Model Online / On-Site
Duration
Modules 5 modules

Course Objective & Scope

Turkish is a richly agglutinative language: a single inflected word can split into three to five subword tokens in English-centric tokenizers, shrinking effective context length and multiplying inference latency and API cost. Moreover, datasets created via naive machine translation suffer from unnatural 'translationese' phrasing and miss Turkey-specific regulatory and cultural context.

This course covers Turkish tokenizer fertility profiling across model families, SentencePiece/BPE vocabulary extension, continual pre-training and SFT on native Turkish corpora, filtering out translationese artefacts, and benchmarking on Turkish evaluation suites. Participants adapt and benchmark a model for a Turkish enterprise task.

Who Should Attend?

  • AI engineers building LLM systems for Turkish-speaking users and local regulatory domains
  • Data scientists working on Turkish NLP and language model adaptation
  • Technical teams developing Turkish customer service, legal, financial or public-sector assistants
  • Prerequisite: Python, PyTorch and foundational SFT/LoRA experience

Key Takeaways

  • Measure Turkish token inflation (fertility and bytes per token) across model families to pick the most efficient base checkpoint
  • Evaluate when to extend a tokenizer's vocabulary with Turkish morphemes versus relying on native multilingual vocabularies
  • Curate authentic Turkish instruction data while detecting and eliminating machine-translated 'translationese'
  • Adapt models to Turkish formal register, idioms and local domain contexts (KVKK, banking, legal, healthcare)
  • Benchmark adapted models using Turkish evaluation suites and calibrated Turkish LLM judges

Curriculum

Turkish Morphology and Tokenizer Efficiency Analysis
  • Agglutination, vowel harmony and subword fragmentation in BPE/SentencePiece tokenizers
  • Comparing Turkish fertility (tokens per word) and inference speed across Qwen, Gemma, Llama and Mistral
  • When vocabulary extension pays off versus when modern large-vocabulary base models suffice
Vocabulary Extension and Turkish Continual Pre-Training
  • Initialising newly added Turkish subword embeddings in embed_tokens and lm_head
  • Bilingual replay mixing (Turkish + English/code) to boost Turkish fluency without losing reasoning strength
Native Turkish SFT Data and Avoiding 'Translationese'
  • How machine-translated instruction sets corrupt natural Turkish syntax and discourse
  • Synthesising native Turkish corporate correspondence, legal/financial terminology and local context
  • Handling natural Turkish-English technical code-switching in enterprise prompts
Evaluating Turkish Language Models
  • Native Turkish benchmarks versus the pitfalls of translated English benchmarks
  • Morphology-aware grading and calibrating LLM-as-a-judge rubrics in Turkish
Workshop: Adapting and Benchmarking a Turkish Enterprise LLM
  • Profiling candidate base models on Turkish token efficiency and zero-shot accuracy
  • Fine-tuning with native Turkish domain data and producing a comparative evaluation report

Other Courses in This Category

  • 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
  • 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

Get in touch