input context -2k

#2
by LittleTaro - opened

I have one question about the context-length. why 2k? When encountering long texts exceeding 2k tokens, is it necessary to split them before translation?

Hi @LittleTaro
Yes, you absolutely must split texts longer than 2k tokens and the safest way is to split your long text is into chunks of paragraphs .
2k limit is a deliberate design choice as translateGemma is designed to run locally on consumer hardware so a smaller context window significantly reduces memory usage (KV cache) and latency.
Thanks

Sign up or log in to comment