flide 0.1.0

English · 简体中文 · 日本語

News · Speed · Exact answers · Quick start · Models · Options

Flide serves decision models. A request holds a state, such as an agent's trajectory or a document, and typed questions about it. Flide forks the state into all of its questions inside one forward pass, so the state is read once and every question gets the answer it would get if it were asked alone. Answers come back as probabilities through the /v1/systemone API that vLLM and SGLang also serve.

🔥 News

• 🔥 2026-10-24 We released Flide 0.1.0 on PyPI. Welcome to have a try.

Speed

Exact answers

Quick start

1. Install

pip install flide

The NVIDIA driver is all a machine needs. PyTorch brings the CUDA runtime, and Flide ships its kernels compiled for the H100 and H200 and for the A40, A10 and RTX 30 series. On other GPUs they compile once at the first start. Images and video also need torchvision, installed from the same index as torch.

PyTorch's wheels on PyPI need driver 580 or newer. With an older driver, install PyTorch's CUDA 12.6 build first, with torchvision.

pip install torch torchvision --index-url https://download.pytorch.org/whl/cu126
pip install flide

2. Start a server

Any chat model on Qwen3.5 or Gemma 4 serves as it is and answers through its next token.

flide serve --model Qwen/Qwen3.5-4B

OneJev and Open-Jev models read their answers their own way, and Flide serves them as they are too.

flide serve --model OmniJev/OneJev1.6-4B

flide serve listens on all interfaces at port 8000, as vLLM does; pass --host 127.0.0.1 to keep it local.

3. Send a request

curl http://localhost:8000/v1/systemone -H "Content-Type: application/json" -d '{
  "state": "Order #1182 arrived with a cracked screen. The customer asks for a refund and attached two photos.",
  "questions": {
    "action": {"type": "choice", "instructions": "What should support do next?",
               "criteria": {"refund": "refund the order", "replace": "send a replacement", "escalate": "ask a human"}},
    "urgent": {"type": "noul", "instructions": "Does this need a reply within the hour?"}
  }
}'

Every answer carries the probability of each option, ready for a threshold.

Models

OneJev and Open-Jev, which read their answers their own way, come with Flide. A new family is one folder in plugins/ and one line in pyproject.toml; a separate package can also join through the flide.formats entry point.

Options

License

Apache-2.0. flide/kernels/gdn.py holds a kernel derived from two kernels of flash-linear-attention (MIT); THIRD_PARTY_NOTICES lists it with the other code Flide builds on.

Stars