DEDUPED Hot AI repos, turned into use cases Get 5 a day

How to speed up chat replies with draft tokens with Janas

Enable draft decoding for supported Qwen and Gemma models to get faster chat replies.

Who it is forA user chatting with a Qwen3.5, Qwen3.6, Qwen3-Next, or Gemma 4 model who wants faster answers to ordinary questions.

The problemWriting a reply one token at a time is slow on a model this large, even when the machine can run it.

The outcomeChat replies at a higher rate, with the same text, when the model supports drafts.

What you need first

  • Linux x86-64 (Ubuntu 22.04, Debian 12 and later) or Windows 10 (1903 or later) or 11, x86-64
  • A terminal on the machine
  • Enough free disk space for the model you pick; the README lists Qwen3-Next-80B at 48.4 GB
  • A machine with 32 GB of RAM is the one the README measured on; other machines are untested

Steps

  1. Install and start the chat as in the Try it section
  2. Choose a supported model such as Qwen3-Next-80B-A3B or Gemma-4-26B-A4B-it
  3. Turn drafts on for the chat (inferred flag form): ./janas-chat --mtp (inferred, check the README)
  4. Ask an ordinary English question and time the reply
  5. Compare the tokens-per-second with and without drafts

Prompt for your agent

In the janas repo, start ./janas-chat --mtp with Qwen3.6-35B-A3B selected and ask it to explain how bread is made at home. Report the tokens-per-second shown and confirm the reply is complete.

Caveat. The README says the draft gain depends on the text, and on Italian prose one model kept only about a third of its drafts and gained little.

Share: X LinkedIn Hacker News Email Copy link

Disclaimer

Deduped lists open-source repositories that belong to other people. We do not own, maintain, audit, or vouch for any of them, and a repo appearing here is not an endorsement or a safety check. The use cases, steps, and prompts are written by AI and may be incomplete, outdated, or just wrong. Where a use case is marked "Seen in the wild", we link to public posts where someone described something similar, but we have not verified those posts and the use case itself is still our AI's own write-up. Use all of this at your own risk. Only run code, commands, or agents on systems and accounts you are authorized to use, and read the code and every command before you run it, especially anything that touches credentials, files, or the network. Nothing here is professional, security, legal, or financial advice. All trademarks and project names belong to their respective owners, and their use here does not imply affiliation. Content is provided as is, without warranties of any kind. Found a mistake, a safety problem, or something you want removed? Email [email protected] and we will look at it promptly. Full disclaimer page