Gabriel WuandKimi Code 396c5a5632 [Bugfix] Lazy-import FlashInfer PCIe IPC all-reduce in kernel_warmup (#54869) ...

Signed-off-by: Zihua Wu <13583761+lucifer1004@users.noreply.github.com>
Co-authored-by: Kimi Code <noreply@moonshot.cn>

2026-09-02 04:31:03 +00:00

..

determinism

[Perf] Autotune batch invariance triton kernel in blackwell, 33.6% E2E latency reduction (#53649)

2026-08-25 15:06:01 -04:00

hw_agnostic

Hardware-agnostic model definition via HF transformer backend (1/N) (#49458)

2026-08-13 11:21:50 +01:00

kernels

[Performance] Register Triton W4A16 GEMM as a custom op (#51453)

2026-09-01 22:55:34 -05:00

layers

[Bugfix] Restore weight_dtype in QuarkW8A8Fp8MoEMethod to fix GPT-OSS FP8 MoE weight loading (#54824)

2026-09-01 16:08:11 -05:00

model_loader

[Model] Support Qwen3.8-Flash-Next (#53896)

2026-08-31 05:57:56 +00:00

models

[Perf] Avoid more h2d copies from non-pinned tensors (#54660)

2026-09-01 20:37:09 -07:00

offloader

[Offloader] Offload submodules that make_layers never reaches (#53120)

2026-08-26 18:37:00 +08:00

warmup

[Bugfix] Lazy-import FlashInfer PCIe IPC all-reduce in kernel_warmup (#54869)

2026-09-02 04:31:03 +00:00

__init__.py

[Platform] Deprecate seed_everything (#31659)

2026-01-04 18:34:04 -08:00

custom_op.py

[Bugfix] Reject contradictory custom-op directives (#49134)

2026-07-26 08:42:14 -04:00

parameter.py

Speed up docs build (#44635)

2026-06-05 14:51:44 +00:00

utils.py

[Weight processing] Copy over new_data attributes in replace_parameter (#49601)

2026-08-06 13:03:37 -04:00