News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
Llama-3-ELYZA-JP-8B

Llama-3-ELYZA-JP-8B

Japanese-tuned Llama 3 model by ELYZA

ELYZA, Inc developed Llama-3-ELYZA-JP-8B to improve Japanese language performance. It builds on Meta’s Llama 3 8B Instruct model through extra pre-training and instruction tuning.

Share on WhatsApp

  • Built by ELYZA, Inc for Japanese users
  • Based on Meta Llama-3-8B-Instruct
  • Enhanced via extra pre-training and tuning
  • Released under Meta Llama 3 Community License
  • Defaults to Japanese language responses

What it is

Llama-3-ELYZA-JP-8B is a large language model created by ELYZA, Inc. It is designed specifically for Japanese usage. The model is based on the Meta Llama-3-8B-Instruct architecture. ELYZA enhanced the base model with additional pre-training and instruction tuning. This process aims to make the assistant more effective in Japanese contexts. The model contains eight billion parameters. It serves as a specialized version of the global Llama 3 family.

How it came about

The model was developed by a team at ELYZA, Inc. Key contributors include Masato Hirakawa, Shintaro Horie, Tomoaki Nakamura, Daisuke Oba, Sam Passaglia, and Akira Sasaki. They built upon the existing Meta Llama 3 framework. The team focused on adapting the model for Japanese speakers. Details of the development process are described in a blog post by the company. The model leverages the foundational work done by Meta on Llama 3.

Inside the model

This model uses a decoder-only transformer architecture. It starts from the Meta Llama-3-8B-Instruct checkpoint. The developers applied additional pre-training on Japanese data. They also performed instruction tuning to align responses with user needs. The default system prompt instructs the model to act as a sincere and excellent Japanese assistant. It is set to reply in Japanese unless told otherwise. The tokenizer and chat template follow the standard Llama 3 format.

What it can do

The model is optimized for Japanese language tasks. It can generate ideas, answer questions, and assist with various text-based requests. The default behavior prioritizes Japanese output. It retains the general capabilities of the underlying Llama 3 8B model. Performance improvements are focused on linguistic nuance and cultural context specific to Japan. Benchmark scores are not provided in the available documentation.

How to use it

The model weights are available for download. It is distributed under the Meta Llama 3 Community License. This license allows for community use with certain conditions. Users can access the model through platforms like Hugging Face. Developers can integrate it into applications using standard transformer libraries. The code examples show how to load the model and tokenizer for inference.

Questions people ask about Llama-3-ELYZA-JP-8B

Who developed Llama-3-ELYZA-JP-8B?

ELYZA, Inc developed the model. The team included Masato Hirakawa, Shintaro Horie, Tomoaki Nakamura, Daisuke Oba, Sam Passaglia, and Akira Sasaki.

What is the base model for this version?

It is based on meta-llama/Meta-Llama-3-8B-Instruct. ELYZA enhanced this foundation with additional training focused on Japanese language usage.

What license covers this model?

The model is released under the Meta Llama 3 Community License. This permits use under the terms set by Meta for the Llama 3 family.

Good reading and sources

    Where this page comes from

    There is no English Wikipedia article on Llama-3-ELYZA-JP-8B yet, so the facts on this page come from the sources listed above: huggingface.co/elyza/Llama-3-ELYZA-JP-8B. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.

    It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

    The weekly digest

    Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.