AI Coding2026
Acts as a proxy that incrementally compresses LLM conversation history for coding agents so a single session can run for months while using far fewer tokens. Compression is model-driven, reversible, and designed to preserve a stable KV-cache prefix so attention and billing remain efficient.
AI Dataset2026
Provides 16 weeks of anonymized production agent-session traces (12,002 sessions, ~1.19M LLM requests, ~1.21M tool calls, 209B input tokens) released as block-level prefix IDs plus flattened Parquet tables for KV-cache, scheduling and serving-system research.
Serves token-level routed LLM inference by dispatching requests to per-model asynchronous subservers and using delayed-batching scheduling to reduce admission latency and step desynchronization. Exposes a request-centric route-send-receive API and reports 2.01–64.15× decoding throughput gains versus single-LLM servers.
AI Image2026
Runs text-to-image generation and instruction-guided image editing in 8 denoising steps. An accelerated checkpoint of Qwen-Image-2.1 that preserves the same 7B visual generator, native RGBA support, Diffusers QwenImage21Pipeline compatibility, a saved 8-step sampling schedule (CFG=1), and prefix KV cache reuse for multi-reference editing.
AI Video Papers2026
Converts pretrained multimodal LLMs into a System One decision interface that outputs structured, software-friendly decisions for video (temporal spans or multiple-choice answers); uses a shared scoring head and cached/joint inference to raise temporal-grounding mIoU and drastically speed up repeated queries.