nano-laser 0.1.0__tar.gz

nano_laser-0.1.0/.github/workflows/prcheck.yml

@@ -0,0 +1,29 @@
+name: PR Checks
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+on:
+  pull_request:
+    branches:
+      - main
+    paths:
+      - 'src/nano_laser/**'
+
+jobs:
+  changelog:
+    name: CHANGELOG
+    runs-on: ubuntu-latest
+    if: github.event_name == 'pull_request'
+
+    steps:
+    - uses: actions/checkout@v4
+      with:
+        fetch-depth: 0
+
+    - name: Check that CHANGELOG has been updated
+      run: |
+        # If this step fails, this means you haven't updated the CHANGELOG.md
+        # file with notes on your contribution.
+        git diff --name-only $(git merge-base origin/main HEAD) | grep '^CHANGELOG.md$' && echo "Thanks for helping keep our CHANGELOG up-to-date!"

nano_laser-0.1.0/.github/workflows/pre-commit.yml

@@ -0,0 +1,18 @@
+name: Prek checks
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+jobs:
+  prek:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.13"
+
+      - uses: j178/prek-action@v1

nano_laser-0.1.0/.github/workflows/publish.yml

@@ -0,0 +1,50 @@
+name: publish
+
+# Publishes to PyPI when a release tag is pushed, with trusted publishing: no API token is stored.
+# PyPI side: add a trusted publisher for BurkimbIA/nano-laser, workflow publish.yml, environment pypi.
+on:
+  push:
+    tags: ["v*.*.*"]
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          version: "latest"
+
+      - name: Check that the tag is the project version
+        run: |
+          version="$(uv version --short)"
+          if [ "${GITHUB_REF_NAME}" != "v${version}" ]; then
+            echo "Tag ${GITHUB_REF_NAME} does not match project.version ${version}" >&2
+            exit 1
+          fi
+
+      - name: Build
+        run: uv build
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+  publish:
+    needs: build
+    runs-on: ubuntu-latest
+    environment:
+      name: pypi
+      url: https://pypi.org/project/nano-laser/
+    permissions:
+      id-token: write
+    steps:
+      - uses: actions/download-artifact@v4
+        with:
+          name: dist
+          path: dist/
+
+      - uses: pypa/gh-action-pypi-publish@release/v1

nano_laser-0.1.0/.github/workflows/tests.yml

@@ -0,0 +1,42 @@
+name: tests
+
+on:
+  pull_request:
+  push:
+    branches: [main]
+
+jobs:
+  tests:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.13"
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: false
+          version: "latest"
+
+      - name: Cache uv
+        uses: actions/cache@v4
+        with:
+          path: ~/.cache/uv
+          key: uv-${{ runner.os }}-${{ hashFiles('uv.lock') }}
+          restore-keys: |
+            uv-${{ runner.os }}-
+
+      - name: Install just
+        uses: extractions/setup-just@v3
+
+      - name: Install dependencies (dev)
+        run: just install
+
+      - name: Run tests
+        run: just test

nano_laser-0.1.0/.gitignore

@@ -0,0 +1,10 @@
+# Python-generated files
+__pycache__/
+*.py[oc]
+build/
+dist/
+wheels/
+*.egg-info
+
+# Virtual environments
+.venv

nano_laser-0.1.0/.pre-commit-config.yaml

@@ -0,0 +1,23 @@
+repos:
+- repo: https://github.com/astral-sh/uv-pre-commit
+  rev: 0.13.4
+  hooks:
+  - id: uv-lock
+
+- repo: https://github.com/astral-sh/ruff-pre-commit
+  rev: v0.14.10
+  hooks:
+  - id: ruff
+    args: [ --fix ]
+    types_or: [ python, pyi ]
+
+- repo: https://github.com/pre-commit/pre-commit-hooks
+  rev: v6.0.0
+  hooks:
+  - id: check-merge-conflict
+  - id: mixed-line-ending
+
+exclude: |
+  unsloth_compiled_cache/.*
+  |llms-finetune/gemma3/unsloth_compiled_cache/.*
+  |.*\.ipynb$

nano_laser-0.1.0/.python-version

@@ -0,0 +1 @@
+3.13

nano_laser-0.1.0/CHANGELOG.md

@@ -0,0 +1,21 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## Unreleased
+
+## v0.1.0 - 2026-10-26
+
+### Added
+
+- `nano_laser`: LASER2 and LASER3 encoders in plain torch, a drop-in replacement for `laser_encoders` 0.0.2 without fairseq.
+- Parity tests against embeddings from `laser_encoders`, run when the models are cached locally.
+- A Colab notebook comparing `nano_laser` with `laser_encoders` on all 147 LASER3 models; all match.
+- A workflow that publishes to PyPI with trusted publishing when a `vX.Y.Z` tag is pushed.
+
+### Security
+
+- Bumped the `uv-pre-commit` hook to uv 0.13.4. Earlier versions ship the compromised emSign ECC Root CA - G3 root, which certifi 2026.10.19 and uv 0.13.4 remove.

nano_laser-0.1.0/CONTRIBUTING.md

@@ -0,0 +1,8 @@
+Contributing
+============
+
+Unless you explicitly state otherwise, any contribution intentionally submitted
+for inclusion in the work by you shall be licensed as in [LICENSE](LICENSE),
+without any additional terms or conditions.
+
+See [the readme](README.md#contributing) for contribution workflow suggestions.

nano_laser-0.1.0/LICENSE

@@ -0,0 +1,29 @@
+BSD 3-Clause License
+
+Copyright (c) 2026, BurkimbIA
+Copyright (c) Facebook, Inc. and its affiliates (LASER, the code adapted in nano_laser)
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+3. Neither the name of the copyright holder nor the names of its
+   contributors may be used to endorse or promote products derived from
+   this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

nano_laser-0.1.0/PKG-INFO

@@ -0,0 +1,108 @@
+Metadata-Version: 2.5
+Name: nano-laser
+Version: 0.1.0
+Summary: LASER sentence encoders in plain torch: a drop-in replacement for laser_encoders, without fairseq
+Author: Facebook AI Research
+Author-email: MNIKIEMA <nikiemamahamadi01@gmail.com>, Salif Sawadogo <sawallesalfo@gmail.com>
+License-Expression: BSD-3-Clause
+License-File: LICENSE
+Requires-Python: >=3.10
+Requires-Dist: numpy>=1.21.3
+Requires-Dist: sacremoses==0.1.0
+Requires-Dist: sentencepiece>=0.1.99
+Requires-Dist: torch>=2.5
+Requires-Dist: unicategories>=0.1.2
+Description-Content-Type: text/markdown
+
+# nano-laser
+
+The [LASER](https://github.com/facebookresearch/LASER) sentence encoders (LASER2 and the LASER3 language
+models) in plain torch. It is a drop-in replacement for [`laser_encoders`](https://pypi.org/project/laser-encoders/),
+without fairseq: the same API, the same checkpoints, the same embeddings.
+
+## Install
+
+The package is not on PyPI yet; install it from GitHub:
+
+```bash
+pip install git+https://github.com/BurkimbIA/nano-laser.git
+# or, in a uv project
+uv add git+https://github.com/BurkimbIA/nano-laser.git
+```
+
+Append `@<commit or tag>` to the URL to pin a version. It needs Python ≥ 3.10 and torch ≥ 2.5.
+
+## Usage
+
+Replace the import, nothing else:
+
+```python
+from nano_laser import LaserEncoderPipeline  # was: from laser_encoders import LaserEncoderPipeline
+
+encoder = LaserEncoderPipeline(lang="mos_Latn")
+embeddings = encoder.encode_sentences(["Laafi bala?"], normalize_embeddings=True)  # (1, 1024)
+```
+
+`initialize_tokenizer`, `initialize_encoder`, `SentenceEncoder`, `LaserTokenizer` and the module layout
+(`nano_laser.models`, `nano_laser.laser_tokenizer`, `nano_laser.download_models`, `nano_laser.language_list`)
+mirror `laser_encoders`. The models are downloaded from the same place into the same cache,
+`~/.cache/laser_encoders`, so models already downloaded by `laser_encoders` are reused.
+
+Like `laser_encoders`, the encoder takes one sentence per batch by default. Batch on GPU with the
+`SentenceEncoder` options, which `initialize_encoder` and `LaserEncoderPipeline` now pass on:
+
+```python
+encoder = LaserEncoderPipeline(lang="dyu_Latn", max_sentences=64)  # or max_tokens=..., cpu=True, fp16=True
+```
+
+## Parity
+
+Against `laser_encoders` 0.0.2 (fairseq 0.12.2, torch 2.1.2), on 85 sentences of 0 to 2,000 tokens:
+
+| model | max abs. difference (cpu) | min cosine (cpu and gpu, batched or not) |
+| --- | --- | --- |
+| LASER2 (LSTM) | 1.3e-6 | 0.9999996 |
+| LASER3 mos_Latn, dyu_Latn, fuv_Latn (transformer) | 1.4e-5 | 0.9999998 |
+
+Every LASER3 model is also checked, on a Colab T4 with 18 sentences, one at a time and in batches of 8
+([`notebooks/check_laser3.ipynb`](notebooks/check_laser3.ipynb)): all 147 give the same tokens and embeddings,
+with a min cosine of 0.9999997 and a max abs. difference of 2.0e-5. They come in three kinds:
+
+| LASER3 models | languages |
+| --- | --- |
+| 12-layer transformer, `maxpool` pooling | 135 |
+| LSTM, like LASER2 | 10: `als_Latn`, `cym_Latn`, `hye_Armn`, `kat_Geor`, `khk_Cyrl`, `khm_Khmr`, `ltz_Latn`, `mlt_Latn`, `quy_Latn`, `ydd_Hebr` |
+| 12-layer transformer, `cls` pooling on a leading `<s>` | 2: `lij_Latn`, `wol_Latn` |
+
+The results per language are in [`scripts/results/laser3_parity.jsonl`](scripts/results/laser3_parity.jsonl).
+
+`tests/test_parity.py` checks this against `tests/data/reference.npz` whenever the models are in the cache
+(or in `LASER_MODEL_DIR`); it is skipped otherwise, so CI downloads nothing.
+
+## Differences from laser_encoders
+
+- No fairseq: the transformer encoder and fairseq's `Dictionary` are re-implemented in `nano_laser/models.py`.
+- Checkpoints load with `torch.load(weights_only=True)`: they only hold tensors and an `argparse.Namespace`.
+- LASER3 works with `max_sentences > 1`. In `laser_encoders` on torch ≥ 1.12, fairseq's BetterTransformer path fails on
+  left-padded batches (`RuntimeError: Mask Type should be defined`).
+- With `sentemb_criterion="cls"`, each sentence of a batch gets its own `<s>` state whatever the padding side.
+  `laser_encoders` takes the `<s>` states in position order, which is only right when they share a position:
+  true for the two `cls` models (`lij_Latn`, `wol_Latn`), which pad on the right.
+- Importing the package does not call `logging.basicConfig`, and a failed download raises instead of saving the
+  error page as the model.
+
+## Development
+
+```bash
+just install  # uv sync and the pre-commit hooks
+just test
+just lint
+```
+
+To regenerate the parity reference, see `tests/data/make_reference.py`.
+
+## License
+
+[BSD 3-Clause](LICENSE), Copyright (c) 2026, BurkimbIA. The code adapted from LASER keeps its copyright,
+Copyright (c) Facebook, Inc. and its affiliates, also under the BSD license; each such file says so in its header.
+The LASER model weights are downloaded from Facebook and are not covered by this license.

nano_laser-0.1.0/README.md

@@ -0,0 +1,92 @@
+# nano-laser
+
+The [LASER](https://github.com/facebookresearch/LASER) sentence encoders (LASER2 and the LASER3 language
+models) in plain torch. It is a drop-in replacement for [`laser_encoders`](https://pypi.org/project/laser-encoders/),
+without fairseq: the same API, the same checkpoints, the same embeddings.
+
+## Install
+
+The package is not on PyPI yet; install it from GitHub:
+
+```bash
+pip install git+https://github.com/BurkimbIA/nano-laser.git
+# or, in a uv project
+uv add git+https://github.com/BurkimbIA/nano-laser.git
+```
+
+Append `@<commit or tag>` to the URL to pin a version. It needs Python ≥ 3.10 and torch ≥ 2.5.
+
+## Usage
+
+Replace the import, nothing else:
+
+```python
+from nano_laser import LaserEncoderPipeline  # was: from laser_encoders import LaserEncoderPipeline
+
+encoder = LaserEncoderPipeline(lang="mos_Latn")
+embeddings = encoder.encode_sentences(["Laafi bala?"], normalize_embeddings=True)  # (1, 1024)
+```
+
+`initialize_tokenizer`, `initialize_encoder`, `SentenceEncoder`, `LaserTokenizer` and the module layout
+(`nano_laser.models`, `nano_laser.laser_tokenizer`, `nano_laser.download_models`, `nano_laser.language_list`)
+mirror `laser_encoders`. The models are downloaded from the same place into the same cache,
+`~/.cache/laser_encoders`, so models already downloaded by `laser_encoders` are reused.
+
+Like `laser_encoders`, the encoder takes one sentence per batch by default. Batch on GPU with the
+`SentenceEncoder` options, which `initialize_encoder` and `LaserEncoderPipeline` now pass on:
+
+```python
+encoder = LaserEncoderPipeline(lang="dyu_Latn", max_sentences=64)  # or max_tokens=..., cpu=True, fp16=True
+```
+
+## Parity
+
+Against `laser_encoders` 0.0.2 (fairseq 0.12.2, torch 2.1.2), on 85 sentences of 0 to 2,000 tokens:
+
+| model | max abs. difference (cpu) | min cosine (cpu and gpu, batched or not) |
+| --- | --- | --- |
+| LASER2 (LSTM) | 1.3e-6 | 0.9999996 |
+| LASER3 mos_Latn, dyu_Latn, fuv_Latn (transformer) | 1.4e-5 | 0.9999998 |
+
+Every LASER3 model is also checked, on a Colab T4 with 18 sentences, one at a time and in batches of 8
+([`notebooks/check_laser3.ipynb`](notebooks/check_laser3.ipynb)): all 147 give the same tokens and embeddings,
+with a min cosine of 0.9999997 and a max abs. difference of 2.0e-5. They come in three kinds:
+
+| LASER3 models | languages |
+| --- | --- |
+| 12-layer transformer, `maxpool` pooling | 135 |
+| LSTM, like LASER2 | 10: `als_Latn`, `cym_Latn`, `hye_Armn`, `kat_Geor`, `khk_Cyrl`, `khm_Khmr`, `ltz_Latn`, `mlt_Latn`, `quy_Latn`, `ydd_Hebr` |
+| 12-layer transformer, `cls` pooling on a leading `<s>` | 2: `lij_Latn`, `wol_Latn` |
+
+The results per language are in [`scripts/results/laser3_parity.jsonl`](scripts/results/laser3_parity.jsonl).
+
+`tests/test_parity.py` checks this against `tests/data/reference.npz` whenever the models are in the cache
+(or in `LASER_MODEL_DIR`); it is skipped otherwise, so CI downloads nothing.
+
+## Differences from laser_encoders
+
+- No fairseq: the transformer encoder and fairseq's `Dictionary` are re-implemented in `nano_laser/models.py`.
+- Checkpoints load with `torch.load(weights_only=True)`: they only hold tensors and an `argparse.Namespace`.
+- LASER3 works with `max_sentences > 1`. In `laser_encoders` on torch ≥ 1.12, fairseq's BetterTransformer path fails on
+  left-padded batches (`RuntimeError: Mask Type should be defined`).
+- With `sentemb_criterion="cls"`, each sentence of a batch gets its own `<s>` state whatever the padding side.
+  `laser_encoders` takes the `<s>` states in position order, which is only right when they share a position:
+  true for the two `cls` models (`lij_Latn`, `wol_Latn`), which pad on the right.
+- Importing the package does not call `logging.basicConfig`, and a failed download raises instead of saving the
+  error page as the model.
+
+## Development
+
+```bash
+just install  # uv sync and the pre-commit hooks
+just test
+just lint
+```
+
+To regenerate the parity reference, see `tests/data/make_reference.py`.
+
+## License
+
+[BSD 3-Clause](LICENSE), Copyright (c) 2026, BurkimbIA. The code adapted from LASER keeps its copyright,
+Copyright (c) Facebook, Inc. and its affiliates, also under the BSD license; each such file says so in its header.
+The LASER model weights are downloaded from Facebook and are not covered by this license.

nano_laser-0.1.0/RELEASE.md

@@ -0,0 +1,100 @@
+# Release process
+
+This guide describes how to release `nano-laser`.
+
+## Versioning
+
+The project follows [Semantic Versioning](https://semver.org/):
+
+- `MAJOR`: breaking API, configuration, or workflow changes.
+- `MINOR`: backwards-compatible features.
+- `PATCH`: backwards-compatible bug fixes, dependency fixes, and security patches.
+
+The release version is the value of `project.version` in [`pyproject.toml`](pyproject.toml). The release tag and changelog heading must use the same version, formatted as `vX.Y.Z`.
+
+## Prepare the release
+
+Start from an up-to-date `main` branch and create a release branch:
+
+```bash
+git checkout main
+git pull --ff-only origin main
+git checkout -b chore/release-vX.Y.Z
+```
+
+Choose the appropriate version bump:
+
+```bash
+uv version --bump major
+uv version --bump minor
+uv version --bump patch
+```
+
+This updates `pyproject.toml` and `uv.lock`. Do not edit only one of them.
+
+Update `CHANGELOG.md` as follows:
+
+1. Move the release-ready entries from `Unreleased` into `## vX.Y.Z - YYYY-MM-DD`.
+2. Add a new empty `## Unreleased` section directly below the changelog intro.
+3. Review the entries for user-facing accuracy and remove internal-only detail that does not help users upgrade.
+
+## Verify and open the release PR
+
+Run the release checks:
+
+```bash
+uv lock --check
+just format
+just lint
+just typecheck
+just test
+just pre-commit
+```
+
+Review the resulting diff. It should include the intended version change, lockfile update, and changelog update, with any relevant documentation or configuration changes.
+
+```bash
+git add pyproject.toml uv.lock CHANGELOG.md
+git commit -m "chore: release vX.Y.Z"
+git push -u origin chore/release-vX.Y.Z
+```
+
+Open a pull request targeting `main` and merge it after review and all enabled CI checks pass.
+
+## Tag and publish
+
+After the release PR is merged, tag the exact `main` commit with an annotated tag, then push it:
+
+```bash
+git checkout main
+git pull --ff-only origin main
+git tag -a vX.Y.Z -m "Release vX.Y.Z"
+git push origin vX.Y.Z
+```
+
+Create the GitHub Release using generated notes, then review and edit the notes before publishing:
+
+```bash
+gh release create vX.Y.Z --title "vX.Y.Z" --generate-notes
+```
+
+## Post-release checks
+
+- Confirm the Git tag points to the merged release commit.
+- Confirm the GitHub Release title, version, and notes are correct.
+- Confirm that `CHANGELOG.md` has an empty `Unreleased` section for the next change.
+
+## Publish to PyPI
+
+Pushing the tag runs [`publish.yml`](.github/workflows/publish.yml): it checks that the tag is `v` + `project.version`,
+builds the sdist and wheel with `uv build`, and publishes them to PyPI through trusted publishing.
+
+One-time setup, before the first release:
+
+- On PyPI, add a trusted publisher for the project `nano-laser`: owner `BurkimbIA`, repository `nano-laser`,
+  workflow `publish.yml`, environment `pypi` (a "pending publisher" if the project does not exist yet).
+- On GitHub, create the `pypi` environment (Settings → Environments), optionally with required reviewers so that
+  each upload waits for an approval.
+
+After the workflow succeeds, confirm that https://pypi.org/project/nano-laser/ shows the new version and that
+`pip install nano-laser==X.Y.Z` installs it.

nano_laser-0.1.0/justfile

@@ -0,0 +1,35 @@
+# display help information
+default:
+    @just --list
+
+
+# clean the project
+clean:
+    rm -rf .venv/
+
+
+# install the dependencies
+install:
+    uv sync --all-groups
+    uvx prek install
+
+
+# format with ruff
+format:
+    uvx ruff format
+
+# lint with ruff
+lint:
+    uvx ruff check
+
+# Typecheck using ty
+typecheck:
+    uvx ty check
+
+# Run pre-commit [lint, format]"
+pre-commit: lint format
+    uvx prek run
+
+# test with pytest
+test:
+    uv run pytest