Adding LLM private fine-tuning example

examples/llm_private_fine_tuning/.gitattributes

@@ -0,0 +1,2 @@
+components/model-reference-relocation-server/models/gpt2-large/pytorch_model.bin filter=lfs diff=lfs merge=lfs -text
+components/model-relocation-server/models/gpt2/pytorch_model.bin filter=lfs diff=lfs merge=lfs -text

examples/llm_private_fine_tuning/.gitignore

@@ -0,0 +1,162 @@
+.DS_Store
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/

examples/llm_private_fine_tuning/README.md

@@ -0,0 +1,86 @@
+# Private Fine-Tuning
+
+Privately fine-tune a LLM using Parameter Efficient Fine-Tuning with private data.
+
+```mermaid
+graph TD;
+    dataset-relocation-server-->dataset-pvc[(dataset-pvc)];
+    model-relocation-server-->base-model-pvc[(base-model-pvc)];
+    tokenizer-relocation-server-->base-tokenizer-pvc[(base-tokenizer-pvc)];
+    dataset-pvc-->model-peft-server;
+    base-model-pvc-->model-peft-server;
+    base-tokenizer-pvc-->model-peft-server;
+
+    model-peft-server-->fine-tuned-model-pvc[(fine-tuned-model-pvc)];
+
+    model-reference-relocation-server-->reference-model-pvc[(reference-model-pvc)];
+    reference-model-pvc-->model-evaluation-server;
+
+    tensorboard(((tensorboard)))-->fine-tuned-model-pvc;
+    fine-tuned-model-pvc-->model-evaluation-server;
+    fine-tuned-model-pvc-->inference-server;
+    base-tokenizer-pvc-->inference-server;
+```
+
+## [Pushing components to registries](docs/registries.md)
+
+## Updating the models
+
+Updating the base and reference models from GPT2 is a matter of updarting the files and the occurrences of `gpt2` in the code.
+
+> Because we are storing the full model, we need to track the large files using `git lfs` before pushing to a different GitHub repository:  
+
+```bash
+git lfs track components/model-reference-relocation-server/models/gpt2-large/pytorch_model.bin
+git lfs track components/model-relocation-server/models/gpt2/pytorch_model.bin
+```
+
+
+## Local development with KinD
+
+Single node required:
+
+```bash
+kind create cluster
+```
+
+### Installing
+
+With skaffold everything can be built and run with one command for local iteration, the first time will take a while
+because all the images are being created.
+
+```bash
+skaffold run --port-forward=true
+```
+
+> Note that the evaluation server takes over 80 minutes.
+But since the inference server is created in parallel, we can use it while evaluation happens.
+
+
+After running the `skaffold` command, it should display something like:
+```bash
+Waiting for deployments to stabilize...
+ - deployment/model-inference-server is ready.
+Deployments stabilized in 4.135 seconds
+Port forwarding service/model-inference-service in namespace default, remote port 5000 -> http://127.0.0.1:5002
+```
+
+And the inference server is ready to receive requests:
+```bash
+ curl -XPOST 127.0.0.1:5002/prompt -d '{"input_prompt":"What is the best hotel in Seville?"}' -H 'Content-Type: application/json'
+```
+
+## Using KubeFlow
+
+KubeFlow automates all the creation of Kubernetes objects, their synchronization and adds utilities to help us with experiments and visualization.
+
+### Building the pipeline
+
+[Using the notebook](kubeflow/GeneratePipeline.ipynb) or running [pipeline.py](kubeflow/pipeline.py)
+
+### Visualising data in TensorBoard
+
+Select the PVC from `fine-tuned-model-pvc`.
+With the following mount path: `model/gpt2/logs/`.
+
+> Note that since some environments can't easily create Persistent Volumes with `ReadWriteMany`, we have to wait for completion or to delete the board once we analyzed it.

examples/llm_private_fine_tuning/components/dataset-relocation-server/.dockerignore

@@ -0,0 +1 @@
+k8s/

examples/llm_private_fine_tuning/components/dataset-relocation-server/Dockerfile

@@ -0,0 +1,11 @@
+FROM python:3.9
+
+WORKDIR /code 
+COPY ./requirements.txt /code/requirements.txt
+RUN pip install --upgrade pip
+RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
+
+WORKDIR $HOME/app
+COPY . $HOME/app
+
+CMD ["python3", "app.py"]

examples/llm_private_fine_tuning/components/dataset-relocation-server/app.py

@@ -0,0 +1,11 @@
+import argparse
+import shutil
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--dataset_destination_path', type=str, help='dataset destination path')
+
+    opt = parser.parse_args()
+    dataset_destination_path = opt.dataset_destination_path
+    print("Relocating to: " + dataset_destination_path)
+    shutil.copytree("./datasets", dataset_destination_path, dirs_exist_ok=True)