Tag
This article discusses the integration of value classes as a preview feature in JDK 28 and analyzes the optimization opportunities and limitations in the JVM, emphasizing the need for careful compiler support.
This article provides a step-by-step guide on running the Qwen3.8-27B AI model on a 16GB VRAM laptop using tools like exllamav3 and tabbyAPI, including installation, configuration, and performance benchmarks.
This paper introduces Response Renormalization, a backward-pass framework to improve training stability in Deep Equilibrium Models by addressing near-singular Jacobian issues, demonstrated across various multiphysics applications.
The paper introduces the 'Compression Trinity' framework, jointly applying sparsity, quantization, and low-rank approximations to compress Large Language Models for improved efficiency and performance.
This paper introduces ExTS, a tree-search policy for budget-constrained agentic search in LLM agents that improves over standard baselines in tasks like prompt optimization and code generation with an average +5.5% gain.
vLLM introduces a native sharded weight transfer engine using Ray Direct Transport (RDT) for efficient weight syncing in large-scale online RL setups, enhancing performance for models like Kimi K2.
The paper introduces the sync heap, a novel data structure that decouples deletion from revealing the deleted element, enabling constant amortized time for insertions and deletions under limited inspections, and improving a textbook scheduling problem from O(n log n) to O(n).
The author recounts a costly mistake from underutilizing GPUs and highlights how AI monitoring agents from the Viktor team can prevent such inefficiencies in real-time.
This blog post describes how replacing a Rust enum with a 64-bit word in the Plush interpreter led to a 17% speed improvement by reducing memory usage and enhancing cache efficiency.
This tutorial demonstrates how to create minimal ELF executables on Linux using assembly code and optimization techniques to reduce file size.
JetBrains officially launches Junie Local, a local AI Coding Agent that optimizes local models like Qwen3.6-27B for offline operation, tuned for speed and stability.
The article proposes speculative programmatic tool calling (sPTC), a technique to optimize tool calls in AI harnesses by overlapping execution with token generation, reducing latency inspired by speculative execution in CPUs and LLMs.
Introducing Speculative Programmatic Tool Calling (sPTC), a technique to speculate on tool calls during code generation to overlap with token generation and execution time, improving efficiency in AI harnesses.
The paper introduces RODE, an optimizer that decouples radial and directional components for matrix updates, demonstrating superior performance over Muon variants in language modeling and image classification tasks.
This paper provides a theoretical analysis of the edge-of-stability phenomenon for the Adam optimizer on a one-dimensional quadratic function, proving that Adam exhibits a restoring mechanism that pushes sharpness toward a stability threshold.
GRAFT introduces a draft-tree construction framework for diffusion language model-based speculative decoding, optimizing edge selection and budget allocation to achieve 2.13×–6.36× speedup over autoregressive decoding with low overhead.
This paper proposes DASO, a tree-aware post-training method for generative recommendation that addresses difficulty mismatch in GRPO by profiling rollout groups and reallocating based on prefix-match depth, improving performance on public benchmarks.
The article discusses the impact of Stephen Boyd's 2006 mathematics textbook on convex optimization, used by Google, Renaissance Technologies, and others, and promotes his free Stanford course EE364A available on YouTube.
The article explains the key differences between DevOps, MLOps, and LLMOps, highlighting how each addresses distinct challenges in software development, machine learning, and LLM applications, with a focus on unique monitoring and optimization in LLMOps.
A fork of llama.cpp has been created and optimized for AMD GFX906 GPUs, improving performance on Mi50, Mi60, Radeon VII, and GCN HIP for machine learning and LLM applications.