Qwen3.8-27B latent attention retrofits Collection Training-free MLA / GLA retrofits of Qwen3.8-27B with a 2x smaller KV cache. Weights preview; vLLM plugin, quants and code coming early October 2026. • 8 items • Updated 5 days ago • 6
Improved Distribution Matching Distillation for Fast Image Synthesis Paper • 2405.14867 • Published May 23, 2024 • 20
Iris-3B: Going Beyond the Latent with Pixel-Space Diffusion Training, Conversion and Fine-Tuning Paper • 2610.09450 • Published 13 days ago • 32
Learning Few-Step Diffusion Models by Trajectory Distribution Matching Paper • 2503.06674 • Published Mar 9, 2025 • 10
Running on Zero Agents 13 Iris-3B Turbo ⚡ 13 4-step pixel-space text-to-image, distilled from Iris-3B