# Ggml-org merges Vulkan kernel fusion for Qwen4exp's SCALE‑sigmoid‑SCALE chain

Digest AI · Hardware & Compute · published 2026-09-28T12:59:41Z

Canonical: https://digestai.news/story/ggml-org-merges-vulkan-kernel-fusion-for-qwen4exp-s-scalesigmoidscale

## Summary

A pull request (#29520) was merged into the ggml‑org/llama.cpp repository on Sep 28 2026, adding a fused Vulkan kernel that combines the SCALE‑SIGMOID‑SCALE‑hcpost operations used by the Qwen4exp model. The change was approved by reviewers Jeffbolznv and 0cc4m and automatically tested by the github‑actions Bot. By collapsing the four separate tensor operations into a single GPU kernel, the update aims to reduce memory traffic and improve inference latency on Vulkan‑compatible devices.

The modification is part of ongoing efforts to optimise ggml’s Vulkan backend, which underpins many open‑source LLM inference projects. Faster execution of the Qwen4exp chain can benefit developers deploying the model on a range of hardware, from desktop GPUs to mobile devices that support Vulkan. The merge does not introduce new model weights or change model behavior, but it streamlines the compute path for existing deployments.

## Key points

- PR #29520 merged on Sep 28 2026 adds a fused Vulkan kernel for Qwen4exp's SCALE‑SIGMOID‑SCALE‑hcpost chain.
- Reviewers Jeffbolznv and 0cc4m approved the change; github‑actions Bot ran automated tests.
- Fusion reduces memory moves and should lower inference latency on Vulkan‑compatible GPUs.

## Why it matters

The kernel fusion speeds up Qwen4exp inference on Vulkan GPUs, helping developers run large language models more efficiently on diverse hardware.

## Sources

1. [vulkan: fuse qwen4exp's SCALE -> SIGMOID -> SCALE -> hc_post chain by fxgsell · Pull Request #29520 · ggml-org/llama.cpp](https://github.com/ggml-org/llama.cpp/pull/29520) (github.com, 2026-09-28, primary source)

## Cite

Digest AI, "Ggml-org merges Vulkan kernel fusion for Qwen4exp's SCALE‑sigmoid‑SCALE chain", 28 September 2026, https://digestai.news/story/ggml-org-merges-vulkan-kernel-fusion-for-qwen4exp-s-scalesigmoidscale

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/ggml-org-merges-vulkan-kernel-fusion-for-qwen4exp-s-scalesigmoidscale.json