plug.directory
Add my server

Qwen3.8 27B reference

by bottlecapaico.huggingface.bottlecapai/qwen3.8-27b-reference

  • Remote endpoint
  • 2 tools
  • Qwen3_8_27B_reference_respond Get a whole reply in one call; streams chat messages (reasoning bubble, then answer). A reply that needs more than a few minutes of GPU time can fail with "Expired ZeroGPU proxy token"; use /step for those.
  • Qwen3_8_27B_reference_step Advance one reply by a single ZeroGPU reservation. Call repeatedly to finish it. Returns: Streams {"ids", "seconds", "think_seconds"} for this call so far: the new token ids,, the decode time since this call's first new token, and when (on that clock), `</think>` (id 248069, which separates reasoning from the answer) was generated,, or null. Append the last "ids" to, `generated` and call again until the reply ends with an EOS id (248046 or 248044), or reaches max_new_tokens.

As the server's own listing gives them. We never call a tool.