TJian 1806d1adfc [ROCm] [DSv4] [Perf] Support DeepSeek v4 MTP (#43385) ...

Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>

2026-05-24 18:43:08 +08:00

..

__init__.py

[1/N][Attention] Restructure attention: move files (#31916)

2026-01-09 13:10:24 -08:00

chunked_prefill_paged_decode.py

[ROCm] Cap Triton paged attention block size to fix ROCm shared memory OOM (#38502)

2026-05-10 10:03:00 +00:00

common.py

[Feat] DeepSeek V4 Rebased (#40860)

2026-04-26 18:31:08 -07:00

dcp_alltoall.py

[Kernel] Pack output and LSE in DCP A2A (#41160)

2026-05-01 09:01:17 -04:00

flashmla.py

[Refactor] Remove unused dead code (#38842)

2026-04-06 11:52:05 -04:00

merge_attn_states.py

[Kernel] Fuse FP8 output quantization into merge_attn_states (#36518)

2026-04-03 01:47:04 +00:00

paged_attn.py

[XPU][9/N] clean up existing ipex code/doc (#34111)

2026-02-11 00:27:15 -08:00

prefix_prefill.py

[ROCm] Support non-causal attention in ROCM_ATTN (#40176)

2026-04-22 12:57:12 +09:00

rocm_aiter_mla_sparse.py

[ROCm] [DSv4] [Perf] Support DeepSeek v4 MTP (#43385)

2026-05-24 18:43:08 +08:00

triton_attention_helpers.py

[Refactor] Unify 2D/3D kernels in triton_unified_attention (#40631)

2026-04-24 17:18:06 +02:00

triton_decode_attention.py

[ROCm][Bugfix]: dynamically align BLOCK_DMODEL with Lv in MLA decode kernel (#41119)

2026-05-11 11:14:19 +08:00

triton_merge_attn_states.py

[Kernel] Fuse FP8 output quantization into merge_attn_states (#36518)

2026-04-03 01:47:04 +00:00

triton_prefill_attention.py

[Performance] Improve Triton prefill attention kernel's performance (#32403)

2026-01-17 20:19:59 -08:00

triton_reshape_and_cache_flash.py

[Feature] KV cache per-token-head INT8/FP8 quantization (#38378)

2026-04-02 08:13:26 -04:00

triton_turboquant_decode.py

[Attention][TurboQuant] Share dequant buffers, eliminate float16_copy (#40941)

2026-04-27 13:48:36 +08:00

triton_turboquant_store.py

[ROCm] Fix TurboQuant on ROCm: backend routing, flash-attn compat, int64 overflow (#39953)

2026-04-17 13:08:37 -07:00

triton_unified_attention.py

[feat] Add FP8 per-tensor Q scale support to Triton attention backend (#42080)

2026-05-19 09:02:05 -07:00

vit_attn_wrappers.py

[Perf] FP8 FlashInfer Attn for ViT (#38065)

2026-04-27 13:44:15 +08:00

xpu_mla_sparse.py

Add XPU MLA Sparse backend for DeepSeek v3.2 (#33230)

2026-03-11 19:19:15 +08:00