hotels

RHEL AI Model Training Scenario: A Fictional Hotel Group

A fictional example for the Training Large Language Models with Red{nbsp}Hat Enterprise Linux AI (AI0005L) and Deploying Models with Red Hat Enterprise Linux AI (AI0006L) lessons.

These lessons present students with a scenario where a hotel group owning three hotels must train their own LLM, aligned with their business needs.

• The taxonomy with skills and knowledge is at https://github.com/RedHatTraining/AI296-taxonomy-hotels. We cannot store the the taxonomy in a monorepo because InstructLab/RHEL AI needs each taxonomy to live in its own dedicated repository.

• The knowledge documents that support the knowledge contributed to the taxonomy are stored in the business_docs directory.

• The results directory contains the intermediate outputs of the SDG phase to save the student time. With the provided taxonomy, the SDG phase takes ~ 2 hours in a g6e.12xlarge AWS instance.

• The trained model is available at https://huggingface.co/RedHatTraining/AI296-m3diterraneo-hotels. This model has been trained with RHEL AI using a small subset of the complete synthetic dataset and a limited number of epochs, so its performance is limited. The model is available in two formats:

  • The Hugging Face safetensors format that results from training a model with RHEL AI.
  • A quantized version in GGUF (samples_89973_Q4_K_M.gguf), created for serving the model on the CPU-only lab environment of Red Hat Training.