Sakana AIEvoVLM-JP
Llama-3-EvoVLM-JP-v2
The second EvoVLM release merges three Llama 3-based checkpoints. It supports sequences that alternate between text and images.
Research release
The input can contain more than one image
The provided loader uses Mantis. The card credits Japanese visual question-answering data and translated ShareGPT4V data, as well as the three source model developers.Llama-3-EvoVLM-JP-v2 official model card
Working with this release
Practical guidance. Give each image a clear role in the prompt. Check whether a comparison refers to the correct image instead of combining details from separate inputs.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | SakanaAI/Llama-3-EvoVLM-JP-v2 |
| Ancestry | Meta-Llama-3-8B-Instruct, Mantis-8B-siglip-llama3 and Bunny-v1.1-Llama-3-8B-V.Llama-3-EvoVLM-JP-v2 official model card |
| Terms | Meta Llama 3 Community License; the card describes experimental research use.Llama-3-EvoVLM-JP-v2 official model card |
Related models
- EvoVLM-JP-v1-7B
Sakana AI ยท EvoVLM-JP
A vision-language model that combines Japanese language understanding with LLaVA through evolutionary merging.
Launch date not confirmedResearch releaseDownloadable