Flide 0.1.10 released, accelerating Jev-style decision models up to 9.5x faster

Flide Accelerates Decision Model Inference

Flide is an inference infrastructure designed for Jev-style decision models. It utilizes exact state forking to achieve faster inference speeds, claiming to be up to 9.5 times faster than existing solutions like vLLM and SGLang. This approach ensures that each answer is the direct output of the model for a given question, without external influences.

Recent Updates and Model Additions

The latest release, Flide 0.1.10, introduced several new decision models, including Ainvest-Decision, Darwin-27B-RSI-Decision, and KnowLine-31B-Gen1.5. Additionally, the KnowLine family of models now has the capability to read images. Previous updates in versions 0.1.9 and 0.1.8 added models such as pplx-decider-v1.1-27b, JEV-35B, AutoTrust, Clef, and drex-max-v1, and enabled the /v1/systemone API to handle multiple requests in a single body.

Broader Model Support and Features

Since its initial release (0.1.0) on PyPI, Flide has expanded its support to include numerous decision model families. These include SynACK Decide, Wald, Xor, Gemma 4 mixture-of-experts models, Drex, and KnowLine. The platform also provides a `flide verify` command to check exact forking on custom models and requests. The API supports various models, including those from Perplexity AI, and allows for questions with up to 256 options.