Application for Filecoin ProPGF Batch 2

OpenModel

This project aims to build a decentralized AI model distribution and transaction infrastructure based on Filecoin Onchain Cloud (FOC), addressing fundamental challenges in real-world AI model usage such as high-frequency downloads, version synchronization, creator-driven distribution, and native payment settlement. The system uses Filecoin Warm Storage as the core storage layer for model weights, leverages Beam to provide low-latency and high-bandwidth retrieval, and integrates on-chain registries with Filecoin Pay to form a complete pipeline from verifiable storage to metered and payable distribution. The project will be developed in clearly defined stages. In the current phase, the primary focus is on optimizing AI model distribution and download performance, rather than dataset browsing or model training and inference. The work will begin by validating the stability and usability of Warm Storage for GB-scale AI model downloads, followed by the construction of automated data pipelines to ensure timely synchronization and retrieval of newly released models. Building on this foundation, the project will gradually open creator-driven upload and management capabilities to form a two-sided model distribution ecosystem. In later stages, pricing and paid download mechanisms will be introduced, enabling transparent on-chain settlement and revenue distribution through Filecoin Pay. The project does not aim to perform model training or inference. Instead, it positions itself as a foundational infrastructure layer within the AI ecosystem, validating and extending Filecoin’s practical applicability in AI model distribution and value exchange scenarios.

https://github.com/6block

Zheng Yi

shannon-6block

6block.com

1.What is the problem? The distribution of AI models—such as LLMs, diffusion models, and embedding models—remains heavily dependent on centralized platforms. While Filecoin has mature capabilities at the data storage layer, it still lacks readily usable, end-to-end infrastructure examples for high-frequency retrieval, model version synchronization, creator-driven distribution, and native paid downloads in real-world AI workflows. 2.Why is this important? AI model files are large in size and updated frequently, creating strong requirements for low-latency access, predictable synchronization, and stable bandwidth. At the same time, an increasing number of models are exploring paid downloads or controlled distribution. Without integration with verifiable storage and on-chain settlement, these use cases remain locked into centralized platforms, limiting Filecoin’s adoption in practical AI workloads. 3.Why hasn’t this been well addressed yet? Most existing Filecoin applications are still focused on cold storage and archival use cases. Although FOC components such as Warm Storage, Beam, and Filecoin Pay are already technically available, they have not yet been systematically integrated and validated within a real AI model distribution pipeline. As a result, there is still no reference implementation that fully connects the flow from model storage → retrieval → synchronization → creator uploads → paid settlement in a cohesive, production-oriented manner.

1.AI developers and researchers: More stable and predictable access to large-scale AI models. 2.Model creators and open-source maintainers: Transparent, trust-minimized distribution and payment mechanisms without reliance on centralized platforms. 3.Storage Providers (SPs): New high-read-frequency use cases for Warm Storage, expanding beyond archival workloads. 4.Filecoin protocol and FOC teams: Real-world validation of Warm Storage, Beam, and Filecoin Pay under practical AI workloads. 5.Ecosystem tooling and SDK teams: Reusable patterns and reference implementations for decentralized distribution and payment flows.

1.Read-only AI model library based on Filecoin Warm Storage (Web prototype) Supports stable downloads of major open-source models, with file integrity verifiable via on-chain CIDs. 2.Model retrieval and distribution integration examples Download workflows integrating Filecoin Beam and IPFS sharding, used to validate decentralized distribution performance for GB-scale model files. 3.Automated model synchronization pipeline (Mirror Bot) A backend service that automatically fetches, packages, and synchronizes models from external sources into Filecoin Warm Storage. 4.On-chain model registry contract (Registry Contract, v1) Records model metadata such as CIDs, versions, and timestamps, and supports real-time queries from the frontend. 5.Creator upload and management prototype Includes wallet-based authentication, model uploads, version management, and basic metadata/README editing. 6.Paid download workflow prototype based on Filecoin Pay Demonstrates end-to-end model pricing, payment-triggered retrieval, and revenue settlement. 7.Public technical documentation and demo materials Architecture descriptions, usage guides, demo videos, and test records to enable third-party auditing and reproducibility.

$50,000 USD

Hack FEVM: FNS: Filecoin Naming Service

150,000

This project incrementally builds an AI model distribution infrastructure that embeds Filecoin’s storage and retrieval capabilities directly into real-world AI development workflows. By remaining compatible with existing Hugging Face client tools and preserving established usage patterns, the project redirects model distribution traffic to the Filecoin network without introducing additional learning overhead for developers. In the initial phase, mainstream open-source models are deployed to Warm Storage and optimized through the combined use of Beam and IPFS sharding mechanisms. This is expected to deliver approximately 2–3× improvements in effective distribution efficiency, significantly improving download stability and predictability for large model files. As a result, developers face lower time-to-access and fewer failed downloads, increasing practical usage frequency. Building on this foundation, the project introduces an automated model synchronization pipeline that enables newly released models to become retrievable shortly after publication, reducing the “distribution lag window” commonly observed in decentralized environments. This shifts Filecoin from a primarily static storage role into a continuous distribution layer across the AI model lifecycle, enabling higher-frequency access and updates. With the introduction of creator-managed uploads and model administration, model supply expands from a fixed, preloaded set into a multi-participant ecosystem. Distribution evolves from a one-way publishing model into a multi-source circulation network, where creators and users no longer rely on centralized hosting platforms but instead leverage verifiable storage and retrieval paths—lowering entry barriers on both sides. Finally, by enabling pricing and paid downloads, model distribution becomes directly coupled with value exchange. Improvements in performance and reliability are expected to translate into higher download conversion rates, with early-stage projections indicating a 1.5–2× increase in effective paid download conversions. This positive feedback loop encourages sustained model publishing and updates, while providing a reusable infrastructure pattern for future ecosystem participants.

Within 30 days after M1 completion: At least 5–10 mainstream models (each ≥10 GB) are downloadable via the Web prototype. Under identical network conditions, download tests using IPFS sharded parallel retrieval demonstrate a ≥2× improvement in effective throughput compared to single-stream retrieval, supported by test scripts and logs/reports. Upon M2 completion: At least 20 models are automatically synchronized by the Mirror Bot and registered on-chain. The median synchronization time from upstream release to retrievable availability is ≤6 hours, with the 95th percentile ≤24 hours, verified via on-chain timestamps and synchronization logs. Within 30 days after M3 completion: At least 20 creators successfully complete self-service uploads, publishing a cumulative total of ≥50 model versions. Each model version is auditable via its on-chain CID and registry record, with corresponding on-chain references and frontend page links provided.

After the grant period concludes, the project will continue operating and scaling through multiple complementary paths. First, as the number of model creators and users grows, the platform will charge a minimal service fee on paid model downloads, without compromising openness. This fee is intended to cover essential operational, storage, and bandwidth costs, enabling a self-sustaining operating model that scales naturally with platform usage. Second, the project will pursue long-term collaborations with Filecoin ecosystem storage providers (SPs) and FOC-related teams to establish stable supply–demand alignment across Warm Storage, retrieval acceleration, and data distribution. These partnerships aim to reduce per-unit storage and distribution costs while supporting the continuous onboarding of larger volumes of models and datasets. Finally, based on the validated technical capabilities and usage data generated during this grant, the project will continue to seek follow-on funding within the Filecoin ecosystem (e.g., future ProPGF rounds or FOC-specific programs) and engage in joint development or integrations with AI infrastructure providers, model publishers, and ecosystem tooling teams. These efforts will further extend the applicability of this distribution pattern across a broader range of AI workloads.

6block Team Core contributors to multiple blockchain infrastructure projects, including Aleo.info, Filfox Explorer, and ZK.Work mining pool, with extensive experience in protocol-level tooling, large-scale infrastructure operations, and ecosystem-facing developer platforms.

1. Protocol and component evolution risk The project depends on the continued evolution of Filecoin Onchain Cloud components (including Warm Storage, Beam, and Filecoin Pay). Changes to interfaces or behavior may require integration adjustments. This risk will be mitigated through close coordination with the FOC team, modular system design, and incremental integration strategies. 2. External platform and ecosystem dependency The automated model synchronization phase relies on public interfaces and access policies of external model publishing platforms (e.g., Hugging Face). Changes to these interfaces may affect synchronization efficiency. The project will reduce dependency on any single platform through configurable data sources and multi-source support. 3. Technical uncertainty in large-file distribution High-concurrency distribution of GB-scale model files in a decentralized network is sensitive to network conditions and node availability. The project will address this through performance optimization using Warm Storage, Beam, and IPFS sharded parallel retrieval, with an early focus on validating stability boundaries. 4. Early-stage adoption uncertainty As an infrastructure-focused project, initial usage may be influenced by overall ecosystem maturity. This risk will be mitigated by prioritizing mainstream models, minimizing onboarding friction, and collaborating with ecosystem partners to onboard seed users.

https://github.com/6block

1. Filecoin Onchain Cloud team Can verify that the project actively uses core components such as Warm Storage and Beam, and evaluate their real-world performance in AI model distribution scenarios. 2. AI model users and creators (early adopters) Developers and creators who download or upload models through the platform can validate usability and distribution effectiveness within real development workflows. 3. Core contributors to Warm Storage / Beam Can technically assess the project’s integration approach and performance across storage, retrieval, and distribution paths. 4. Filecoin storage providers (SPs) involved in Warm Storage and retrieval As the project directly depends on their nodes for model storage and retrieval, participating SPs can serve as infrastructure-level validators. 5. ProPGF review team (on-chain verification) Can independently verify milestone completion and relevance through on-chain CIDs, storage states, and transaction records.