Default Branch

4765f0f189 · [Bugfix] Fix sequence_parallel_chunk_impl custom op aliasing its input (#44130) · Updated 2026-06-05 23:56:36 +00:00

Branches

dee959c14d · minor comment · Updated 2026-06-06 00:03:01 +00:00

4
3

fa54d88f51 · Fix cumem allocator default to probe C extension availability · Updated 2026-06-05 19:24:12 +00:00

80
10

ef61b9735f · Merge branch 'main' into wentao-fix-v2-CohereASRDecoder · Updated 2026-06-05 19:18:28 +00:00

5
2

f62c1bb622 · remove dead mxfp8 code · Updated 2026-06-05 18:57:18 +00:00

5
1

cc6968a926 · Bump the minor-update group across 1 directory with 149 updates · Updated 2026-06-05 16:59:49 +00:00

7
1

5919036d8c · Pack KV caches into contiguous per-block allocations for DeepSeek V4 · Updated 2026-06-05 16:19:05 +00:00

15
1

81931a2cef · Merge branch 'main' into wentao-fp8-scaled-mm-oddM · Updated 2026-06-05 15:25:58 +00:00

10
2

e010548b9d · Remove unused is_kv_layout_blocks_first from TransferTopology · Updated 2026-06-05 15:19:23 +00:00

69
12

5ec1606eb7 · Merge branch 'main' into lwilkinson/kv-layout/bucket-layers-refactor · Updated 2026-06-05 14:19:20 +00:00

14
5

f191ebee69 · Run tool parser tests before tokenizer tests · Updated 2026-06-05 13:46:47 +00:00

28
7

76c973e13c · refactor & simplify · Updated 2026-06-05 06:39:10 +00:00

205
37

7b375c8502 · Fix ROCm artifact test dependencies · Updated 2026-06-05 02:30:21 +00:00

36
1

3d7a85964c · Merge branch 'main' into wentao-add-reset-cache-for-v1 · Updated 2026-06-04 19:40:13 +00:00

39
3

45dfecbe55 · Merge branch 'main' into wentao-optimize-per-token-group-quant · Updated 2026-06-04 19:09:36 +00:00

39
10

5e3d8ee466 · Merge branch 'main' into wentao-fix-es-v2-bug · Updated 2026-06-04 18:36:12 +00:00

41
6

d47419733a · Merge branch 'main' into wentao-mrv2-quantized · Updated 2026-06-04 18:33:51 +00:00

42
2

7d6f3025b0 · Merge branch 'main' into wentao-enable-all-dense-for-mrv2 · Updated 2026-06-04 18:33:39 +00:00

43
3

d5022f3904 · [KVCache] Standardize KV cache layout and remove legacy shape/stride APIs · Updated 2026-06-04 04:05:57 +00:00

69
9

5b204aec14 · [KVCache] Remove _update_hybrid_attention_layout (dead after K/V packing) · Updated 2026-06-04 03:59:42 +00:00

69
8

0decac0d96 · fix: resolve CUTLASS fmin compatibility for DeepSeek-V4 init · Updated 2026-06-04 00:11:47 +00:00

386
27