Tuesday, Sep 8, 2026

Top stories

Speculative Decoding in vLLM on AMD GPUs

vLLM published a benchmark-heavy guide to speculative decoding on AMD Instinct MI300X/MI355X GPUs (ROCm), comparing five drafting approaches — native and separate MTP modules plus dedicated speculator networks (EAGLE-3, DFlash, DSpark) — against baseline autoregressive decoding across GSM8K, MATH500, HumanEval, and MBPP. Speedups varied sharply by model and method: up to 2.87x on Gemma 4 26B-A4B and 2.68x on Kimi-K2.5 with DFlash on MATH500, while Qwen models ranged from a modest 1.08x to 2.20x depending on drafter and proposal length. It's not a new vLLM release — speculative decoding already existed — but it's a rare task-grounded, tunable-by-tunable look at squeezing real throughput out of AMD hardware, an area with much thinner public benchmarking than NVIDIA gets.

Sources & depth

Two vibe-coded builds: GPT-6 Astra and Claude Fable 5.1

Simon Willison posted two same-day "I had a model build this" writeups. First: an animated Mercator-to-Equal-Earth map-projection transition built with GPT-6 Astra (medium) in ChatGPT Work using D3, prompted by the UN's recent vote to adopt the Equal Earth projection. Second: a browser-based video compressor built with Claude Fable 5.1 in Claude Code for the web, wrapping the WebAssembly build of FFmpeg to shrink a phone-recorded demo clip for his blog. Neither is a formal benchmark, but both are first-hand, reproducible data points — from a credible, technical author — on what a single vibe-coding session with each model can actually produce today.

Sources & depth
Also today