Merge remote-tracking branch 'upstream/main'

Author: dchourasia

examples/kfto-sft-feast-rag/README.md

@@ -0,0 +1,70 @@
+# Fine-Tuning a RAG Model with Feast on OpenShift AI
+
+This project provides an end-to-end example of how to fine-tune a Retrieval-Augmented Generation (RAG) model on **OpenShift AI**. It uses the **Feast** (Feature Store) for efficient retrieval of context and the **Kubeflow Training SDK** to orchestrate the distributed fine-tuning job on the cluster.
+
+The core idea is to enhance a generator model (like BART) by providing it with relevant documents retrieved from a knowledge base at runtime. This notebook handles the entire lifecycle: ingesting data into the feature store, fine-tuning the RAG model on synthetically generated Q&A pairs, and testing the final artifact.
+
+***
+
+## Prerequisites
+
+Before you begin, ensure you have the following setup:
+
+* An OpenShift cluster with OpenShift AI (RHOAI) 2.20+ installed:
+  * The `dashboard`, `trainingoperator` and `workbenches` components enabled.
+* Workbench with medium size container, 1 NVIDIA GPU / 1 AMD GPU accelerator, and cluster storage of 200GB.
+* Sufficient worker nodes for your configuration(s) with NVIDIA GPUs (Ampere-based or newer recommended) or AMD GPUs depending on your environment.
+* A dynamic storage provisioner supporting RWX PVC provisioning.
+* A standalone Milvus deployment. See example [here](https://github.com/rh-aiservices-bu/llm-on-openshift/tree/main/vector-databases/milvus#deployment).
+
+***
+
+## Workbench Setup
+
+You must run this notebook from within an OpenShift AI Workbench. Follow these steps to create one:
+
+* Access the OpenShift AI dashboard:
+* Log in, then go to _Data Science Projects_ and create a project.
+* Once the project is created, click on _Create a workbench_.
+* Then create a workbench with a preferred name and with the following settings:
+  * Select the `PyTorch` (or the `ROCm-PyTorch`) workbench image with the recommended version.
+  * Select the `Medium` as the deployment container size.
+  * Add one NVIDIA / AMD accelerator (GPU) depending on environment.
+  * Create a storage that'll be shared between the workbench and the fine-tuning runs.
+    Make sure it uses a storage class with RWX capability and give it enough capacity according to the size of the model you want to fine-tune.
+    > [!NOTE]
+    > You can attach an existing shared storage if you already have one instead.
+  * Review the storage configuration and click "Create workbench"
+* From the "Workbenches" page, click the icon ![Open icon](https://raw.githubusercontent.com/primer/octicons/main/icons/link-external-16.svg) on the workbench you've just created once it becomes ready.
+* From the workbench, clone this repository, i.e., `https://github.com/opendatahub-io/distributed-workloads.git`
+* Navigate to the `distributed-workloads/examples/kfto-sft-feast-rag` directory and open the `sft_feast_rag_model` notebook
+
+You can now proceed with the instructions from the notebook.
+***
+
+## Workflow Overview
+
+The notebook is structured to guide you through the following key stages:
+
+* **feature_repo/feature_store.yaml**
+  This is the core configuration file for the RAG project's feature store, configuring a Milvus online store on a local provider.
+  * In order to configure Milvus you should:
+    - Update `feature_store.yaml` with your Milvus connection details:
+      - host
+      - port (default: 19530)
+      - credentials (if required)
+* **Environment Setup**: Installs all necessary Python libraries, including the Kubeflow Training SDK, Feast, and Hugging Face Transformers.
+* **Data Ingestion with Feast**:
+  * Loads the `facebook/wiki_dpr` dataset.
+  * Chunks the documents into smaller, manageable passages.
+  * Generates vector embeddings for each passage using a `sentence-transformers` model.
+  * Initializes a Feast feature store and ingests the passages and their embeddings into the online store.
+* **Distributed Training**:
+  * The core training logic is defined in a `main` function.
+  * It defines a `RagSequenceForGeneration` model, combining a question-encoder with a generator model.
+  * It uses a custom `FeastRAGRetriever` to connect the RAG model to the Feast feature store.
+  * The notebook uses the Kubeflow `TrainingClient` to submit this `main` function as a distributed `PyTorchJob` to the OpenShift cluster.
+* **Monitoring**: You can monitor the job's progress directly through its logs and visualize metrics using the integrated TensorBoard dashboard.
+* **Inference and Testing**: After the training job is complete, the final, fine-tuned RAG model is loaded from shared storage for testing.
+
+***

examples/kfto-sft-feast-rag/feature_repo/feature_store.yaml

@@ -0,0 +1,18 @@
+project: ragproject
+provider: local
+registry: data/registry.db
+online_store:
+  type: milvus
+  host: # Insert Milvus route host
+  username: # Insert Milvus username if required
+  password: # Insert Milvus password if required
+  port: 19530
+  vector_enabled: true
+  embedding_dim: 384
+  index_type: FLAT
+  metric_type: COSINE
+offline_store:
+  type: file
+entity_key_serialization_version: 3
+auth:
+    type: no_auth

examples/kfto-sft-feast-rag/feature_repo/ragproject_repo.py

@@ -0,0 +1,48 @@
+from datetime import timedelta
+
+from feast import Entity, FeatureView, Field, FileSource, ValueType
+from feast.data_format import ParquetFormat
+from feast.types import Array, Float32, String
+
+# Define your entity (primary key for feature lookup)
+wiki_passage = Entity(
+    name="passage_id",
+    join_keys=["passage_id"],
+    value_type=ValueType.STRING,
+    description="Unique ID of a Wikipedia passage",
+)
+
+parquet_file_path = "data/wiki_dpr.parquet"
+
+# Define offline source
+wiki_dpr_source = FileSource(
+    name="wiki_dpr_source",
+    file_format=ParquetFormat(),
+    path=parquet_file_path,
+    timestamp_field="event_timestamp",
+)
+
+# Define the feature view for the Wikipedia passage content
+wiki_passage_feature_view = FeatureView(
+    name="wiki_passages",
+    entities=[wiki_passage],
+    ttl=timedelta(days=1),
+    schema=[
+        Field(
+            name="passage_text",
+            dtype=String,
+            description="Content of the Wikipedia passage",
+        ),
+        Field(
+            name="embedding",
+            dtype=Array(Float32),
+            description="vectors",
+            vector_index=True,
+            vector_length=384,
+            vector_search_metric="COSINE",
+        ),
+    ],
+    online=True,
+    source=wiki_dpr_source,
+    description="Content features of Wikipedia passages",
+)