Search results

How to set kv data in mla_decode_fwd()?

Status: Open.

#200 In ROCm/aiter;

· ghostplant opened on Mar 14, 2025

[Feature]: TopK weights scaled on tokens with Shared Experts Support

Status: Open.

#195 In ROCm/aiter;

· sijiac opened on Mar 13, 2025

[Feature]: Optional topk_weights arg in Fused MoE B2B GEMM

Status: Open.

#193 In ROCm/aiter;

· sijiac opened on Mar 11, 2025

[Feature]: support seqlen_q large than 1 for PagedAttention with FP8KV

Status: Open.

#181 In ROCm/aiter;

· sijiac opened on Mar 6, 2025

[Issue]: Build errors for gfx90a (MI250) architecture

Status: Open.

#179 In ROCm/aiter;

· OscarSavNS opened on Mar 5, 2025

[Feature]: enable multi-blocks execution for moe align kernel

Status: Open.

#107 In ROCm/aiter;

· yiakwy-xpu-ml-framework-team opened on Feb 7, 2025

[Issue]: mixtral 8x7b fp8 tp1 performance degraded with aiter

Status: Open.

#91 In ROCm/aiter;

· andyluo7 opened on Feb 5, 2025

Inference attention tech path

Status: Open.

#79 In ROCm/aiter;

· carlushuang opened on Jan 27, 2025

deepseek-v3/R1 tech path

Status: Open.

#75 In ROCm/aiter;

· carlushuang opened on Jan 23, 2025