ByteDanceSeed-OSS
Seed-OSS 36B Instruct
Seed-OSS 36B Instruct is a downloadable language model with a controllable reasoning budget.
Model design
The 36B family was trained on 12 trillion tokens and supports a native 512K context. It uses grouped-query attention and provides separate base and instruction-tuned releases.Seed-OSS 36B Instruct official model card
Practical use
Practical guidance. Start with a modest reasoning budget for a defined task. Increase it only if the result improves, and measure the actual memory use of long inputs.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | Seed-OSS-36B-Instruct |
| Context window | 524,288 tokens Published checkpoint context limit. Seed-OSS 36B Instruct official model card |
| License | Apache 2.0Seed-OSS 36B Instruct official model card |
| Access | Weights are published in the official ByteDance Seed repository.Seed-OSS 36B Instruct official model card |
Related models
- Seed-OSS 36B Base
ByteDance · Seed-OSS
Seed-OSS 36B Base is the pretrained checkpoint that includes synthetic instruction data.
Launch date not confirmedDownloadable - Seed-OSS 36B Base woSyn
ByteDance · Seed-OSS
Seed-OSS Base woSyn omits the synthetic instruction data used in the other 36B base release.
Launch date not confirmedDownloadable