Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
54 changes: 52 additions & 2 deletions .github/workflows/ci.yml
Original file line number Diff line number Diff line change
Expand Up @@ -127,6 +127,55 @@ jobs:
python examples/run_quickstarts.py --profile cpu
--output-dir "${RUNNER_TEMP}/cuphoton-quickstart"

test-nooa:
name: xLens NOOA / ${{ matrix.name }} / py${{ matrix.python }}
runs-on: ${{ matrix.runner }}
timeout-minutes: 15
strategy:
fail-fast: false
matrix:
include:
- name: linux-x64
runner: ubuntu-24.04
python: "3.12"
- name: linux-x64
runner: ubuntu-24.04
python: "3.13"
- name: linux-arm64
runner: ubuntu-24.04-arm
python: "3.13"
env:
CUPHOTON_REQUIRE_NOOA_TESTS: "1"
steps:
- uses: actions/checkout@df4cb1c069e1874edd31b4311f1884172cec0e10 # v6
with:
persist-credentials: false
- uses: astral-sh/setup-uv@d0cc045d04ccac9d8b7881df0226f9e82c39688e # v6
with:
version: "0.12.1"
enable-cache: true
- name: Install the isolated worker runtime
run: |
sudo apt-get update
sudo apt-get install -y bubblewrap apparmor libseccomp2
uv python install "${{ matrix.python }}"
uv sync --locked --python "${{ matrix.python }}" --extra dev --extra xlens-nooa
- name: Allow bubblewrap user namespaces on Ubuntu
run: |
sudo tee /etc/apparmor.d/cuphoton-ci-bwrap >/dev/null <<'PROFILE'
abi <abi/4.0>,
profile cuphoton-ci-bwrap /usr/bin/bwrap flags=(unconfined) {
userns,
}
PROFILE
sudo apparmor_parser --replace /etc/apparmor.d/cuphoton-ci-bwrap
- name: Verify bubblewrap namespace support
run: >-
bwrap --unshare-all --unshare-user --disable-userns --die-with-parent
--ro-bind / / true
- name: Exercise real NOOA with local model fixtures
run: uv run --no-sync pytest -q -rs tests/xlens

test-gpu:
name: xFit, xRay and GPU quickstarts / L40G / py3.12
# Only vetted copies and maintained branches may use self-hosted GPUs.
Expand Down Expand Up @@ -497,7 +546,7 @@ jobs:
ci-required:
# PR checks do not satisfy the merge gate before package and GPU qualification.
name: ${{ github.event_name == 'pull_request' && 'ci-pr-checks' || 'ci-required' }}
needs: [lint, test-cpu, test-gpu, coverage, package, conda]
needs: [lint, test-cpu, test-nooa, test-gpu, coverage, package, conda]
if: ${{ always() }}
runs-on: ubuntu-24.04
timeout-minutes: 5
Expand All @@ -506,13 +555,14 @@ jobs:
env:
LINT_RESULT: ${{ needs.lint.result }}
CPU_RESULT: ${{ needs.test-cpu.result }}
NOOA_RESULT: ${{ needs.test-nooa.result }}
GPU_RESULT: ${{ needs.test-gpu.result }}
COVERAGE_RESULT: ${{ needs.coverage.result }}
REQUIRE_FULL_CI: ${{ github.event_name == 'push' && github.repository == 'NVIDIA/cuPhoton' }}
PACKAGE_RESULT: ${{ needs.package.result }}
CONDA_RESULT: ${{ needs.conda.result }}
run: |
for result in "${LINT_RESULT}" "${CPU_RESULT}" "${COVERAGE_RESULT}" "${CONDA_RESULT}"; do
for result in "${LINT_RESULT}" "${CPU_RESULT}" "${NOOA_RESULT}" "${COVERAGE_RESULT}" "${CONDA_RESULT}"; do
test "${result}" = success || exit 1
done
if [ "${REQUIRE_FULL_CI}" = true ]; then
Expand Down
15 changes: 10 additions & 5 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
Expand Up @@ -2,13 +2,18 @@

## Unreleased

### xLens

- Add NOOA as an opt-in `xlens-nooa` extra for iterative evidence queries
and isolated Python on Linux with Python 3.12/3.13, bubblewrap and libseccomp.
Select it with `backend = "nooa"`; basic inference remains the default.
- Add `api = "responses"` alongside the default Chat Completions transport, with
documentation for optional `gpt-6-rosalind` configuration. Model access and
endpoint compatibility depend on the user's provider; a live Rosalind
integration has not been qualified.

### Breaking changes

- Remove the optional NOOA xLens engine, its installation extra, and its
backend and generated-code configuration. xLens uses its built-in
evidence-based inference engine. Remove `backend`, `max_iterations`,
`max_tool_calls`, and `max_evidence_bytes` from existing `[inference]`
tables, including `backend = "basic"`; these settings are no longer accepted.
- Move Pandas and NumExpr into the `catalog` extra for xScan ingestion,
and Pydantic into the `ximage` extra for xImage configuration. Select
these extras for the corresponding workflows.
Expand Down
30 changes: 30 additions & 0 deletions THIRD_PARTY_NOTICES.md
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,9 @@ requirements, compiled header code, and the CFITSIO library bundled in Linux
wheels are documented below. `uv.lock` records the reproducible resolution for
the project Python dependency profiles and the `native-build` and `notebook`
groups, including the optional DragonHPC and mpi4py packages.
The optional NOOA investigation profile is also recorded in the lock; its
Python dependencies, system bubblewrap executable and libseccomp library are
installed separately and are not bundled in cuPhoton artifacts.
The site-provided MPI implementation is not part of that lock. Isolated
PEP 517 builds resolve build-system requirements separately from `uv.lock`;
their `build` rows are labeled `not locked` below.
Expand Down Expand Up @@ -87,6 +90,8 @@ more than one license.
| `ximage`, `dev` | `pydantic>=2.11,<3` | `2.14.1` | `MIT` | [Pydantic](https://github.com/pydantic/pydantic) | `uv / PyPI` |
| `ximage` | `zarr>=3,<4` | `3.4.1` | `MIT` | [Zarr](https://github.com/zarr-developers/zarr-python) | `uv / PyPI` |
| `rubin` | `lsst-images[https]==30.0.10` | `30.0.10` | `BSD-3-Clause`; optional dependency tree includes `starlink-pyast` under `LGPL-3.0-or-later` | [LSST Images](https://github.com/lsst/images) | `uv / PyPI` |
| `xlens-inference` | `httpx>=0.28,<1` | `0.28.1` | `BSD-3-Clause` | [HTTPX](https://github.com/encode/httpx) | `uv / PyPI; also selected by xlens-nooa` |
| `xlens-nooa` | `nooa==0.0.12` | `0.0.12` | `Apache-2.0`; upstream distribution includes its own third-party notices | [NVIDIA OO Agents](https://github.com/NVIDIA-NeMo/labs-OO-Agents) | `uv / PyPI; Linux, Python 3.12/3.13 only` |
| `base` | `PyYAML>=6.0` | `6.0.3` | `MIT` | [PyYAML](https://github.com/yaml/pyyaml) | `uv / PyPI` |
| `base` | `scipy>=1.13` | `1.18.0` | `BSD-3-Clause`; distributions include separately licensed components | [SciPy](https://github.com/scipy/scipy) | `uv / PyPI` |
| `torch` | `torch>=2.13,<3` | `2.13.0` | `Apache-2.0 AND Apache-2.0 WITH LLVM-exception AND BSD-2-Clause AND BSD-3-Clause AND BSL-1.0 AND MIT` | [PyTorch](https://github.com/pytorch/pytorch) | `uv / PyPI` |
Expand Down Expand Up @@ -114,6 +119,31 @@ more than one license.
| `dev` | `types-PyYAML>=6.0.12` | `6.0.12.20260906` | `Apache-2.0` | [typeshed](https://github.com/python/typeshed) | `uv / PyPI` |
| `dev` | `coverage>=7.10.6,<8` | `7.16.2` | `Apache-2.0` | [Coverage.py](https://github.com/coveragepy/coveragepy/tree/7.16.2) | `uv / PyPI; development only` |

## Optional NOOA investigation runtime

The `xlens-nooa` extra installs NOOA and its declared Python dependencies from
upstream distributions. cuPhoton loads NOOA in an isolated worker for iterative
evidence queries and Python execution. NOOA source, model weights and dependency
package trees are not bundled in cuPhoton wheels or source archives.

NOOA 0.0.12 is Apache-2.0 licensed. Its declared dependencies include AnyIO,
HTTPX, LiteLLM, MessagePack, OpenInference instrumentation, Pydantic and urllib3;
their dependencies and licenses are separate. In particular:

| Component | Reference version | License and upstream notice | Use |
| --- | --- | --- | --- |
| NOOA | `0.0.12` | [Apache-2.0](https://github.com/NVIDIA-NeMo/labs-OO-Agents/blob/main/LICENSE); NVIDIA Corporation & Affiliates | Agent loop in the worker. Its distribution includes additional third-party notices. |
| LiteLLM | `1.109.0` | [MIT](https://github.com/BerriAI/litellm/blob/main/LICENSE); its distribution also contains tokenizer assets with their own notices | NOOA dependency, installed separately. Model requests in this integration are brokered by cuPhoton's HTTP client. |
| OpenInference LiteLLM instrumentation | `0.1.55` | [Apache-2.0](https://github.com/Arize-ai/openinference/blob/main/LICENSE) | NOOA dependency, installed separately; the cuPhoton adapter suppresses NOOA's automatic tracing initialization. |
| bubblewrap | Site-provided | [LGPL-2.0-or-later](https://github.com/containers/bubblewrap/blob/main/COPYING) | Separate Linux executable used to start the isolated worker. It is not linked into or bundled with cuPhoton. |
| libseccomp | Site-provided | [LGPL-2.1-only](https://github.com/seccomp/libseccomp/blob/main/LICENSE) | System shared library loaded through Python ctypes only for NOOA, to build the worker's syscall filter. Not bundled with cuPhoton. |

The Python versions above reflect the inspected distributions and lock. Consult
each installed distribution's license files and third-party notices for its
complete contents. Bubblewrap and libseccomp come from the user's operating-system
provider; their packages and system dependencies are outside `uv.lock`. See the
[xLens installation and execution requirements](docs/components/xlens.md#configure-optional-investigation).

## Optional distributed runtime inventory

The Dragon executor requires DragonHPC. MPI collective aggregation requires
Expand Down
Loading
Loading