Model Online / On-Site
Duration
Modules 5 modules

Course Objective & Scope

For organisations that require strict data sovereignty, predictable unit economics and freedom from vendor lock-in, open-weight models such as Llama, Qwen, Mistral, Gemma and DeepSeek provide a powerful foundation. However, each model family has distinct chat templates, attention mechanisms, vocabulary quirks and quantisation sensitivities.

This course covers the complete open-model adaptation lifecycle: selecting and downloading checkpoints into private registries, adding custom control tokens, fine-tuning with family-specific best practices, quantising to AWQ, GPTQ, FP8 or GGUF with domain calibration data, and serving via vLLM or SGLang inside an enterprise VPC or on-premise cluster.

Who Should Attend?

  • AI and software engineers building sovereign or on-premise LLM applications
  • Data scientists fine-tuning and optimising open-weight models
  • ML platform engineers deploying models in private cloud (VPC) or air-gapped environments
  • Prerequisite: Python, PyTorch and basic Linux/container familiarity

Key Takeaways

  • Account for architectural and tokenizer differences across Llama, Qwen, Mistral, Gemma and DeepSeek during adaptation
  • Add custom control or domain tokens and properly resize and train embedding/output layers
  • Manage model weights securely in private registries and air-gapped enterprise environments
  • Quantise adapted models to AWQ, GPTQ, FP8 or GGUF using domain-representative calibration sets
  • Deploy the customised open model behind an OpenAI-compatible vLLM or SGLang endpoint

Curriculum

Architectural Comparison of Open-Weight Model Families
  • Llama, Qwen, Mistral, Gemma and DeepSeek: GQA, sliding-window attention, tied embeddings and MoE quirks
  • Family-specific chat templates, special tokens and padding/EOS nuances
Adding Custom Tokens and Vocabulary Adaptation
  • Registering new special tokens for tools or schemas and calling resize_token_embeddings
  • Initialising new token embeddings via mean/subword averaging for stable convergence
Enterprise Adaptation and Air-Gapped Workflows
  • Safetensors verification and internal weight mirroring for VPC and air-gapped clusters
  • Self-contained distillation from large open-weight teachers (e.g., 70B to 8B) inside the firewall
Post-Training Quantisation (PTQ) and Serving Readiness
  • Calibrating AWQ and GPTQ quantisation using domain-specific adaptation samples
  • FP8 quantisation for modern GPUs and llama.cpp GGUF conversion for edge/CPU targets
  • Auditing post-quantisation accuracy retention against the FP16/BF16 checkpoint
Workshop: End-to-End Open Model Customisation and Serving
  • Fine-tuning an open-weight checkpoint on enterprise data and packaging into AWQ/FP8
  • Serving via vLLM and benchmarking latency and throughput

Other Courses in This Category

  • Turkish Language Models: Tokenization, Evaluation and Adaptation

    Course on adapting LLMs for Turkish: morphological challenges in BPE tokenization, vocabulary extension, continual pre-training, Turkish evaluation benchmarks and cultural/regulatory alignment.

    LLM Customisation & Model Adaptation Corporate
    Live Course
    5 modules
    View
  • Distributed LLM Training with PyTorch FSDP and DeepSpeed

    Advanced course on scaling LLM training across multi-GPU and multi-node clusters using Data, Tensor, Pipeline and Context Parallelism, DeepSpeed ZeRO-1/2/3, PyTorch FSDP2 and communication profiling.

    LLM Customisation & Model Adaptation Corporate
    Live Course
    5 modules
    View
  • Reinforcement Learning for Reasoning Models (GRPO and RLVR)

    Advanced hands-on course on training DeepSeek-R1-style reasoning models using Reinforcement Learning with Verifiable Rewards (RLVR) and critic-free Group Relative Policy Optimization (GRPO).

    LLM Customisation & Model Adaptation Corporate
    Live Course
    5 modules
    View
  • Dataset Quality and Decontamination for LLM Training

    Course on auditing and cleaning LLM training datasets: exact and MinHash/semantic deduplication, train-test and benchmark decontamination, perplexity/judge quality filtering and PII scrubbing.

    LLM Customisation & Model Adaptation Corporate
    Live Course
    5 modules
    View
  • Prompting, RAG or Fine-Tuning: LLM Adaptation Decision Framework

    Hands-on course on choosing and combining context engineering, RAG and fine-tuning based on knowledge freshness, style/behaviour alignment, latency and TCO, producing an adaptation decision matrix.

    LLM Customisation & Model Adaptation Corporate
    Live Course
    5 modules
    View
  • Model and Data Readiness: Base Model Selection and Data Suitability

    Course on selecting open-weight or commercial base models by size, licence, tokenizer and hardware footprint, and auditing enterprise data volume, quality and privacy readiness before adaptation.

    LLM Customisation & Model Adaptation Corporate
    Live Course
    5 modules
    View

Get in touch