All issues

Search results

[Issue]: GPT-OSS-120B fails on the ATOM SGLang stack (MXFP4 triton_kernel MoE)

Status: Open.

#2065 In ROCm/ATOM;

· ibeattyo opened on Aug 27, 2026

[Issue]: AITER OPUS FMHA 32-bit buffer-offset overflow on DeepSeek DP-attention at 64k (SGLang 0.5.17 regression)

Status: Open.

#2064 In ROCm/ATOM;

· ibeattyo opened on Aug 27, 2026

[Issue]: DeepSeek-V3.2 TP-attention (DSA) prefill GPU coredump on SGLang at 64k (bs >= 48)

Status: Open.

#2063 In ROCm/ATOM;

· ibeattyo opened on Aug 27, 2026

[Issue]: ATOM: TP-DSA decode hang in the plugin forward-relay

Status: Open.

#2062 In ROCm/ATOM;

· ibeattyo opened on Aug 27, 2026

[Issue]: DeepSeek V4 TP-MoE does not work

Status: Open.

#2022 In ROCm/ATOM;

· ibeattyo opened on Aug 25, 2026

[Issue]: Mori EP-MoE not CUDA-graph-capturable under TP-attention

Status: Open.

#2021 In ROCm/ATOM;

· ibeattyo opened on Aug 25, 2026

[Feature]: Multimodal support in ATOM

Status: Open.

#2014 In ROCm/ATOM;

· cubezhang opened on Aug 25, 2026

[Issue]:

Status: Open.

#1972 In ROCm/ATOM;

· ibeattyo opened on Aug 20, 2026

[Issue]: Kimi K3 ATOM+SGLang has ~309KB per token for KV while SGLang with no atom plugin uses only 26KB

Status: Open.

#1959 In ROCm/ATOM;

· ibeattyo opened on Aug 19, 2026

[Issue]:

Status: Open.

#1957 In ROCm/ATOM;

· ibeattyo opened on Aug 19, 2026

[Issue]: GPT-OSS-120B fails on the ATOM SGLang stack (MXFP4 triton_kernel MoE)

Status: Open.

#1945 In ROCm/ATOM;

· ibeattyo opened on Aug 18, 2026

[Feature]: FP8 block-quant MoE rejects TP shards narrower than block_n (Qwen3.6-35B-A3B)

Status: Open.

#1944 In ROCm/ATOM;

· ibeattyo opened on Aug 18, 2026