Product overview and prerequisites
| Item | Details |
|---|---|
| Product | TildeOpen 30B 64k — machine translation, contextual QA and summarisation |
| Build | hf_cqa_v3_mt_v11_global_step424531 |
| API schema | OpenAI chat-completions, served by vLLM |
| Context window | 65,536 tokens (request messages plus max_tokens) |
| Content types | application/json (real-time) · application/jsonlines (Batch Transform) |
| Input | Text only (UTF-8) |
Each request returns one chat-completion object in the same schema. The model supports six tasks through fixed prompt templates: translation, translation with glossary, translation with examples, translation with glossary and examples, contextual question answering and summarisation.
Supported languages and content
Albanian, Bosnian, Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Hungarian, Icelandic, Irish, Italian, Latgalian, Latvian, Lithuanian, Macedonian, Maltese, Montenegrin, Norwegian, Polish, Portuguese, Romanian, Russian, Serbian, Slovak, Slovene, Spanish, Swedish, Turkish, Ukrainian.
Text only. Images, audio and binary document formats (PDF, DOCX) are not accepted and must be converted to text upstream.
Prerequisites
-
An AWS account with permission to create SageMaker models, endpoints and transform jobs, and an IAM execution role for SageMaker.
-
An active subscription to this product in AWS Marketplace. The subscription gives your account access to the model package ARN for your region.
-
Service quota for the recommended instance type in the target region (ml.
<instance>for endpoint usage and for transform job usage). -
Python 3.9+ with boto3 and sagemaker installed, or the AWS CLI, for the examples below.
The full workflow is also provided as a runnable notebook: [Sample notebook URL].