| name | genai-dac-specialist |
| description | Expert guidance for OCI Dedicated AI Clusters — private LLM hosting, fine-tuning, sizing, and cost optimisation for enterprise deployments |
| version | 1.0.0 |
| platform | ["claude-code","cline","cursor","roocode"] |
| activation | {"cline":"@skills/genai-dac-specialist/SKILL.md","cursor":"@skills/genai-dac-specialist/SKILL.md"} |
OCI GenAI Dedicated AI Cluster (DAC) Specialist
You are an expert in Oracle's Dedicated AI Clusters for private LLM hosting and fine-tuning.
When to Use
- Deploying private LLM infrastructure on OCI
- Sizing and costing DAC deployments
- Fine-tuning models with proprietary data
- High-throughput inference requirements
Overview
A DAC is a private, isolated GPU cluster in OCI for:
- Hosting LLMs without shared infrastructure
- Fine-tuning with LoRA adapter training on your data
- High throughput with multiple endpoints per cluster
- Data sovereignty — data never leaves your tenancy
Sizing Guide
| Workload | Units | Approx Monthly | Use Case |
|---|
| Dev/Test | 2-5 | Verify pricing | Experimentation |
| Production | 5-15 | Verify pricing | Standard workloads |
| High Volume | 15-30 | Verify pricing | Enterprise scale |
| Enterprise | 30-50 | Verify pricing | Maximum throughput |
IMPORTANT: Always verify current pricing at oracle.com/cloud/price-list
Model Selection for DAC
| Model | Best For | Context |
|---|
| Cohere Command A | Complex reasoning, enterprise | 256K |
| Cohere Command R/R+ | RAG-optimized, retrieval | 128K |
| Llama 3.3 70B | Fine-tuning, customization | 128K |
| Llama 3.1 405B | Maximum capability | 128K |
Fine-Tuning
Methods
| Method | Models | Use Case |
|---|
| LoRA | Llama 3.3 70B, Llama 3.1 405B | Domain-specific adaptation |
| T-few | Cohere Command | Quick enterprise customization |
| Vanilla | Cohere Command | Full fine-tune |
Process
1. Upload training data → Object Storage (JSONL format)
2. Create fine-tuning job → DAC processes
3. Deploy custom endpoint → Use fine-tuned model
4. Monitor performance → Iterate
Also Consider: AI Blueprints LoRA Fine-Tuning
Before setting up a DAC, check if the AI Blueprints LoRA Fine-Tuning blueprint meets your needs:
https://github.com/oracle-quickstart/oci-ai-blueprints
Terraform Quick Start
resource "oci_generative_ai_dedicated_ai_cluster" "main" {
compartment_id = var.compartment_id
display_name = "production-dac"
type = "HOSTING"
unit_count = 10
unit_shape = "LARGE_COHERE"
}
resource "oci_generative_ai_endpoint" "chat" {
compartment_id = var.compartment_id
dedicated_ai_cluster_id = oci_generative_ai_dedicated_ai_cluster.main.id
model_id = "cohere.command-a-03-2025"
display_name = "chat-endpoint"
}
Cost Optimization
- Start with on-demand endpoints — validate before committing to DAC
- Begin with 5 units, scale based on actual usage
- Multi-model endpoints — share DAC across use cases
- Monitor utilization — scale down if under 60%
Documentation
Cline Activation
To use this skill in Cline, reference it at the start of your message:
@skills/genai-dac-specialist/SKILL.md
Size a Dedicated AI Cluster for a production Cohere Command A deployment with 500 concurrent users.
Or in a .clinerules workflow:
## DAC Sizing
When asked about private LLM hosting on OCI, load @skills/genai-dac-specialist/SKILL.md. Always verify pricing, recommend starting small (5 units), and check AI Blueprints before recommending a full DAC.
Triggers: Dedicated AI Cluster, DAC, private LLM hosting, OCI fine-tuning, GPU cluster sizing, model fine-tuning OCI