Model Online / On-Site
Duration
Modules 5 modules

Course Objective & Scope

Most failed fine-tuning initiatives stem from picking the wrong foundation model, overlooking poor tokenizer efficiency on target languages or domain terms, missing restrictive model licence clauses, or feeding noisy corporate logs into training. This course teaches how to run a rigorous model and data suitability audit before spending GPU hours.

Participants compare the Llama, Qwen, Mistral, Gemma and DeepSeek families across architecture, licensing and tokenizer fertility, size models from 1B to 70B+ against hardware constraints, and audit enterprise datasets for coverage, label noise and PII/KVKK compliance. The deliverable is a model comparison and data readiness report.

Who Should Attend?

  • Data scientists and AI engineers leading model selection and adaptation projects
  • ML platform engineers planning GPU capacity and model-serving infrastructure
  • Technical leads evaluating data governance, licensing and privacy constraints
  • Prerequisite: Python and foundational LLM knowledge

Key Takeaways

  • Select the right parameter scale and model family for a task's complexity and latency budget
  • Decide whether to start adaptation from a base, instruct or reasoning checkpoint
  • Measure tokenizer fertility and token inflation on Turkish and domain-specific corpora
  • Audit internal datasets for sample sufficiency, distribution coverage, label noise and PII exposure
  • Evaluate open-weight model licences (Apache 2.0, MIT, Llama, Gemma terms) for commercial deployment

Curriculum

Foundation Model Selection Criteria
  • Open-weight models (Qwen, Llama, Gemma, Mistral) versus managed fine-tuning APIs
  • Base, Instruct and Reasoning checkpoints: choosing the right starting point
  • Dense versus Mixture-of-Experts (MoE) trade-offs in training and inference
Sizing, Hardware Footprint and Tokenizer Suitability
  • Matching task narrowness to 1B–8B small models versus 32B–70B+ models
  • Testing tokenizer fertility (tokens per word) on Turkish and domain corpora
  • Estimating training and serving VRAM footprints upfront
Data Suitability and Readiness Audit
  • Estimating minimum high-quality sample counts by task type (100 vs 10,000+ rows)
  • Input-output distribution profiling, edge-case coverage and label consistency
  • Detecting PII, confidential secrets and memorisation risks in training corpora
Licensing, Governance and Data Residency
  • Open-source versus open-weight licences and commercial usage thresholds
  • Terms-of-service restrictions on distilling or training with outputs from proprietary models
Workshop: Model Comparison and Data Readiness Report
  • Profiling token lengths, fertility and quality on a sample organisational dataset
  • Running zero-shot suitability checks across candidate models and writing the selection report

Other Courses in This Category

  • Quality, Cost and Risk Analysis in LLM Adaptation

    Course on modelling total cost of ownership (data curation, training and 24/7 serving), quality-versus-regression trade-offs and operational/regulatory risks for LLM customisation projects.

    LLM Customisation & Model Adaptation Corporate
    Live Course
    5 modules
    View
  • Baselines and Success Metrics: Pre-Adaptation Benchmarking for LLMs

    Course on building golden test sets, establishing zero/few-shot baselines, defining deterministic and LLM-as-a-judge metrics and setting acceptance thresholds before starting model adaptation.

    LLM Customisation & Model Adaptation Corporate
    Live Course
    5 modules
    View
  • Fine-Tuning Fundamentals: Introduction to Customising LLMs

    Foundational hands-on course on how LLM fine-tuning works: dataset formatting, chat templates, running a first adapter training job with Hugging Face TRL and Unsloth, and avoiding overfitting.

    LLM Customisation & Model Adaptation Corporate
    Live Course
    5 modules
    View
  • Advanced Fine-Tuning: Continual Pre-Training, Long Context and Model Merging

    Advanced course covering continual pre-training (CPT), RoPE long-context extension, multi-turn tool-calling fine-tuning, catastrophic forgetting mitigation and model merging (SLERP, TIES, DARE) with MergeKit.

    LLM Customisation & Model Adaptation Corporate
    Live Course
    5 modules
    View
  • Supervised Fine-Tuning (SFT): Engineering and Optimisation

    Deep-dive engineering course on Supervised Fine-Tuning (SFT): completion-only loss masking, sequence packing, FlashAttention-2, NEFTune, learning rate schedules and production workflows with TRL SFTTrainer and Axolotl.

    LLM Customisation & Model Adaptation Corporate
    Live Course
    5 modules
    View
  • Parameter-Efficient Fine-Tuning (PEFT): Architectures and Multi-Adapter Serving

    Course on PEFT architectures—LoRA, DoRA, rsLoRA, VeRA, LoftQ, IA3 and Prefix Tuning—VRAM memory budgeting and serving dozens of dynamic adapters on a single base model with vLLM and SGLang.

    LLM Customisation & Model Adaptation Corporate
    Live Course
    5 modules
    View

Get in touch