Repository navigation

All pull requests

Labels (16)Milestones (0)New pull request

Open

(61)

Closed

(1,270)

[ROCm][RDNA] Skinny GEMMs: no VGPR spills on gfx11/gfx12 (bf16/fp16, int4)
#1401roberteg16 opened Oct 24, 2026
[ROCm][gfx1151] ROCM_RDNA_ATTN for multimodal encoders (non-causal)
#1400roberteg16 opened Oct 23, 2026
[ROCm][MiniCPM-V] Skip the ViT attention mask when no patch is padded
#1398stoivonen-amd opened Oct 23, 20263 of 5 tasks
[ROCm][gfx1151] ROCM_RDNA_ATTN: multimodal ranges in batched decode
#1396roberteg16 opened Oct 23, 2026
[ROCm][gfx1151] RDNA3.5 attention: decode and prefill rows for 17 more shapes
#1389roberteg16 opened Oct 20, 2026
4
wo_a optimization
#1387ahmed-bsod opened Oct 20, 20265 tasks
[ROCm][Perf] Use AITER GEMMs for bf16->fp32 MoE router and DSv4.1 KV-compressor projections on gfx1250
#1386ahmed-bsod opened Oct 20, 2026
[ROCm][RDNA4] ROCM_RDNA_SINGLESEQ_ATTN: gfx1200/gfx1201, per-family tables, tuned decode and prefill
#1383roberteg16 opened Oct 19, 2026
5
[ROCm][RDNA] RDNA single-seq: occupancy and tables of their own for 1024-VGPR archs
#1371roberteg16 opened Oct 14, 2026
1
Check supported backends in kernel test
#1364JadenMathias opened Oct 9, 2026
Fix overflow when running default context len for dsr1
#1358JadenMathias opened Oct 8, 2026
switch to triton a4w4 moe prefill kernel
#1348danichan-mkm opened Oct 6, 2026
[GFX12] DSV4-Pro Optimizations
#1345k50112113 opened Oct 6, 2026
chore: align foundational repo files with ROCm repo template
#1343dllehr-amd opened Oct 2, 2026Collaborator
[ROCm][gfx1151] rdna35_rope_cache: fused decode q/k RMSNorm + RoPE + KV cache write
#1342roberteg16 opened Oct 2, 20266 tasks done
[ROCm][W4A16] Add an opt-in hipBLASLt path for the gfx11 W4A16 GEMM
#1321mgehre-amd opened Sep 23, 2026
initial enablement and gates for dsv4.1f
#1317danichan-mkm opened Sep 21, 2026
[ROCm][gfx1151] Default the KV cache layout to HND for decode
#1315roberteg16 opened Sep 21, 20266 of 8 tasks
[Benchmarks] attention: keep the KV working set out of cache
#1311roberteg16 opened Sep 18, 20264 tasks done
1
Remove MOE_BOUND env being set
#1309JadenMathias opened Sep 17, 2026
Remove stale upstream-triton override in Dockerfile.rocm_gfx1250 (clobbers ROCm-built wheel)
#1307kiran-thumma opened Sep 16, 2026Collaborator2 tasks
Remove stale upstream-triton override in Dockerfile.rocm_gfx1250 (clobbers ROCm-built wheel)
#1306kiran-thumma opened Sep 16, 2026Collaborator2 tasks
routing through triton attn and padding gemm_afp4wfp4
#1285danichan-mkm opened Sep 9, 2026
1
fixes needed to enable kimi2.6 on gfx1250
#1282ahmed-bsod opened Sep 8, 20265 tasks
1
fix(amd-ci): DPX CI fixes — test isolation, memory wait, buildkite config
#1248mkunredd opened Sep 3, 20262 tasks
1

Pull requests