projects

Research projects and open-source work.

research
project thumbnail

DRDA

Discriminative Radial Domain Adaptation
project thumbnail

MuLTI

Latent Processes Identification From Multi-View Time Series
project thumbnail

iDAG

Invariant DAG Searching for Domain Generalization

work
project thumbnail

Rubicon-preview (30B-A3B)

Rubicon-preview — a 30B-A3B model trained with a novel reinforcement learning framework built on rubric anchors, targeting open-ended, creative and humanities-centric tasks where verifiable rewards are hard to define.
project thumbnail

LLaDA-MoE-7B-A1B

The first open-source sparse MoE diffusion language model — LLaDA-MoE-7B-A1B has 7B total parameters with only ~1.4B activated per token, pre-trained from scratch on ~20T tokens (Base / Instruct / Instruct-TD).
project thumbnail

LLaDA2.0-flash (100B-A6B)

LLaDA2.0-flash — a 100B-A6B MoE diffusion language model, the instruction-tuned flagship of the LLaDA2.0 series (alongside LLaDA2.0-Uni and LLaDA2.0-mini), averaging 79.32 across knowledge, reasoning, coding, math, agent and alignment benchmarks.
project thumbnail

LLaDA2.2-flash (128K)

LLaDA2.2-flash — an agent-oriented diffusion LM with a 128K context window, introducing Levenshtein editing (DELETE / INSERT control tokens) for long-context tool use, multi-turn interaction and robust error correction.
project thumbnail

GroveMoE-Base / -Inst (33B-3.28B)

GroveMoE — a sparse MoE architecture with adjugate experts for dynamic computation allocation, with 33B total parameters and only 3.14–3.28B active per token, up-cycled from Qwen3-30B-A3B-Base via mid-training and SFT.

fun
project thumbnail

青甘大环线动态路书

一份面向桌面与 iPad 的 7 天 6 夜自驾路书，支持分日路线、动态地图、卫星底图、高德定位与打印。