4214b7e40e66ff937a6fd; weight binaries not downloaded or executed. This does not settle backbone/data or other checkpoint terms. [30]
documentationavailable · MIT code; selected base checkpoint card Apache-2.0
Declaration for the scoped source implementation/documentation; dependencies, external services, model/data assets and inherited file-specific terms require separate checks. No blanket redistribution or operational clearance. [1] [2] [3]

This profile is scoped to source implementation/model descriptions and an initial inspection workflow; no physical manufacturing design, pretrained checkpoint or dataset is selected for redistribution or tested deployment.

References

Each source links to the exact revision we read. “Used for” shows where this page relies on it.

Correction guidance · Discussion is not enabled · Profile format 0.7.0-draft.5

Change history