PolyLink: A Blockchain Based Decentralized Edge AI Platform for LLM Inference

TL;DR

PolyLink tackles the centralization of LLM inference by introducing a blockchain-driven decentralized edge AI platform that supports single- and cross-device execution, a TIQE protocol for trustless quality evaluation, and a dynamic token-based incentive model. It combines EdgeShard-based cross-device inference with a hybrid quality assessment (Cross-Encoder + LLM-as-a-Judge) to balance latency and accuracy, and uses VRF-based validator elections with median-score consensus and slashing to maintain integrity. Real-world deployment across geo-distributed edge devices demonstrates practical latency, scalable TIQE performance, and reward distributions that align with contribution quality and stake. The work advances decentralized AI deployment in DePIN contexts and offers a scalable, incentive-aligned framework for edge-based LLM services, while highlighting constraints such as the <1/3> security assumption and network-induced latency in cross-device setups.

Abstract

The rapid advancement of large language models (LLMs) in recent years has revolutionized the AI landscape. However, the deployment model and usage of LLM services remain highly centralized, creating significant trust issues and costs for end users and developers. To address these issues, we propose PolyLink, a blockchain-based decentralized AI platform that decentralizes LLM development and inference. Specifically, PolyLink introduces a decentralized crowdsourcing architecture that supports single-device and cross-device model deployment and inference across heterogeneous devices at the edge. Moreover, to ensure the inference integrity, we design the TIQE protocol, which combines a lightweight cross-encoder model and an LLM-as-a-Judge for a high-accuracy inference evaluation. Lastly, we integrate a comprehensive token-based incentive model with dynamic pricing and reward mechanisms for all participants. We have deployed PolyLink and conducted an extensive real-world evaluation through geo-distributed deployment across heterogeneous devices. Results indicate that the inference and verification latency is practical. Our security analysis demonstrates that the system is resistant to model degradation attacks and validator corruptions. PolyLink is now available at https://github.com/IMCL-PolyLink/PolyLink.

Figures (7)

• Figure 1: Overview of PolyLink, a blockchain-based decentralized edge AI platform. Service users send inference requests via the API. The API server batches requests targeting the same model, which are then processed by a worker running that model. The responses are returned and evaluated for inference quality by the validators.
• Figure 2: Overview of the TIQE Protocol in each Epoch
• Figure 3: The prompt of LLM-as-a-Judge to evaluate the decentralized LLM inference quality.
• Figure 4: Average per-batch evaluation latency of the Cross-Encoder under different batch sizes.
• Figure 5: Per-Batch Latency and Cost of LLM-as-a-Judge under different batch sizes.

Theorems & Definitions (4)

• Theorem 1: Model Degradation Attack Resistance
• Proof 1
• Theorem 2: Validator Corruption Attacks Resistance
• Proof 2