IWSLT 2026 Submission — Access Request
These weights are shared for the IWSLT 2026 Simultaneous Speech Translation Shared Task evaluation only. Access is granted manually by the maintainer. Please provide accurate contact details below so that legitimate requests can be reviewed quickly.
Log in or Sign Up to review the conditions and access this model content.
IWSLT 2026 Submission — Model Weights
This repository contains the fine-tuned model weights accompanying our submission to the IWSLT 2026 Simultaneous Speech Translation Shared Task, for the English → Chinese and English → German directions (main track).
Contents
| Directory | Direction | Approximate size |
|---|---|---|
model_en_zh/ |
English → Chinese | ~66 GB |
model_en_de/ |
English → German | ~66 GB |
Each directory follows the standard Hugging Face / vLLM layout
(config.json, tokenizer files, sharded *.safetensors) and can be
loaded directly with vllm serve <path> once downloaded.
Usage
These weights are designed to be served by the Docker container shipped
separately to the IWSLT 2026 organising committee
(submission_main_track.tar.gz). The container's entrypoint downloads
the relevant sub-directory automatically given a Hugging Face token.
Access
This repository is gated: please click "Request access" above and fill in your affiliation. Requests are reviewed manually and typically processed within one business day.