Skip to main content

eclise

Shared research infrastructure: one eclise wheel under the PEP 420 namespace eclise.*. The folder taxonomy is organizational only — imports are always flat eclise.<name>. Heavy stacks live in pixi environments, not in wheel metadata.

Extras (the published wheel)

Extra Import Purpose
utils eclise.utils Research utilities (tensor/module batching, linalg, ARE/ODE, labeled arrays, plotting)
ode eclise.utils.ode torchdiffeq for batch_odeint
datasets eclise.datasets Adapters for consuming external datasets (ImageNet-1k val sampling, label tables)
clips eclise.clips First-party clip-shard data contract (mp4-in-HDF5 + per-frame state labels)
vla eclise.vla Sim-/model-agnostic VLA obs/action schema + websocket env/policy serving
octo / smolvla / pi0 / libero (client aliases) eclise[vla] — talking to a server, not installing its stack
all utils + datasets + clips + vla
dev all + pytest

eclise.octo / smolvla / libero / phyworld / openvid / pi0 / launch ship as inert source in the same wheel. Running a server or generator needs the matching pixi environment (or python -m eclise.launch).

models/vision/ is deliberately absent: a vision-model contract will be created only once it has at least two real consumers.

Pixi environments (from a checkout)

Environment Python Kind Purpose
dev 3.12 library eclise[dev] + contract/launcher tests
libero 3.10 socket (env-server, port 9000) LIBERO + robosuite/MuJoCo harness
octo 3.10 socket (policy, port 8000) Octo JAX 0.4.20 policy server
smolvla 3.12 socket (policy, port 8000) SmolVLA (LeRobot) policy server
pi0 3.11 socket (policy, port 8000) Tooling for upstream openpi (uv / git-lfs / gsutil)
phyworld 3.10 batch Box2D families → eclise.clips shards
phyworld-phyre 3.9 batch PHYRE combinatorial family (phyre==0.2.2)
openvid 3.10 batch OpenVid-1M mp4 → 256px fixed-FPS shards

Installing

Client-side (numpy-1.x-safe eclise.vla only — no torch/jax):

pip install "eclise[octo,libero]"
python -m eclise.launch start octo --setup

--setup clones this repo at eclise-v<wheel version> (unless ECLISE_ROOT already points at a checkout), installs pixi if needed, and runs pixi install -e <env>. start returns a host/port; the caller wires WebsocketPolicyClient / WebsocketEnvClient.

From a checkout:

pixi install -e dev
pixi run -e dev test
pixi run -e libero setup
pixi run -e libero serve --port 9000

Developing

pixi run -e dev test
pixi run -e libero serve
pixi run -e phyworld generate --family collision --grid in_dist_l0 --experiment smoke --limit 32

Editable extras without pixi: pip install -e ".[dev]". Releases are tagged eclise-v0.1.0.

Design rules

  • One distribution, environments at the pixi boundary. Incompatible stacks (JAX vs torch, numpy 1.x vs 2.x) do not share an environment. Wheel extras stay light; pixi features own the heavy pins.
  • Imports are flat, folders carry taxonomy. Moving a package between folders is never a breaking change for consumers.
  • eclise.clips and eclise.vla are contracts. Writers and readers may live in different repos/envs; document schema changes in CHANGELOG.md. eclise[vla] must remain numpy 1.x-compatible.

Package archetypes

Every package is one of three kinds. New packages pick one and follow its rules rather than inventing a fourth layout.

Library extra — published wheel, installed into a consumer's env, no server. utils, datasets, clips, vla.

Interactive socket service — pixi env isolates an incompatible stack behind an eclise.vla websocket. octo, smolvla, pi0, libero.

  • Env: pixi install -e <name> (and pixi run -e <name> setup when the feature defines a setup task).
  • Server logic lives in the package as eclise.<name>.serve (pi0 wraps upstream openpi instead).
  • Launch: python -m eclise.launch start <name>pixi run -e <env> serve.

Batch producer — pixi env; decoupled via the on-disk eclise.clips shard contract (no socket). phyworld, openvid.

  • Env: pixi install -e <name> + pixi run -e <name> setup (scratch data/ / outputs/ symlinks).
  • Job entrypoint: scripts/<verb>.py (+ .sbatch via pixi run -e <env>).
  • Producer verify: scripts/verify_shards.py.
Package Archetype
utils, datasets, clips, vla library extra
octo, smolvla, libero interactive socket service (eclise.<name>.serve)
pi0 interactive socket service (setup_pi0.sh + serve_pi0.sh wrap openpi)
phyworld batch producer (phyworld-phyre is the PHYRE secondary env)
openvid batch producer

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

eclise-0.1.0.tar.gz (311.0 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

eclise-0.1.0-py3-none-any.whl (235.2 kB view details)

Uploaded Python 3

File details

Details for the file eclise-0.1.0.tar.gz.

File metadata

  • Download URL: eclise-0.1.0.tar.gz
  • Upload date:
  • Size: 311.0 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for eclise-0.1.0.tar.gz
Algorithm Hash digest
SHA256 e8f32a5f69eaa68af44f4f597afd938310fa00f08934719a7b8fe57b141804a5
MD5 3eb21b5c8b12b1529c48c2d93acf62bb
BLAKE2b-256 adcfd9240d0c377b64b52f164f2a01b3018790e616ce5b33337bf2a03826b803

See more details on using hashes here.

File details

Details for the file eclise-0.1.0-py3-none-any.whl.

File metadata

  • Download URL: eclise-0.1.0-py3-none-any.whl
  • Upload date:
  • Size: 235.2 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for eclise-0.1.0-py3-none-any.whl
Algorithm Hash digest
SHA256 d6e20e248e8dee96d55c5995b5b5bf53c32b41ca2c2f578280c783dc9593f2cb
MD5 c6cf1c460eaf5d9705bab1d65fc0b08e
BLAKE2b-256 3e995675faaa1a0e5d7cd707e194e1005d795451ec0ae3aae58cee47685cbbc7

See more details on using hashes here.

Release history Release notifications | RSS feed

0.2.0

2 files

This release

0.1.0 This release

2 files

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page