Sign in Subscribe

Bob M

Flash-dLLM's 11x Speedup for Diffusion Language Models

Flash-dLLM's 11x Speedup for Diffusion Language Models

Flash-dLLM reports 11x faster diffusion language model inference. And the claim comes with a named baseline: a 5.1x speedup on GSM8K and an 11.0x speedup on HumanEval over the prior strongest acceleration method, Elastic-Cache (paper listing). The framework, from authors Quan Nguyen-Tri, Mukul Ranjan. And Zhiqiang Shen, is

The Context Database for AI Agents: What OpenViking Actually Stores

The Context Database for AI Agents: What OpenViking Actually Stores

Not the model. The memory. Every argument I had about agent quality this year collapsed into that distinction eventually. And OpenViking is the project that finally names it: the repo lives in volcengine's GitHub org, it calls itself "The Context Database for AI Agents". Six words,

KV Cache Cost Attribution: Who's Actually Paying for GPU Memory

KV Cache Cost Attribution: Who's Actually Paying for GPU Memory

KV cache cost attribution means tying the memory side of your inference bill to the requests, sessions. And tenants that actually created it. That's the whole idea. And almost nobody does it, because the bill you get doesn't show memory at all. It shows tokens. The

Agentic Coding Environments: CodeMidas Builds 5,545 Tasks

Agentic Coding Environments: CodeMidas Builds 5,545 Tasks

CodeMidas turned 3,185 open-source codebases into 5,545 training tasks for agentic coding environments. And it never read an issue, a commit, or a written task record to do it. The paper behind that number, CodeMidas: Scaling Agentic Coding RL Environments from Code Itself, describes an agentic pipeline that

AI Agent Payment Authorization After 4,371 Attacks

AI Agent Payment Authorization After 4,371 Attacks

APort Vault replayed 4,371 human-written attacks against 14 language models from 8 labs. And the takeaway for AI agent payment authorization is blunt: the decision to move money should never live inside the model. The Open Agent Passport (OAP), the open specification APort built and stress-tested in that exercise,

OpenAI Multimodal Fine-Tuning Is Shutting Down: What the Notice Says

OpenAI Multimodal Fine-Tuning Is Shutting Down: What the Notice Says

OpenAI closed multimodal fine-tuning to new users on May 8, 2026. Existing users got "the coming months" to keep creating training jobs, which is a duration, not a date. Every fine-tuned model already in production stays alive exactly as long as its base model does. Not longer. Start