A PyTorch native library for training speculative decoding models

Fluxon is a distributed data acceleration plane built in Rust for AI-native systems. It features three standardized interfaces: KV/RPC for inference KV / latent Cache, MQ for state passing in training and data pipelines, and FS for caching and S3-compatible access to model files and checkpoints.

MoocHelper is a High-performance MoonCake/Lua plugin, Language Server Protocol for MoonCake/Lua, on the shoulders of LuaHelper.

Instant setup scripts for cloud-based LLM development.

standard C library I/O FFI binding. 对C标准库中stdio的绑定

Mooncake、vLLM、vLLM Ascend 相关的中文教程。

Self-contained PoC: cross-tenant timing oracle against shared content-hashed KV caches (Mooncake, Perplexity KV Messenger, DeepSeek 3FS).