korean-llm-v3
A Korean LLM project sharing lessons from its implementation across versions
SITE CAPTUREBased on information published on the service's website. Selection and review policy
WHAT IT DOES
What does it do?
Korean LLM v4 is a PyTorch-based project for implementing and training a Korean LLM. The current page describes its progression from an early experiment of about 50M, through v1 at about 541M and v2 at about 1.09B, to v4. It also notes that the default sequence lengths in the model class and actual training settings differ.
KEY FEATURES
Key features
- Korean LLM implementation and training project
- Development history from early experiments through v4
- Explanation of default sequence lengths in the model class and training settings