The EAGLE team, vLLM, and TorchSpec jointly release EAGLE 3.1 to fix speculative decoding instability in production.

The post Meet EAGLE 3.1: The Speculative Decoding Algorithm That Fixes Attention Drift in LLM Inference appeared first on MarkTechPost.

Researchers from NUS, MIT, and A*STAR propose MEMO, a modular framework that encodes corpus knowledge into a separate trainable MEMORY model.

The post MEMO: A Modular Framework for Training a Dedicated Memory Model on New Knowledge Without Modifying LLM Parameters appeared first on MarkTechPost.

In this tutorial, we use zeroentropy/zerank-2-reranker, a 4B Qwen3-based cross-encoder reranker, to improve retrieval quality. We start by setting up the runtime, loading the reranker, and understanding how it scores query-document pairs. Then, we move from simple pairwise scoring to a practical two-stage retrieve-and-rerank pipeline, where a fast bi-encoder first retrieves candidates and zerank-2 reranks […]

The post Design a High-Precision Retrieve-and-Rerank Pipeline with ZeroEntropy Zerank-2 Reranker appeared first on MarkTechPost.

Stability AI has released Stable Audio 3, a family of latent diffusion models for instrumental music and sound effects generation. The release includes open weights for the small and medium variants. Small runs on a MacBook Pro M4 CPU. Medium fits on consumer GPUs with 8 GB of VRAM. Both generate stereo audio at 44.1 kHz using a three-stage training pipeline: flow matching, distillation warmup, and adversarial post-training. On the BBC Sound Effects benchmark at 5 seconds, SA3 medium scores FAD 0.369 — lower than every open-weight baseline evaluated in the paper.

The post Stability AI Releases Stable Audio 3: A Family of Fast Latent Diffusion Models for Audio Generation and Editing appeared first on MarkTechPost.

OmniVoice Studio runs voice cloning, video dubbing, real-time dictation, and speaker diarization entirely on your own hardware. No API keys, no cloud account, and no subscription required. The project supports 646 languages for TTS and exposes an MCP server for integration with Claude, Cursor, or any MCP client.

The post Meet OmniVoice Studio: A Local, Open-Source Alternative to ElevenLabs appeared first on MarkTechPost.

In this tutorial, we explore the TuringEnterprises/Open-MM-RL dataset as a practical foundation for multimodal reasoning and reinforcement learning with verifiable rewards. We load the dataset, inspect its schema, analyze domains, formats, question lengths, answer types, and image distributions, and visualize representative examples from each domain. We also build a lightweight reward function that checks exact, […]

The post Design a Complete Multimodal RLVR Pipeline with Open-MM-RL, Vision-Language Prompting, Reward Scoring, and GRPO Export appeared first on MarkTechPost.

Together AI has released OSCAR (Offline Spectral Covariance-Aware Rotation), an INT2 KV cache quantization method for long-context LLM serving. Unlike prior rotation-based approaches that apply data-oblivious Hadamard transforms, OSCAR derives separate rotations for keys and values from attention-aware covariance structures estimated offline. At 2.28 bits per KV element, OSCAR reduces the BF16 accuracy gap to 3.78 points on Qwen3-4B-Thinking-2507 and 1.42 points on Qwen3-8B, while delivering approximately 8× KV memory reduction and up to 3× decode speedup at 100K context length.

The post Together AI Open-Sources OSCAR: An Attention-Aware 2-Bit KV Cache Quantization System for Long-Context LLM Serving appeared first on MarkTechPost.

In this tutorial, we build an advanced federated learning experiment with NVIDIA FLARE. We compare FedAvg and FedProx on a non-IID CIFAR-10 setup, where client data is split using a Dirichlet distribution to simulate realistic label imbalance across federated sites. We use the NVFlare Job API to define and launch federated jobs, while the Client […]

The post Step by Step Guide to Build and Compare FedAvg and FedProx Federated Learning on Non-IID CIFAR-10 with NVIDIA FLARE appeared first on MarkTechPost.

As MCP crosses 97 million monthly SDK downloads and AI agents move into production workflows, authentication has become the most critical infrastructure decision teams face. This guide ranks the eight leading platforms — WorkOS, Stytch, Auth0 by Okta, Composio, Nango, Arcade, TrueFoundry, and Cloudflare — on spec compliance, enterprise identity depth, integration breadth, and real-world fit for 2026 deployments.

The post Best Authentication Platforms for AI Agents and MCP Servers in 2026 appeared first on MarkTechPost.

Most web applications still have no structured way for an AI agent to register. auth.md proposes a fix: a Markdown file apps publish at their domain that tells agents which registration flows are supported, which scopes to request, and how to get credentials tied to a real user — without a human filling out a form.

The post WorkOS Releases auth.md: An Open Agent Registration Protocol Built on OAuth Standards appeared first on MarkTechPost.

In this tutorial, we implement the Langfuse (an open-source LLM engineering platform) pipeline for tracing, prompt management, scoring, datasets, and experiments. We build a complete workflow that works with either a real OpenAI key or a deterministic mock LLM, so we can understand every major Langfuse feature without depending on paid model access. We start […]

The post Build a Complete Langfuse Observability and Evaluation Pipeline for Tracing, Prompt Management, Scoring, and Experiments appeared first on MarkTechPost.

StepFun, the Shanghai-based AI lab, released StepAudio 2.5 Realtime in May 2026 — an end-to-end real-time speech large language model with fully customizable persona capabilities. The model connects via a WebSocket API, supports Chinese and English, and ranked first across all five benchmark dimensions tested in April 2026, including an 80.41 human evaluation score and 82.18 on paralinguistic comprehension.

The post StepFun Releases StepAudio 2.5 Realtime: An End-to-End Voice Model with Roleplay-Specific RLHF and Paralinguistic Comprehension appeared first on MarkTechPost.

Microsoft Research introduces Webwright, a terminal-native browser agent framework that replaces click-trace web automation with reusable Playwright scripts. Using a single agent loop across three modules and roughly 1,000 lines of code, Webwright powered by GPT-5.4 reaches 60.1% on the long-horizon Odysseys benchmark and 86.7% on Online-Mind2Web — the highest AutoEval score among open-sourced harness recipes.

The post Microsoft Research Releases Webwright: A Terminal-Native Web Agent Framework That Scores 60.1% on Odysseys, Up from Base GPT-5.4’s 33.5% appeared first on MarkTechPost.

Linear attention squeezes the unbounded KV cache into a fixed-size recurrent state, but editing that memory without scrambling existing associations is hard. Prior delta-rule models like Gated DeltaNet and KDA use one scalar gate to control both erasing old content and writing new content. NVIDIA's Gated DeltaNet-2 decouples these into a channel-wise erase gate b_t on the key axis and a channel-wise write gate w_t on the value axis. At 1.3B parameters trained on 100B FineWeb-Edu tokens, it outperforms Mamba-2, Gated DeltaNet, KDA, and Mamba-3 across language modeling, commonsense reasoning, and long-context retrieval — with the largest gains on RULER S-NIAH and multi-key needle retrieval.

The post NVIDIA AI Releases Gated DeltaNet-2: A Linear Attention Layer That Decouples Erase and Write in the Delta Rule appeared first on MarkTechPost.

Tencent has open-sourced TencentDB Agent Memory, a fully local memory system for AI agents released under the MIT license. The project pairs symbolic short-term memory, which offloads verbose tool logs into a compact Mermaid task canvas, with a 4-tier long-term memory pyramid (L0 Conversation → L1 Atom → L2 Scenario → L3 Persona). It ships as an OpenClaw plugin and a Hermes Docker image, runs on local SQLite + sqlite-vec by default, and uses hybrid BM25 + vector retrieval with RRF fusion. Tencent's own benchmarks report a 61.38% token reduction and 51.52% relative pass-rate gain on WideSearch with OpenClaw, alongside PersonaMem accuracy moving from 48% to 76%.

The post Tencent Open-Sources TencentDB Agent Memory: A 4-Tier Local Memory Pipeline for AI Agents appeared first on MarkTechPost.

In this tutorial, we build an advanced workflow using the SuperClaude Framework as a structured layer on top of the Anthropic API.

The post Build a SuperClaude Framework Workflow with Commands, Agents, Modes, and Session Memory appeared first on MarkTechPost.

Nous Research releases Contrastive Neuron Attribution (CNA), a method that identifies and ablates sparse MLP neuron circuits to steer LLM behavior — no sparse autoencoder training, no weight modification, and no degradation of general capability benchmarks.

The post Nous Research Releases Contrastive Neuron Attribution (CNA): Sparse MLP Circuit Steering Without SAE Training or Weight Modification appeared first on MarkTechPost.

Perplexity has open-sourced Bumblebee, an internal security tool it uses to protect the developer systems behind its search product, Comet, and Computer. Bumblebee is a read-only inventory collector for macOS and Linux developer endpoints. It scans npm, PyPI, Go modules, MCP configs, editor extensions, and browser extensions — without invoking any package manager or running any code.

The post Perplexity Open-Sources Bumblebee: A Read-Only Supply-Chain Scanner for Developer Endpoints appeared first on MarkTechPost.

AI agents start every session from zero — no memory of meetings, notes, or decisions. GBrain, the open-source memory layer Y Combinator's Garry Tan built to power his own OpenClaw and Hermes deployments, fixes that with a markdown-first knowledge graph that wires itself through regex inference, not LLM calls. This step-by-step coding tutorial walks through installing GBrain v0.38.2.0, building a brain repo, running hybrid search, and connecting it to Claude Code via MCP — about 20 minutes, all terminal output captured live.

The post A Step-by-Step Coding Tutorial to Implement GBrain: The Self-Wiring Memory Layer Built by Y Combinator’s Garry Tan for AI Agents appeared first on MarkTechPost.

Microsoft Research released Fara1.5, a family of browser computer-use agents in 4B, 9B, and 27B sizes. Fara1.5-27B scores 72% on Online-Mind2Web, outperforming OpenAI Operator, Gemini 2.5 Computer Use, and Yutori Navigator n1. The release also includes FaraGen1.5, a synthetic data pipeline that trains agents on gated

The post Microsoft Releases Fara1.5: A Family of Browser Computer-Use Agents (4B/9B/27B) That Outperform OpenAI Operator and Gemini 2.5 Computer Use on Online-Mind2Web appeared first on MarkTechPost.