Thank you for sending your enquiry! One of our team members will contact you shortly.
Thank you for sending your booking! One of our team members will contact you shortly.
Duration 14 hours
Course Outline
AI Sovereignty and Localized LLM Deployment
- Risks associated with cloud-based LLMs: data retention, model training on user inputs, and foreign jurisdiction.
- Ollama architecture: model server, registry management, and OpenAI-compatible API endpoints.
- Comparative analysis with vLLM, llama.cpp, and Text Generation Inference frameworks.
- Model licensing considerations: Terms for Llama, Mistral, Qwen, and Gemma.
Installation and Hardware Configuration
- Deploying Ollama on Linux with CUDA and ROCm accelerator support.
- CPU-only fallback mechanisms and AVX/AVX2 optimization strategies.
- Docker implementation and persistent volume mapping configurations.
- Multi-GPU environments and VRAM allocation protocols.
Model Lifecycle Management
- Retrieving models from the Ollama registry: execution of 'ollama pull llama3'.
- Importing GGUF formatted models from HuggingFace and TheBloke repositories.
- Quantization level selection: Tradeoffs between Q4_K_M, Q5_K_M, and Q8_0 formats.
- Dynamic model switching and concurrent loading constraints.
Custom Modelfile Development
- Syntax for Modelfiles: Utilization of FROM, PARAMETER, SYSTEM, and TEMPLATE directives.
- Tuning parameters: Temperature, top_p, and repeat_penalty adjustments.
- System prompt engineering for defined, role-specific operational behaviors.
- Creation and local registry publication of customized models.
API Integration Standards
- OpenAI-compatible endpoint configuration: /v1/chat/completions structure.
- Implementation of streaming responses and JSON mode outputs.
- Integration with LangChain, LlamaIndex, and bespoke application architectures.
- Authentication protocols and rate limiting via reverse proxy infrastructure.
Performance Optimization Strategies
- Context window sizing and KV cache management techniques.
- Batch inference processing and parallel request handling.
- CPU thread allocation and NUMA topology awareness.
- Monitoring GPU utilization and memory pressure metrics.
Security and Compliance Posture
- Network isolation strategies for model serving endpoints.
- Implementation of input filtering and output moderation pipelines.
- Comprehensive audit logging of prompts and model completions.
- Model provenance tracking and cryptographic hash verification.
Requirements
- Intermediate proficiency in Linux and container administration.
- High-level understanding of machine learning principles and transformer architectures.
- Familiarity with RESTful APIs and JSON data formats.
Audience
- AI engineers and developers seeking to replace cloud LLM APIs with sovereign alternatives.
- Organizations with strict data sensitivity requirements that preclude the use of cloud-based models.
- Government and defense teams requiring air-gapped language model capabilities.
