Skip to content

OCI AI Blueprints v1.0.7

Choose a tag to compare

View all tags
@grantneumanoracle grantneumanoracle released this 18 Aug 16:59
· 9 commits to main since this release
v1.0.7
0bf705e
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Release Change Audit Report

New Version: v1.0.7
Previous Version: v1.0.5
Generated: 2025-08-07 14:56:56

Repositories Analyzed:

  • Quickstarts Repo (oci-ai-blueprints)
  • Control Plane Repo (corrino)
  • Portal Repo (oci_ai_blueprints_portal)

Version v1.0.7 — Release Notes (Upgrade from v1.0.5)

Release Date: 2025-08-07

1. Executive Summary

v1.0.7 expands GPU-inference capabilities, hardens the secrets workflow, and streamlines API & UX touch-points across the stack.

Key highlights:

  • AMD MI300x support end-to-end (Quickstarts blueprints & exporters, Control Plane node-pool provisioning, Portal library entries)
  • Inference Benchmarking blueprints for offline (vLLM, LLMPerf) and online runs with MLflow logging
  • Secrets Management v2 — namespaced Secrets API plus turnkey examples for opaque, registry, Hugging Face, and NVIDIA NIM credentials
  • Blueprint Library evolution — new categories with richer filtering
  • Auto-generated OpenAPI spec via DRF-Spectacular for easier client generation
  • Portal UI overhaul for faster blueprint discovery and clearer deployment status

2. Repository-Specific Changes

Quickstarts Repo (v1.0.7)

New Features

  • AMD MI300x
    • Metrics exporter & plugin
    • llama4_maverick_amd.json sample blueprint
    • MI300x specs added to RDMA tables
  • Inference Benchmarking
    • Offline (vLLM, LLMPerf) & online GPU benchmarks
    • Sample configs, MLflow integration, dedicated READMEs
  • Secrets Integration
    • Examples covering opaque, registry, Hugging Face, and NIM secrets
  • Partner Blueprints
    • First entry: Llama Stack (partner_blueprints category)

Improvements

  • Parameter rename: local_directory_path → node_directory_path
  • Expanded JSON schema (gpu_benchmarking, model_serving, partner_blueprints)
  • Clearer offline-inference Object Storage mounting docs & IAM policy guidance

Technical Changes

  • Terraform/Helm modules updated for AMD exporter & MI300x node pools
  • docs/sample_blueprints re-organized in quickstart

Bug Fixes & Documentation

  • Purged outdated inference configs
  • READMEs enriched (benchmarking, secrets, version matrix)

Control Plane Repo (v1.0.7)

New Features

  • MI300x hardware detection & recipe deployment
  • Local filesystem volumes (input/output)
  • Subnet-ID syntax validator
  • DRF-Spectacular integration (auto-gen schema.yml)
  • Secrets API adds namespace field with richer examples

Improvements

  • Simplified Swagger routing
  • Extended JSON schema with Partner category & new classifications
  • Clearer documentation around the deployment API control plane endpoint (examples & error messaging)

Bug Fixes

  • Fixed OpenAPI schema generation tests
  • Resolved DRF-Spectacular validation edge cases

Technical Changes

  • Re-structured api/models/*; removed utils/models.py
  • Added drf-spectacular dependency; legacy docs removed
  • New README.openapi.schema.generation.md guide

Documentation

  • Top-level schema.yml with inline examples
  • Deployment logs API docs; grammar & header consistency pass

Portal Repo (v1.0.7)

New Features

  • Blueprint Library
    • New blueprint categories: "Model Serving", "Model Finetuning", "GPU Health Check", "GPU Benchmarking", "Partner Blueprints", "Platform Features" and "Other"
    • Enhanced category filters; description panel moved right
  • Deployments UI
    • Single deployments in table view; grouped deployments in accordion

Improvements

  • Robust content-type detection for Deployment-Content API with explicit error messages

Bug Fixes & Documentation

  • No user-visible defects this cycle; minor doc polish

3. Cross-Repository Impact

  • Unified MI300x Workflows – blueprint samples, provisioning logic, and UI entries align for seamless GPU deployments.
  • Consistent Secrets Pipeline – namespaced Secrets API in Control Plane, turnkey examples in Quickstarts, and Portal-level validation.
  • Blueprint Taxonomy Refresh – new categories for blueprints added; Portal surfaces "Model Serving", "Model Finetuning", "GPU Health Check", "GPU Benchmarking", "Partner Blueprints", "Platform Features" and "Other" categories for discovery.
  • OpenAPI Schema – DRF-Spectacular spec underpins Quickstarts/Portal docs and simplifies client codegen.
  • Local Filesystem Volumes – Quickstarts blueprints leverage new Control Plane volume declarations for high-throughput data access.

4. Upgrade Considerations

  • Parameter Renames – Update any blueprint referencing local_directory_path to node_directory_path.
  • Terraform & Helm – Pull the latest oci_ai_blueprints modules; MI300x node-pool creation requires updated scripts.
  • Secrets API – Add the required namespace field to all Secrets payloads; update client libraries.
  • JSON Schema Updates – Validate blueprints against the updated schema (gpu_benchmarking, model_serving, partner_blueprints).
  • IAM & RBAC – Review updated policy examples to align roles with the new secrets and MI300x workflows.

Upgrade with confidence and enjoy streamlined MI300x GPU workflows, powerful inference benchmarking, and a richer developer experience with OCI AI Blueprints v1.0.7!

Assets 4
Loading