arXiv subjects

Mingliang Tao

Publications and source records attributed to Mingliang Tao.

3 recordsLinked to original sources

arXiv2026

The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence

We introduce the MiniMax-M2 series, a family of Mixture-of-Experts language models built around the principle that mini activations can unleash maximum real-world intelligence. The flagship M2 contains 229.9B total parameters with only 9.8B activated per token. Designed end-to-end for agentic deployment, the M2 series rests on three components: (i) agent-driven data pipelines producing large-scale, verifiable trajectories across agentic coding and agentic cowork, each grounded in an executable workspace and an artifact-aligned reward; (ii) Forge, a scalable agent-native RL system that adapts to long-horizon agent trajectories, paired with windowed-FIFO scheduling, prefix-tree merging, inference optimization, and a clean training-inference-agent decoupling that supports both white-box and black-box agents; (iii) the latest M2.7 checkpoint takes an early step toward self-evolution -- autonomously debugging training runs and modifying its own scaffold. Across M2 through M2.7, this combination translates a mini-activation footprint into frontier-tier performance on agentic coding, deep search, office-task, and reasoning benchmarks.

cs.AI↗

arXiv2025

MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention

We introduce MiniMax-M1, the world's first open-weight, large-scale hybrid-attention reasoning model. MiniMax-M1 is powered by a hybrid Mixture-of-Experts (MoE) architecture combined with a lightning attention mechanism. The model is developed based on our previous MiniMax-Text-01 model, which contains a total of 456 billion parameters with 45.9 billion parameters activated per token. The M1 model natively supports a context length of 1 million tokens, 8x the context size of DeepSeek R1. Furthermore, the lightning attention mechanism in MiniMax-M1 enables efficient scaling of test-time compute. These properties make M1 particularly suitable for complex tasks that require processing long inputs and thinking extensively. MiniMax-M1 is trained using large-scale reinforcement learning (RL) on diverse problems including sandbox-based, real-world software engineering environments. In addition to M1's inherent efficiency advantage for RL training, we propose CISPO, a novel RL algorithm to further enhance RL efficiency. CISPO clips importance sampling weights rather than token updates, outperforming other competitive RL variants. Combining hybrid-attention and CISPO enables MiniMax-M1's full RL training on 512 H800 GPUs to complete in only three weeks, with a rental cost of just $534,700. We release two versions of MiniMax-M1 models with 40K and 80K thinking budgets respectively, where the 40K model represents an intermediate phase of the 80K training. Experiments on standard benchmarks show that our models are comparable or superior to strong open-weight models such as the original DeepSeek-R1 and Qwen3-235B, with particular strengths in complex software engineering, tool utilization, and long-context tasks. We publicly release MiniMax-M1 at https://github.com/MiniMax-AI/MiniMax-M1.

cs.CL↗

arXiv2020

On the Mutual Interference between Spaceborne SARs: Modeling, Characterization, and Mitigation

Since by the International Telecommunications Union (ITU) regulatory the radio spectrum available to spaceborne synthetic aperture radar (SAR) is restricted to certain limited frequency intervals, there are many different spaceborne SAR systems sharing common frequency bands. Due to this reason, it is reported that two spaceborne SARs at orbit cross positions can potentially cause severe mutual interference. Specifically, the transmitting signal of a SAR, typically linear frequency modulated (LFM), can be directly received by the side or back lobes of another SAR's antenna, causing radiometric artefacts in the focused image. This paper tries to model and characterize the artefacts, and study efficient methods for mitigating them. To this end, we formulate an analytical model for describing the artefact, which reveals that the mutual interference can introduce a two-dimensional LFM radiometric artefact in image domain with a limited spatial extent. We show that the artefact is low-rank based on range-azimuth decoupling analysis and two-dimensional high-order Taylor expansion. Based on the low rank model, we show that two methods, i.e., principle component analysis and its robust variant, can be adopted to efficiently mitigate the artefact via processing in image domain. The former method has the advantage of fast processing speed, for example, a sub-swath of Sentinel-1 interferometric wide swath image can be processed within 70 seconds via block-wise operation, whereas the latter provides improved accuracy for sparse point-like scatterers. Experiment results demonstrate that the radiometric artefacts caused by mutual interference in Sentinel-1 level-1 images can be efficiently mitigated via the proposed methods.

eess.SP↗