Nekomata 7B
Nekomata 7B continues pretraining Qwen-7B on about 30B tokens of Japanese and English text.
Research release
The smaller release has the longer context
The original Qwen vocabulary has more than 150,000 tokens. rinna highlights more efficient Japanese tokenization than its earlier Youri series. The card specifies a 32,768-token sequence limit.Nekomata 7B official model card
Check token usage on real Japanese documents
Practical guidance. Use the actual tokenizer to compare document lengths. Japanese character counts do not map to a fixed token count, and differences in vocabulary can affect cost and usable input length.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | nekomata-7bnekomata-7b-gguf |
| Context window | 32,768 tokens Maximum sequence length stated in the model card. Nekomata 7B official model card |
| License | Tongyi Qianwen License Agreement; this is not an Apache-licensed Qwen2 release.Nekomata 7B official model card |
| Ancestry | Qwen/Qwen-7B.Nekomata 7B official model card |
Related models
- Nekomata 14B
rinna · Nekomata
Nekomata 14B continues pretraining Qwen-14B on about 66B tokens of Japanese and English text.
Launch date not confirmedResearch releaseDownloadable - Nekomata 7B Instruction
rinna · Nekomata
This model instruction-tunes Nekomata 7B, retaining its Qwen ancestry and using the Alpaca task format.
Launch date not confirmedDownloadable