LLM-jp-3 VILA 14B
LLM-jp-3 VILA 14B adds image understanding to LLM-jp-3 13B Instruct. It combines a 428M SigLIP encoder with a 32M projector.
Research release
Projector training precedes joint language adaptation
The first stage trains the projector. Later stages train the projector and language model using image-text, interleaved and instruction data. The card discloses unresolved synthdog-ja dataset licensing: the authors proceeded on a CC-BY-SA assumption after no clarification was received.LLM-jp-3 VILA 14B official model card
Review the disclosed data caveat
Practical guidance. This release is useful for reproducing the earlier Japanese VLM approach. For a new redistribution plan, inspect the stated source-data conditions as well as the weight license.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | llm-jp-3-vila-14b |
| Language component | LLM-jp-3 13B Instruct.LLM-jp-3 VILA 14B official model card |
| Weights license | Apache 2.0; the card also requires compliance with OpenAI terms because GPT-4 synthetic data was used.LLM-jp-3 VILA 14B official model card |
Related models
- LLM-jp-3 13B Instruct
LLM-jp / NII LLMC · LLM-jp-3
The original 13B LLM-jp-3 instruction model with a 4,096-token context and Apache 2.0 weights.
Launch date not confirmedResearch releaseDownloadable - LLM-jp-4 VL 9B
LLM-jp / NII LLMC · LLM-jp-4 VL
A Japanese-English vision-language reasoning model that changes both the language backbone and training data from the beta.
Launch date not confirmedResearch releaseDownloadable