Flide: Accelerating Decision Inference with Exact State Forking
Paper · https://arxiv.org/abs/2610.42148
Decision models such as Jev have emerged as a highly promising paradigm for LLM agent systems, providing lightweight and reliable categorical reasoning for rapid System One execution. However, existing generation-oriented inference engines fail to support these workloads effectively, suffering from severe prefix recomputation overheads or distorted output probabilities. To address this challenge, we propose Flide, an efficient and exact inference engine that decouples the logical tree structure of a decision request from its physical execution within a single flattened token stream. Through ForkAttention, Flide navigates request dependencies using GPU-resident indices across both standard attention and recurrent Gated DeltaNet layers, guaranteeing in exact arithmetic that every query branch evaluates identically to an isolated forward pass. By expressing branch topologies directly in GPU-resident metadata, Flide accommodates diverse request mixtures with a compact set of pre-recorded execution graphs while enabling expanding agent histories to resume from checkpoints saved in earlier passes. Extensive evaluations on public hybrid and decision models demonstrate that Flide answers queries up to 9.5 times faster than vLLM and SGLang while staying closest to a float32 reference in total variation distance. By unifying physical execution throughput with mathematical exactness, Flide establishes a principled foundation for scalable, latency-critical decision intelligence in agent systems.
First seen by Oriel on .
AI summary
Flide is an exact inference engine for decision models that uses flattened token streams and GPU-resident branch metadata to accelerate agent queries.
- Its ForkAttention mechanism handles dependencies across standard attention and recurrent Gated DeltaNet layers, matching isolated branch execution in exact arithmetic.
- Flide uses compact pre-recorded execution graphs and can resume expanding agent histories from checkpoints saved in earlier passes.
- Evaluations on public hybrid and decision models report queries up to 9.5 times faster than vLLM and SGLang.
- The results stay closest to a float32 reference in total variation distance, addressing output-probability distortion in generation-oriented engines.
#llm agents #gpu inference #attention #decision models
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hugging Face Papers: 37, latest observation ; +8 over observed period since
Comments
Comment count reported by the source.
- Hugging Face Papers: 6, latest observation ; +5 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Papers | 2610.42148 | https://arxiv.org/abs/2610.42148 Read discussion |