Merge upstream→gfx11#1011

Merged

mgehre-amd merged 18 commits into

eble-amd:merge-from-upstream

Jun 19, 2026

Commits on Jun 2, 2026

[FlashAttention] Sync FA with upstream (#44065)
MatthewBonanni
authored
[BugFix] [GDN] Read linear_key_head_dim from hf_text_config for multimodal models (#43978)
IdoAtadTD
authored
[XPU] [Bug] remove xpuw4a16 output size check (#44168)

zufangzhu
and
jikunshang
authored
[Rust Frontend] Support recursive tool parameter conversion (#44299)
BugenZhao
authored
[ROCm] Fix AITER RMSNormQuantFusion for Kimi-Linear (#44308)
pschlan-amd
authored
[Rust Frontend] Cover different thinking modes in roundtrip tests (#44320)
BugenZhao
authored
Migrate header files to torch stable abi (#44013)
cleonard530
authored
[Misc] Support local image encoding in benchmarks (#43843)
xiaozcy
authored
[compressed-tensors] Asymmetric support for MoE WNA16 marlin (#44025)
brian-dellabetta
authored
[Core] Move `max_concurrent_batches` to `VllmConfig` (#44274)
njhill
authored
[Refactor] Remove dead code from parser infrastructure (#44279)
sfeng33
authored
[XPU] Enable rms_norm/act quant fusions (#43963)

zhenwei-intel
and
jikunshang
authored
[BugFix] Fix Humming MoE deploy error (#43100)

authored
[Bugfix] flashinfer: fail fast when --kv-cache-dtype nvfp4 used on unsupported arch (#43669)
Kartavyasonar
authored
[Feature] Support EPLB for DeepSeek v4 Mega Moe (#43339)

wzhao18
and
Wei Zhao (Engrg-Hardware 1)
authored
[Anthropic] Support system role messages inside messages array (#44283)

authored
[MRV2] Also enable MRV2 for Llama and Mistral dense models (#43458)

njhill
and
yewentao256
authored

Commits on Jun 19, 2026

Merge commit 'da107a59e5' into merge-from-upstream
eble-amd
committed