Meituan Tech Team's Selected Papers on Large Language Models and AI (2024-2025)
The article compiles Meituan’s recent 2024‑2025 research on large language models, presenting a diverse set of papers that explore transformer enhancements, scaling laws, safety optimization, instruction fine‑tuning, temporal decay learning, code generation, agent refinement, cost‑efficient MoE inference, quantization, fast parallel inference, speculative decoding, multilingual speech, vision‑language models, evaluation benchmarks, and jailbreak robustness.
This article curates recent academic papers from Meituan's technical team on large language models, covering LLMs, system architecture optimization, multimodal understanding and generation, and evaluation.
Paper list includes: Predictor-Corrector Enhanced Transformers with EMA Coefficient Learning (NeurIPS 2024); Scaling Laws Across Model Architectures: Dense vs MoE Models (arXiv 2024); SEAS: Self-Evolving Adversarial Safety Optimization for LLMs (AAAI 2025); Learning or Self-aligning? Rethinking Instruction Fine-tuning (ACL 2024); Earlier Tokens Contribute More: Learning DPO from Temporal Decay (ICLR 2025); DolphCoder: Echo-Locating Code LLMs with Multi-Objective Instruction Tuning (ACL 2024); AgentRefine: Enhancing Agent Generalization via Refinement Tuning (ICLR 2025); EPS-MoE: Expert Pipeline Scheduler for Cost-Efficient MoE Inference (arXiv 2024); FPTQ: Fine-grained Post-Training Quantization for LLMs (arXiv 2023); A Speed Odyssey for Deployable Quantization of LLMs (arXiv 2023); Flash Communication: Reducing Tensor Parallelization Bottleneck for Fast LLM Inference (arXiv 2024); Speculative Decoding via Early-exiting with Thompson Sampling Control (ACL 2024); Enhancing Multilingual Speech Recognition via Language Prompt Tuning and Frame-level Language Adapter (ICASSP 2024); MobileVLM V2: Faster and Stronger Vision-Language Baseline (arXiv 2024); Denoising with Joint-Embedding Predictive Architecture (ICLR 2025); Lumen: Unleashing Versatile Vision-centric Capabilities of LMMs (NeurIPS 2024); LLaVA-ST: Multimodal LLM for Fine-Grained Spatial-Temporal Understanding (CVPR 2025); Who’s the MVP? Game-theoretic Evaluation Benchmark for Modular Attribution in LLM Agents (arXiv 2025); Leveraging Dual Process Theory in Language Agent Framework for Real-time Simultaneous Human-AI Collaboration (arXiv 2025); Q-Eval-100K: Evaluating Visual Quality and Alignment for Text-to-Vision (CVPR 2025); Hallu-PI: Evaluating Hallucination in Multi-modal LLMs within Perturbed Inputs (ACM MM 2024); A Wolf in Sheep's Clothing: Generalized Nested Jailbreak Prompts can Fool LLMs Easily (NAACL 2024).
Meituan Technology Team
Over 10,000 engineers powering China’s leading lifestyle services e‑commerce platform. Supporting hundreds of millions of consumers, millions of merchants across 2,000+ industries. This is the public channel for the tech teams behind Meituan, Dianping, Meituan Waimai, Meituan Select, and related services.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.