AINeutralarXiv – CS AI · Apr 146/10
🧠Researchers present the first systematic study of performance-energy trade-offs in multi-request LLM inference workflows, using NVIDIA A100 GPUs and vLLM/Parrot serving systems. The study identifies batch size as the most impactful optimization lever, though effectiveness varies by workload type, and reveals that workflow-aware scheduling can reduce energy consumption under power constraints.
🏢 Nvidia
AIBullisharXiv – CS AI · Apr 146/10
🧠Researchers propose Tool-Internalized Reasoning (TInR), a framework that embeds tool knowledge directly into Large Language Models rather than relying on external tool documentation during reasoning. The TInR-U model uses a three-phase training pipeline combining knowledge alignment, supervised fine-tuning, and reinforcement learning to improve reasoning efficiency and performance across various tasks.
AIBullisharXiv – CS AI · Apr 146/10
🧠Researchers have optimized the Bielik v3 language models (7B and 11B parameters) by replacing universal tokenizers with Polish-specific vocabulary, addressing inefficiencies in morphological representation. This optimization reduces token fertility, lowers inference costs, and expands effective context windows while maintaining multilingual capabilities through advanced training techniques including supervised fine-tuning and reinforcement learning.
AIBullisharXiv – CS AI · Apr 146/10
🧠Researchers introduce MEDS, a memory-enhanced reward shaping framework that addresses a critical reinforcement learning failure mode where language models repeatedly generate similar errors. By tracking historical behavioral patterns and penalizing recurring mistake clusters, the method achieves consistent performance improvements across multiple datasets and models while increasing sampling diversity.
AIBullisharXiv – CS AI · Apr 146/10
🧠Researchers propose NExt, a nonlinear extrapolation framework that accelerates reinforcement learning with verifiable rewards (RLVR) for large language models by modeling low-rank parameter trajectories. The method reduces computational overhead by approximately 37.5% while remaining compatible with various RLVR algorithms, addressing a key bottleneck in scaling LLM training.
AIBullisharXiv – CS AI · Apr 146/10
🧠Researchers propose ITEM, an iterative utility judgment framework that enhances retrieval-augmented generation (RAG) systems by aligning with philosophical principles of relevance. The framework improves how large language models prioritize and process information from retrieval results, demonstrating measurable improvements across multiple benchmarks in ranking, utility assessment, and answer generation.
AIBullisharXiv – CS AI · Apr 146/10
🧠Researchers demonstrate that quantization and local inference techniques can reduce LLM energy consumption and carbon emissions by up to 45% without sacrificing performance. The findings address growing sustainability concerns surrounding generative AI deployment, offering practical optimization strategies for resource-constrained environments.
AIBullisharXiv – CS AI · Apr 146/10
🧠Researchers introduce HiPRAG, a training methodology that improves agentic RAG systems by using fine-grained process rewards to optimize search decisions. The approach reduces inefficient search behaviors while achieving 65-67% accuracy across QA benchmarks, demonstrating that optimizing reasoning processes yields better performance than outcome-only training.
🧠 Llama
AINeutralarXiv – CS AI · Apr 136/10
🧠Researchers propose StaRPO, a reinforcement learning framework that improves large language model reasoning by incorporating stability metrics alongside task rewards. The method uses Autocorrelation Function and Path Efficiency measurements to evaluate logical coherence and goal-directedness, demonstrating improved accuracy and reasoning consistency across four benchmarks.
AIBullisharXiv – CS AI · Apr 136/10
🧠Researchers present PETITE, a tutor-student multi-agent framework that enhances LLM problem-solving by assigning complementary roles to agents from the same model. Evaluated on coding benchmarks, the approach achieves comparable or superior accuracy to existing methods while consuming significantly fewer tokens, demonstrating that structured role-differentiated interactions can improve LLM performance more efficiently than larger models or heterogeneous ensembles.
AINeutralarXiv – CS AI · Apr 136/10
🧠A research paper proposes a fundamental shift in how retrieval systems are evaluated, moving from traditional relevance-based metrics toward utility-centric optimization for large language models. This framework argues that retrieval effectiveness should be measured by its contribution to LLM-generated answer quality rather than document ranking alone, reflecting the structural changes introduced by retrieval-augmented generation (RAG) systems.
AIBullisharXiv – CS AI · Apr 136/10
🧠Researchers introduce RecaLLM, a post-trained language model that addresses the 'lost-in-thought' phenomenon where retrieval performance degrades during extended reasoning chains. The model interleaves explicit in-context retrieval with reasoning steps and achieves strong performance on long-context benchmarks using training data significantly shorter than existing approaches.
AIBullisharXiv – CS AI · Apr 136/10
🧠Researchers introduce Chain-in-Tree (CiT), a framework that optimizes large language model tree search by selectively branching only when necessary rather than at every step. The approach reduces computational overhead by 75-85% on math reasoning tasks with minimal accuracy loss, making inference-time scaling more practical for resource-constrained deployments.
AIBullisharXiv – CS AI · Apr 106/10
🧠Researchers introduce RePro, a novel post-training technique that optimizes large language models' reasoning processes by framing chain-of-thought as gradient descent and using process-level rewards to reduce overthinking. The method demonstrates consistent performance improvements across mathematics, science, and coding benchmarks while mitigating inefficient reasoning behaviors in LLMs.
AINeutralarXiv – CS AI · Apr 76/10
🧠Research reveals that adaptive reward mechanisms in AI-guided satellite scheduling systems actually hurt performance, with static reward weights achieving 342.1 Mbps versus dynamic weights at only 103.3 Mbps. The study found that fine-tuned LLMs performed poorly due to weight oscillation issues, while simpler MLP models achieved superior results of 357.9 Mbps.
AIBullisharXiv – CS AI · Mar 276/10
🧠Researchers have developed EcoThink, an energy-aware AI framework that reduces inference energy consumption by 40.4% on average while maintaining performance. The system uses adaptive routing to skip unnecessary computation for simple queries while preserving deep reasoning for complex tasks, addressing sustainability concerns in large language model deployment.
AIBullisharXiv – CS AI · Mar 266/10
🧠Researchers propose APreQEL, an adaptive mixed precision quantization method for deploying large language models on edge devices. The approach optimizes memory, latency, and accuracy by applying different quantization levels to different layers based on their importance and hardware characteristics.
AINeutralarXiv – CS AI · Mar 55/10
🧠Researchers present a blueprint for evaluating and optimizing multi-agent conversational shopping assistants, addressing challenges in multi-turn interactions and tightly coupled AI systems. The paper introduces evaluation rubrics and two prompt-optimization strategies including a novel Multi-Agent Multi-Turn GEPA approach for system-level optimization.
AIBullisharXiv – CS AI · Mar 37/106
🧠Researchers propose Draft-Thinking, a new approach to improve the efficiency of large language models' reasoning processes by reducing unnecessary computational overhead. The method achieves an 82.6% reduction in reasoning budget with only a 2.6% performance drop on mathematical problems, addressing the costly overthinking problem in current chain-of-thought reasoning.
AIBullisharXiv – CS AI · Mar 37/108
🧠Researchers introduce LittleBit-2, a new framework for extreme compression of large language models that achieves sub-1-bit quantization while maintaining performance comparable to 1-bit baselines. The method uses Internal Latent Rotation and Joint Iterative Quantization to solve geometric alignment issues in binary quantization, establishing new state-of-the-art results on Llama-2 and Llama-3 models.
AIBullisharXiv – CS AI · Mar 36/104
🧠Researchers introduce AdaBlock-dLLM, a training-free optimization technique for diffusion-based large language models that adaptively adjusts block sizes during inference based on semantic structure. The method addresses limitations in conventional fixed-block semi-autoregressive decoding, achieving up to 5.3% accuracy improvements under the same throughput budget.
AIBullisharXiv – CS AI · Mar 36/104
🧠Researchers evaluated HiFloat (HiF8 and HiF4) formats for low-bit inference on Ascend NPUs, finding them superior to integer formats for high-variance data and preventing accuracy collapse in 4-bit regimes. The study demonstrates HiFloat's compatibility with existing quantization frameworks and its potential for efficient large language model inference.
AIBullisharXiv – CS AI · Mar 27/1018
🧠Researchers propose Semantic Parallelism, a new framework called Sem-MoE that significantly improves efficiency of large language model inference by optimizing how AI models distribute computational tasks across multiple devices. The system reduces communication overhead between devices by 'collocating' frequently-used model components with their corresponding data, achieving superior throughput compared to existing solutions.
AIBullishHugging Face Blog · Apr 166/107
🧠The article discusses prefill and decode techniques for optimizing Large Language Model (LLM) performance when handling concurrent requests. These methods aim to improve efficiency and reduce latency in AI systems serving multiple users simultaneously.
AINeutralarXiv – CS AI · Apr 105/10
🧠Researchers introduce MSPA-CQR, a machine learning approach that improves conversational query rewriting by aligning preferences across three dimensions: query rewriting, passage retrieval, and response generation. The method uses self-consistent preference data and direct preference optimization to generate more diverse and effective rewritten queries in conversational search systems.
