arXiv subjects

Zetian Li

Publications and source records attributed to Zetian Li.

3 recordsLinked to original sources

arXiv2026

Local or Escalate? Post-Training Expert Routing for Real-Time Duplex Speech Models

A full-duplex model should answer easy questions immediately, but hard questions often need stronger reasoning, which may require waiting. The challenge is to choose between these paths at the moment the model begins to speak. Recent systems learn related mode decisions through special-token generation inside the dialogue model, coupling routing to turn-taking and temporally annotated full-duplex training. We study a post-training alternative. We freeze MiniCPM-o 4.5, using only its speech input and output, and fit a logistic failure readout on hidden states already available at its native listen-to-speak transition. The resulting gate requires no additional model generation and exposes thresholds that can be retuned without changing the duplex policy. A probe fitted on text-input states also reads the states of spoken questions, including unseen human recordings. On the internal benchmark, selective handoff improves answer accuracy by 27.1 points. Compared with always using the expert, it reduces mean server time to first answer audio by 31.5 percent. It also outperforms matched random routing across external speech question-answering pools, as well as a zero-shot GPT-5.4 mini router that rates the transcribed question. The training recipe also transfers to a second duplex model family. These results show that an existing duplex model can be equipped to answer easy turns immediately and wait for stronger reasoning or fresher knowledge on the turns that need it, without further duplex training.

cs.AI↗

arXiv2025

LFM2 Technical Report

We present LFM2, a family of Liquid Foundation Models designed for efficient on-device deployment and strong task capabilities. Using hardware-in-the-loop architecture search under edge latency and memory constraints, we obtain a compact hybrid backbone that combines gated short convolutions with a small number of grouped query attention blocks, delivering up to 2x faster prefill and decode on CPUs compared to similarly sized models. The LFM2 family covers 350M-8.3B parameters, including dense models (350M, 700M, 1.2B, 2.6B) and a mixture-of-experts variant (8.3B total, 1.5B active), all with 32K context length. LFM2's training pipeline includes a tempered, decoupled Top-K knowledge distillation objective that avoids support mismatch; curriculum learning with difficulty-ordered data; and a three-stage post-training recipe of supervised fine-tuning, length-normalized preference optimization, and model merging. Pre-trained on 10-12T tokens, LFM2 models achieve strong results across diverse benchmarks; for example, LFM2-2.6B reaches 79.56% on IFEval and 82.41% on GSM8K. We further build multimodal and retrieval variants: LFM2-VL for vision-language tasks, LFM2-Audio for speech, and LFM2-ColBERT for retrieval. LFM2-VL supports tunable accuracy-latency tradeoffs via token-efficient visual processing, while LFM2-Audio separates audio input and output pathways to enable real-time speech-to-speech interaction competitive with models 3x larger. LFM2-ColBERT provides a low-latency encoder for queries and documents, enabling high-performance retrieval across multiple languages. All models are released with open weights and deployment packages for ExecuTorch, llama.cpp, and vLLM, making LFM2 a practical base for edge applications that need fast, memory-efficient inference and strong task capabilities.

cs.LG↗

arXiv2025

Interplay of $95$ GeV Diphoton Excess and Dark Matter in Supersymmetric Triplet Model

The decay of the Higgs boson and the nature of dark matter remain fundamental challenges in particle physics. We investigate the $95$ GeV diphoton excess and dark matter within the framework of the triplet-extended Minimal Supersymmetric Standard Model (TMSSM). In this model, an additional Hypercharge $Y=0$, $SU(2)_L$ triplet superfield is introduced. Mixing between the triplet and doublet Higgs states enhances the diphoton signal strength of the $95$ GeV Higgs boson, resulting in $μ_{γγ}^{\text{CMS+ATLAS}} = 0.24_{-0.08}^{+0.09}$, which is consistent with experimental observations. This enhancement arises primarily from charged Higgs loop contributions. Additionally, the model accommodates viable dark matter candidates in the form of a bino-dominated neutralino. The relic density is reduced to the observed value through resonance-enhanced annihilation via the Higgs portal or co-annihilation with the triplino or higgsino. This reduction remains consistent with constraints from direct and indirect detection experiments. A comprehensive parameter scan demonstrates that the TMSSM can simultaneously explain the $95$ GeV diphoton excess, the observed $125$ GeV Higgs mass, and the dark matter relic density, establishing a compelling and theoretically consistent framework.

hep-ph↗