Show HN: Flide: Make your decision models fast Flide released version 0.1.9 of its inference infrastructure for Jev-style decision models, which the project says runs up to 9.4 times faster than vLLM and SGLang through exact state forking while returning the same answer the model gives the question alone. Flide 0.1.9 adds the pplx-decider-v1.1-27b and JEV-35B models and lets the /v1/systemone API accept several requests in one body; the project published its paper "Flide: Accelerating Decision Inference with Exact State Forking" on arXiv on 2026-11-02. Earlier releases from 2026-10-25 through 2026-11-08 added model families including AutoTrust, Clef, drex-max-v1, SynACK Decide, Wald, Xor, KnowLine, Drex, Jeffan, JevAny, Kev, Lychee, pplx-decider and StartLux, and version 0.1.6 introduced the `flide verify` command to check exact forking on a user's own model and requests. News -news · Speed speed · Exact answers exact-answers · Quick start quick-start · Models models · Options options Flide https://docs.flide.org is fast inference infrastructure for Jev-style decision models. With exact state forking it is up to 9.4 times faster than vLLM and SGLang, and every answer is the one the model gives the question alone. It serves the Jev /v1/systemone API. - 🔥 2026-11-09 We added pplx-decider-v1.1-27b https://huggingface.co/perplexity-ai/pplx-decider-v1.1-27b and JEV-35B https://huggingface.co/autotrust/JEV-35B in Flide 0.1.9, and /v1/systemone now takes several requests in one body. - 🔥 2026-11-08 We added two decision model families in Flide 0.1.8, including AutoTrust https://huggingface.co/collections/autotrust/autotrust-decision-models-6ada7050d1462de6733b7f98 and Clef https://huggingface.co/collections/Cloudflare/clef-6ac429cac15d70ee2af5d1c6 , and also drex-max-v1 https://huggingface.co/nace-ai/drex-max-v1 . - 🔥 2026-11-02 We released our paper Flide: Accelerating Decision Inference with Exact State Forking https://arxiv.org/abs/2610.42148 on arXiv. - 🔥 2026-10-29 We added three decision model families in Flide 0.1.7, including SynACK Decide https://huggingface.co/SkyPanther/synack-decide-26b-a4b , Wald https://huggingface.co/collections/org2ai/wald and Xor https://huggingface.co/juspay/Xor-26B-A4B , and Gemma 4 mixture-of-experts models such as KnowLine-26B-A4B https://huggingface.co/PelaAI/KnowLine-26B-A4B-Gen2.5 . - 🔥 2026-10-29 We added two decision model families in Flide 0.1.6, including Drex https://huggingface.co/nace-ai/drex-v2.0 and KnowLine https://huggingface.co/PelaAI/KnowLine-35B-A3B-Gen1.5 , and flide verify checks exact forking on your own model and requests. - 🔥 2026-10-28 We added six decision model families in Flide 0.1.5, including Jeffan https://huggingface.co/collections/tattrongvu/decision-model-6ad5c85a79bd246cc1c18c4a , JevAny https://huggingface.co/collections/SimpleJev/jevany-adaptive-decision-systems-6abc8fd39266b1c17b11b4e6 , Kev https://huggingface.co/collections/jaredpalmer/kev-6aad9d0ea49f2589665e07cd , Lychee https://huggingface.co/LycheeAI/Lychee-27B , pplx-decider https://huggingface.co/perplexity-ai/pplx-decider-flash-v1.2 and StartLux https://huggingface.co/collections/startlux-models/startlux-decision-v11-instant-6aca8158623e8e001724cc82 , and images for Lumma https://huggingface.co/collections/FrontiersMind/lumma-decision-models-6ab6356434b3db6856ec7ae6 and mammoth https://huggingface.co/baselabs/mammoth-1 . - 🔥 2026-10-26 We added five decision model families in Flide 0.1.4, including cerno https://huggingface.co/apertixai/cerno-27b , JADE https://huggingface.co/theunnecessarythings/JADE , Linne https://huggingface.co/collections/PNPM244/decision-model-6ad363fb3351cecf1112a5a8 , mammoth https://huggingface.co/baselabs/mammoth-1 and Sieve https://huggingface.co/collections/sthanika-ai/sieve-6abb93a61dd495f69811c586 , and also Quyet-1.0-Large https://huggingface.co/chinhnc/Quyet-1.0-Large . - 🔥 2026-10-26 We added eight decision model families in Flide 0.1.3, including AJev https://huggingface.co/andyzhang232/ajev-gemma4-12b-lora7 , Bespoke-Nimble https://huggingface.co/bespokelabs/Bespoke-Nimble-9B-v3 , Darwin https://huggingface.co/collections/FINAL-Bench/ztc-models-jev-ecosystems-6aafa630afd69b39ead55aa8 , jiwo https://huggingface.co/eljiwo/jiwo-4b , Lumma https://huggingface.co/collections/FrontiersMind/lumma-decision-models-6ab6356434b3db6856ec7ae6 , mogan https://huggingface.co/moganai/mogan-decision-31B-it , Quyet https://huggingface.co/collections/chinhnc/quyet-10-6ac1b597a9a6a6806bb01fa6 and Senova https://huggingface.co/WangSenQWQ/Senova-4B , and LoRA adapters now serve as they are. - 🔥 2026-10-25 We added two decision model families in Flide 0.1.2, including Intern-Decision https://huggingface.co/collections/internlm/intern-decision-6ab75776af19ca84df0bc004 and torchcast https://huggingface.co/torchcast-ai/torchcast-decision-27b , and a question now takes up to 256 options. - 🔥 2026-10-24 We released Flide 0.1.0 on PyPI https://pypi.org/project/flide/ , serving Qwen3.5 https://huggingface.co/collections/Qwen/qwen35-6992e3053c019221cf2d725f and Gemma 4 https://huggingface.co/collections/google/gemma-4-69ce8ad93186d46744cb42f1 chat models and the OneJev https://huggingface.co/collections/OmniJev/onejev16-6ad27669cb868abe0516486d and Open-Jev https://huggingface.co/collections/ZefanCai/open-jev-6ab049b9d43a267bae4dedc8 decision models. Welcome to have a try. With three questions, a generation engine runs three sequences and computes 4 cells of the state again, and multi-item scoring adds 3 delimiters that change the answers. Flide runs one sequence, computes nothing twice, adds no tokens, and gives each question the answer it gets when asked alone. On one H200 with Qwen3.5-0.8B, Flide is up to 9.4 times faster than both vLLM and SGLang QuALITY, where it is 14 times faster than vLLM , 3.6 to 4 times faster on RACE and the Decision Index, and serves 2.4 times the throughput with 64 clients. The more questions a state carries, the bigger the gap. Flide stays within 1.1 points of a float32 reference, the closest of any engine, and flips at most 2.3% of decisions at 0.5. Multi-item scoring moves answers by up to 14.8 points and flips up to 28.9%. With --batch-invariant , a request gets the same bits however it is batched. See the docs https://docs.flide.org for every model, the API and all options. uv venv -p 3.12 && source .venv/bin/activate uv pip install flide torchvision 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 models read their answers their own way, and Flide serves them as they are too. flide serve --model OmniJev/OneJev1.6-0.8B An Open-Jev release is an adapter on a Qwen3.5 base. Merge it into a checkpoint once, then serve that. python -m flide open jev.prepare ZefanCai/Open-Jev-2B open-jev-2b flide serve --model open-jev-2b flide serve listens on all interfaces at port 8000; pass --host 127.0.0.1 to keep it local. 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?"} } }' OneJev models also read images. Put