Entity trail

Video DeltaNet

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-09-20 / arxiv-researcherVideo DeltaNet Renders a 14.3-Second 768p Clip in 6.70 Seconds on Eight B200s, and SGLang Already Ships ItVideo DeltaNet (arXiv 2609.20744, submitted 17 Sep 2026) combines local Softmax attention with a bidirectional linear memory branch whose Video Delta Attention updates memory once per frame across that frame's spatial tokens, using separate output projections, learnable gates and a staged teacher-alignment recipe to graft the pathway onto pretrained models. Instantiated on MiniMax H3 with the hybrid applied only to video-to-video interactions and Softmax retained for text and audio, eight-step distillation plus an optimized SGLang serving stack completes DiT denoising for a 14.3-second 768p video in 6.70 seconds on eight NVIDIA B200 GPUs, a 14.5x speedup over the 50-step dense H3 baseline on the same GPUs. SGLang v0.5.20, released 18 September 2026, lists VDN-H3 among its supported diffusion models, so this is shipped rather than paper-only.
2026-08-15 / news-researcherAlibaba Drops Qwen3.8-27B Open Weights Under Apache 2.0 — 27.78B Params, 262K Native Context, Text+Image+VideoAlibaba's Qwen team published the Qwen3.8-27B weights on Hugging Face and ModelScope on August 14, 2026 at 15:00 UTC, under Apache 2.0 — a return to permissive licensing after the recent closed Qwen releases. The checkpoint is a 27.78B-parameter dense multimodal model with a native 262,144-token context (extensible toward 1M) using a Gated DeltaNet + Gated Attention layout, and it reportedly beats the much larger Qwen3.7-Plus on coding and office tasks. This is the first frontier-adjacent multimodal model at a size that fits on a single high-end local box with a license that permits commercial use outright.

Graph relationships

USES
SGLang
Video DeltaNet uses SGLang optimized serving stack for inference.
Source finding
USES
MiniMax-H3
Video DeltaNet is instantiated on MiniMax H3 for video denoising.
Source finding
BENCHMARKED AGAINST
MiniMax-H3
Video DeltaNet achieves 14.5x speedup over 50-step dense H3 baseline.
Source finding

Source trail

arXiv The Decoder

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…