Meet the team that turns your ideas into work.

Discover EightBots
EIGHTBOTS
Contact salesSign up
rinnaJapanese GPT-NeoX 3.6B

Japanese GPT-NeoX 3.6B SFT v2

SFT v2 adapts rinna's Japanese GPT-NeoX 3.6B to conversation. It uses a different training-data split from the earlier SFT release.

Translated instruction data supports Japanese dialogue

The card lists translated subsets of Anthropic HH-RLHF, FLAN and Stanford Human Preferences. It does not release the training data. The conversation template uses Japanese user and system speaker names.Japanese GPT-NeoX 3.6B SFT v2 official model card

Preserve the trained conversation markers

Practical guidance. Use the documented speaker markers when comparing this release with its PPO successor. The original base and first SFT repositories could not be accessed during this review; their names identify ancestry, not verified current downloads.

Specifications & access

Japanese GPT-NeoX 3.6B SFT v2 specifications
DetailValue and conditions
AccessDownloadable
Model identifiersjapanese-gpt-neox-3.6b-instruction-sft-v2
Architecture36 layers with hidden width 2,816.Japanese GPT-NeoX 3.6B SFT v2 official model card
LicenseMIT.Japanese GPT-NeoX 3.6B SFT v2 official model card

Related models

  • Japanese GPT-NeoX 3.6B PPO

    rinna ยท Japanese GPT-NeoX 3.6B

    A Japanese GPT-NeoX 3.6B conversation model aligned with PPO after supervised fine-tuning.

    Launch date not confirmedDownloadable