large language models

Dual-stream representation learning with large language models for electricity demand forecasting

Designing Growth Mindset-Inspired Feedback Messages Tailored to Learning Behaviors Using LLM and Persuasive Strategies

This work introduces a four-part feedback message structure inspired by the growth mindset principle, which encourages students to reflect, persist, and embrace challenges, and provides an initial foundation for explainable, rule-based feedback generation that can be extended through experimentation with learners.

An Autonomy Aware Metamodel for Human AI Collaboration in Software Engineering

This work provides a conceptual foundation for governance-aware, adaptable, and AI-first software engineering methods by proposing a metamodel that treats autonomy not as a fixed property of an actor, but as a derived, situation-dependent authority assignment determined by task, context, and collaboration pattern.

CrossServe: Cross-Layer Scheduling for SLO Optimization in Multi-Tenant LLM Serving

The deployment of Large Language Models (LLMs) as multi-tenant cloud services is now widespread, but maintaining high Service Level Objective (SLO) attainment across diverse tenants remains challenging. Current serving systems focus on a single layer of the stack, either using iteration-level batching or coarse-grained...

Harmonic Intelligence: Recursive Analytic Meta-Geometry, Residual-Bearing Learning, and the Geometry of Solution Types — Integrated Edition

[wAI ~ wErrors] «All models are wrong, but some are useful.» — George E. P. Box (1976) This preprint synthesizes a large exploratory notebook on metric atlases, spectral and harmonic representations, residual-bearing learning, recursive joint/meta-diagonalization, Galois and monodromy fingerprints, sheaf/cohomological...

Online Scheduling of Battery-Aware Speculative Decoding for Energy-Efficient Cloud-Edge Collaborative LLM Inference

This work designs a learning-centric algorithmic framework which employs bandit learning to address the intractability and manage draft length selection in the first subproblem and applies online learning to address the online uncertainty and manage battery states and grid-energy use in the second subproblem, both with...

GroupKV: Hierarchical KV Cache Management for Long-Context Diffusion LLM Inference

Diffusion large language models (dLLMs) are emerging as a promising generative paradigm that complements autoregressive decoding. In long-context settings, KV cache bloat and offloading transfer overhead have become primary bottlenecks in inference systems. Meanwhile, the periodic full-sequence recomputation and locali...

BR-MoE: A Memory-Budget-Aware Co-Design Framework for Efficient MoE Inference on Resource-Constrained GPUs

The Mixture-of-Experts (MoE) architecture has become the dominant paradigm for scaling Large Language Models, yet its explosive parameter growth routinely overflows the GPU memory budget. Weight quantization offers a practical way to fit these models into limited GPU memory, but reducing precision below 4 bits causes s...

OmniPipe: Efficient, Flexible and Scalable Pipeline Parallelism for Large Model Training

Mixture-of-Experts (MoE) has become the de facto architecture for scaling large language models, offering expanded capacity with manageable compute. Pipeline parallelism (PP) is indispensable for distributed MoE training, but state-of-the-art PP schemes face three major limitations: large pipeline bubbles, high per-sta...

AsymFlow: Enabling Long-Context LLM Serving via CPU-GPU Prefill-Decode Disaggregation

Driven by cost and privacy constraints, many organizations deploy large language model (LLM) inference on local CPU-GPU servers. For long-context requests, the key-value (KV) cache grows linearly with sequence length and can rapidly exhaust GPU memory, reducing both the maximum supported context and request concurrency...

AccESM: Packed-Total Exact Inference for CPU ESM2 Embedding

Protein language models such as ESM2 are now common tools in biology, and a common task is to compute pooled embeddings for large sequence sets. In many labs, this workload runs on shared CPU servers rather than GPUs, making CPU inference an important systems problem. Existing CPU stacks are general-purpose: they are n...

PIM-Forge: Toward Autonomous PIM Kernel Synthesis through Constraint-Guided Hardware Exploration

PIM-Forge is proposed, a constraint-guided memory-augmented multi-agent framework that iteratively explores hardware constraints and refines synthesis knowledge, validating the efficacy of constraint-guided hardware exploration.