Update list of authorized CI users #2152
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Tim Moon <[email protected]>
vthumbe1503
pushed a commit
to vthumbe1503/TransformerEngine
that referenced
this pull request
Sep 8, 2025
Signed-off-by: Tim Moon <[email protected]> Signed-off-by: Varun Thumbe <[email protected]>
vthumbe1503
added a commit
to vthumbe1503/TransformerEngine
that referenced
this pull request
Sep 19, 2025
Signed-off-by: Varun Thumbe <[email protected]> [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci Signed-off-by: Varun Thumbe <[email protected]> Add cuBLASMp-backed GEMM-like API to TE common (NVIDIA#1824) * Pick up cuBLASMp during build Signed-off-by: Vladimir Cherepanov <[email protected]> * Saving... Signed-off-by: Vladimir Cherepanov <[email protected]> * Change lib order to fix link error Signed-off-by: Vladimir Cherepanov <[email protected]> * Saving... Signed-off-by: Vladimir Cherepanov <[email protected]> * Context creation, incomplete... Signed-off-by: Vladimir Cherepanov <[email protected]> * Test fixure Signed-off-by: Vladimir Cherepanov <[email protected]> * Saving... Signed-off-by: Vladimir Cherepanov <[email protected]> * A sanity AgGemm test, failing... Signed-off-by: Vladimir Cherepanov <[email protected]> * Saving... Signed-off-by: Vladimir Cherepanov <[email protected]> * Fix axes Signed-off-by: Vladimir Cherepanov <[email protected]> * Take care of uneven distribution Signed-off-by: Vladimir Cherepanov <[email protected]> * Use MPI to get position of local matrices Signed-off-by: Vladimir Cherepanov <[email protected]> * Refactor Signed-off-by: Vladimir Cherepanov <[email protected]> * Refactor & fixes Signed-off-by: Vladimir Cherepanov <[email protected]> * Saving... Signed-off-by: Vladimir Cherepanov <[email protected]> * Gemm-RS Signed-off-by: Vladimir Cherepanov <[email protected]> * Gemm-AR, not working... Signed-off-by: Vladimir Cherepanov <[email protected]> * Fixes Signed-off-by: Vladimir Cherepanov <[email protected]> * Setting all-reduce epilogue for gemm-ar Signed-off-by: Vladimir Cherepanov <[email protected]> * Use supported shapes for GEMM-AR Signed-off-by: Vladimir Cherepanov <[email protected]> * Tweak tolerance Signed-off-by: Vladimir Cherepanov <[email protected]> * First shot at fp8 Signed-off-by: Vladimir Cherepanov <[email protected]> * Use TensorHolder in tests Signed-off-by: Vladimir Cherepanov <[email protected]> * More test configs Signed-off-by: Vladimir Cherepanov <[email protected]> * Support comm_sm_count Signed-off-by: Vladimir Cherepanov <[email protected]> * Parametrize dtypes for A, B and D separately Signed-off-by: Vladimir Cherepanov <[email protected]> * Tweak scaling Signed-off-by: Vladimir Cherepanov <[email protected]> * Amax ptr Signed-off-by: Vladimir Cherepanov <[email protected]> * Flags parity with cublas_gemm, saving... Signed-off-by: Vladimir Cherepanov <[email protected]> * Cleanup Signed-off-by: Vladimir Cherepanov <[email protected]> * Bias tests Signed-off-by: Vladimir Cherepanov <[email protected]> * Fix bias test Signed-off-by: Vladimir Cherepanov <[email protected]> * Aux, saving... Signed-off-by: Vladimir Cherepanov <[email protected]> * aux_ld Signed-off-by: Vladimir Cherepanov <[email protected]> * A fix Signed-off-by: Vladimir Cherepanov <[email protected]> * Use test::Tensor Signed-off-by: Vladimir Cherepanov <[email protected]> * Set scale inv Signed-off-by: Vladimir Cherepanov <[email protected]> * Remove unsupported test configs Signed-off-by: Vladimir Cherepanov <[email protected]> * Tweak tests Signed-off-by: Vladimir Cherepanov <[email protected]> * Replace libcal with NCCL Signed-off-by: Vladimir Cherepanov <[email protected]> * Add NVTX markers to API functions Signed-off-by: Vladimir Cherepanov <[email protected]> * Tweak GemmAr tests Signed-off-by: Vladimir Cherepanov <[email protected]> * More test config Signed-off-by: Vladimir Cherepanov <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci Signed-off-by: Vladimir Cherepanov <[email protected]> * Fix merge fallout Signed-off-by: Vladimir Cherepanov <[email protected]> * Remove MPI dependency, comment API, add algo parameter Signed-off-by: Vladimir Cherepanov <[email protected]> * Fix nvshmem dependency Signed-off-by: Vladimir Cherepanov <[email protected]> * Fix nvshmem build Signed-off-by: Vladimir Cherepanov <[email protected]> * Excluse CommGemm tests from L0_cppunittest Signed-off-by: Vladimir Cherepanov <[email protected]> * Add cpp_distributed sh file for CI Signed-off-by: Vladimir Cherepanov <[email protected]> * Adapt tp TensorAllocator Signed-off-by: Vladimir Cherepanov <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Skip GemmAr test on unsupported HW Signed-off-by: Vladimir Cherepanov <[email protected]> * Oversibscribe is needed on some clusters Signed-off-by: Vladimir Cherepanov <[email protected]> * Fix incomplete libcal removal Signed-off-by: Vladimir Cherepanov <[email protected]> * Move CI tests to L1 Signed-off-by: Vladimir Cherepanov <[email protected]> * Rename context to include NVTE prefix Signed-off-by: Vladimir Cherepanov <[email protected]> * Remove leftover code Signed-off-by: Vladimir Cherepanov <[email protected]> * NVTE_WITH_CUBLASMP off by default Signed-off-by: Vladimir Cherepanov <[email protected]> * More detailed NVTE_CHECK diag Signed-off-by: Vladimir Cherepanov <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Comment API Signed-off-by: Vladimir Cherepanov <[email protected]> * Include stdbool header for legacy C compilers Signed-off-by: Vladimir Cherepanov <[email protected]> * Remove now unused argument Signed-off-by: Vladimir Cherepanov <[email protected]> * Abstract away cuBLASMp algo behind our own enum Signed-off-by: Vladimir Cherepanov <[email protected]> * More detailed shape diag messages Signed-off-by: Vladimir Cherepanov <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update transformer_engine/common/include/transformer_engine/comm_gemm.h Co-authored-by: Przemyslaw Tredak <[email protected]> Signed-off-by: Vladimir Cherepanov <[email protected]> * Add license Signed-off-by: Vladimir Cherepanov <[email protected]> --------- Signed-off-by: Vladimir Cherepanov <[email protected]> Signed-off-by: Vladimir Cherepanov <[email protected]> Co-authored-by: Vladimir Cherepanov <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Przemyslaw Tredak <[email protected]> Signed-off-by: Varun Thumbe <[email protected]> [PyTorch][CUDA Graph] Fix FP8 Weight Quantization Cache under CUDA Graph (NVIDIA#2119) * add noop to comp amax Signed-off-by: zhongboz <[email protected]> * fix for fp8 blockwise recipe Signed-off-by: zhongboz <[email protected]> * resolve comments Signed-off-by: zhongboz <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Signed-off-by: zhongboz <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Tim Moon <[email protected]> Signed-off-by: Varun Thumbe <[email protected]> [PyTorch] fix cross entropy vanishing gradients (NVIDIA#2139) * fix cross entropy Signed-off-by: Casper <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci Signed-off-by: Casper <[email protected]> * fix comments Signed-off-by: Casper <[email protected]> * fix: few more style issues Signed-off-by: Casper <[email protected]> * fix: remove grad_output_stride (unnecessary) Signed-off-by: Casper <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix: only backward was broken Signed-off-by: Casper <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Generalize cross entropy backward kernel to handle reduced and unreduced loss Signed-off-by: Tim Moon <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Signed-off-by: Casper <[email protected]> Signed-off-by: Tim Moon <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Tim Moon <[email protected]> Co-authored-by: Tim Moon <[email protected]> Signed-off-by: Varun Thumbe <[email protected]> Fix bug when enabling --overlap-grad-reduce in mcore (NVIDIA#2142) * fix bugs when enabling --overlap-grad-reduce in mcore Signed-off-by: Hongbin Liu <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix CI Signed-off-by: Hongbin Liu <[email protected]> * format Signed-off-by: Hongbin Liu <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Signed-off-by: Hongbin Liu <[email protected]> Co-authored-by: Hongbin Liu <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Signed-off-by: Varun Thumbe <[email protected]> Fix CUDA version in setup.py (NVIDIA#2132) * Fix CUDA version in setup.py Signed-off-by: Vladimir Cherepanov <[email protected]> * Re-enable building comm-gemm tests Signed-off-by: Vladimir Cherepanov <[email protected]> * WAR for nvidia-nvshmem package Signed-off-by: Vladimir Cherepanov <[email protected]> --------- Signed-off-by: Vladimir Cherepanov <[email protected]> Co-authored-by: Tim Moon <[email protected]> Signed-off-by: Varun Thumbe <[email protected]> [JAX] NoScaleTensor wrapper for non-quantized data (NVIDIA#2136) * Custom call tests passing Signed-off-by: Jeremy Berchtold <[email protected]> * Fix test_layer.py Signed-off-by: Jeremy Berchtold <[email protected]> * Lint Signed-off-by: Jeremy Berchtold <[email protected]> * Fix comments Signed-off-by: Jeremy Berchtold <[email protected]> * Support using amax on HighPrecision tensor if it exists instead of recomputing for current scaling Signed-off-by: Jeremy Berchtold <[email protected]> * Fix shardy issue with amax being shape 1,1,1 instead of shape (1,) Signed-off-by: Jeremy Berchtold <[email protected]> * Add higher-precision VJP tests to test_distributed_layernorm_mlp Signed-off-by: Jeremy Berchtold <[email protected]> * Cast non-quantized kernels to input dtype in VJPs Signed-off-by: Jeremy Berchtold <[email protected]> * Rename HighPrecisionTensor to NoScaleTensor Signed-off-by: Jeremy Berchtold <[email protected]> * Use NoScaleTensor in pure JAX impls where it was missing Signed-off-by: Jeremy Berchtold <[email protected]> * Fix tests Signed-off-by: Jeremy Berchtold <[email protected]> --------- Signed-off-by: Jeremy Berchtold <[email protected]> Signed-off-by: Varun Thumbe <[email protected]> [JAX] Fix GroupedScaledTensor creation with keyword arg (NVIDIA#2154) Fix GroupedScaledTensor creation Signed-off-by: Phuong Nguyen <[email protected]> Signed-off-by: Varun Thumbe <[email protected]> Fixing few issues with multi-process launching. (NVIDIA#2155) * Fixing few issues with multi-process launching. Signed-off-by: Ming Huang <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Signed-off-by: Ming Huang <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Phuong Nguyen <[email protected]> Signed-off-by: Varun Thumbe <[email protected]> Update list of authorized CI users (NVIDIA#2152) Signed-off-by: Tim Moon <[email protected]> Signed-off-by: Varun Thumbe <[email protected]> a bit of cleanup Signed-off-by: Varun Thumbe <[email protected]>
vthumbe1503
added a commit
to vthumbe1503/TransformerEngine
that referenced
this pull request
Sep 19, 2025
author Varun Thumbe <[email protected]> 1757373536 +0000 committer Varun Thumbe <[email protected]> 1758262513 +0000 parent de9ef2f author Varun Thumbe <[email protected]> 1757373536 +0000 committer Varun Thumbe <[email protected]> 1758262476 +0000 parent de9ef2f author Varun Thumbe <[email protected]> 1757373536 +0000 committer Varun Thumbe <[email protected]> 1758262304 +0000 merge conflict Signed-off-by: Varun Thumbe <[email protected]> [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci Signed-off-by: Varun Thumbe <[email protected]> FP8 AllGather in FP8 GroupedGEMM + Fix Stream Usage Issue. (NVIDIA#2086) * FP8 AllGather in FP8 GroupedGEMM 1. Support current scaling FP8 quantation with a given amax. 2. Support FP8 AG in fwd and BF16 RS in bwd. 3. The workflow is AR-max -> FP8 Quant -> FP8 AG -> FP8 GroupedGEMM. Signed-off-by: Ming Huang <[email protected]> * Slightly refactor Signed-off-by: Ming Huang <[email protected]> * Adding documents of new args. Signed-off-by: Ming Huang <[email protected]> * Adding unit-tests. Signed-off-by: Ming Huang <[email protected]> * Adding license. Signed-off-by: Ming Huang <[email protected]> * Move unit-tests to L1. Signed-off-by: Ming Huang <[email protected]> * Move quantizaer store/reset into FP8 only. Signed-off-by: Ming Huang <[email protected]> * Adding all layout support for Blackwell+ Signed-off-by: Ming Huang <[email protected]> * Adopt the feedback from code-review. Signed-off-by: Ming Huang <[email protected]> * Fixed the wrong stream used by d2d in groupedGEMM FFI. Signed-off-by: Ming Huang <[email protected]> --------- Signed-off-by: Ming Huang <[email protected]> Co-authored-by: Phuong Nguyen <[email protected]> [JAX] Delay MeshResource validation until first usage (NVIDIA#2124) Delay MeshResource validation until first usage Signed-off-by: Jeremy Berchtold <[email protected]> Co-authored-by: Phuong Nguyen <[email protected]> [JAX] `dot_1_output` sharding constraint + use AXIS_IS_UNSHARDED (NVIDIA#2128) * add dot_1_output sharding constraint + use AXIS_IS_UNSHARDED Signed-off-by: Phuong Nguyen <[email protected]> --------- Signed-off-by: Phuong Nguyen <[email protected]> [JAX] Add amax input to DBiasQuantizePrimitive and FFI (NVIDIA#2118) * add amax input to DBiasQuantizePrimitive and FFI Signed-off-by: Phuong Nguyen <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * make sure amax is init with zero Signed-off-by: Phuong Nguyen <[email protected]> * fix sharding rule Signed-off-by: Phuong Nguyen <[email protected]> --------- Signed-off-by: Phuong Nguyen <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Further relax constraints to cuDNN 9.13 for disabling fused attn for kv caching (NVIDIA#2121) Signed-off-by: Kshitij Lakhani <[email protected]> Temporarily remove comm_gemm tests (NVIDIA#2133) Signed-off-by: Vladimir Cherepanov <[email protected]> [PyTorch] Disable determinism for sm100 (NVIDIA#2130) * disable determinism for sm100+ and cudnn<9.14 Signed-off-by: Charlene Yang <[email protected]> * fix remaining CI failures Signed-off-by: Charlene Yang <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * revert some changes Signed-off-by: Charlene Yang <[email protected]> * revert more changes Signed-off-by: Charlene Yang <[email protected]> * remove sm100 from determinism table Signed-off-by: Charlene Yang <[email protected]> --------- Signed-off-by: Charlene Yang <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> [PyTorch] ONNX export of FP8 Current Scaling (NVIDIA#2068) * Compute amax in normalization forward in current scaling in untuned kernels Signed-off-by: Jan Bielak <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix Signed-off-by: Pawel Gadzinski <[email protected]> * fix Signed-off-by: Pawel Gadzinski <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix Signed-off-by: Pawel Gadzinski <[email protected]> * code drop Signed-off-by: Pawel Gadzinski <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix Signed-off-by: Pawel Gadzinski <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix Signed-off-by: Pawel Gadzinski <[email protected]> * apply tims suggestions Signed-off-by: Pawel Gadzinski <[email protected]> --------- Signed-off-by: Jan Bielak <[email protected]> Signed-off-by: Pawel Gadzinski <[email protected]> Co-authored-by: Jan Bielak <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> [PyTorch][MOE] Tentative Fix For Replacing from_blob with empty for experts receiving zero tokens (NVIDIA#2134) use torch empty for empty shape instead of from_blob Signed-off-by: zhongboz <[email protected]> Co-authored-by: Kirthi Shankar Sivamani <[email protected]> build: pull cached wheels (NVIDIA#2127) * build: pull cached wheels Signed-off-by: oliver könig <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Update setup.py Signed-off-by: oliver könig <[email protected]> --------- Signed-off-by: oliver könig <[email protected]> Co-authored-by: Kirthi Shankar Sivamani <[email protected]> [Common] Add checks to CUDA kernel launch and CUDA API calls (NVIDIA#2074) * add checks to cuda kernel launch and cuda API calls Signed-off-by: Xin Yao <[email protected]> * Remove exceptions from destructors Signed-off-by: Tim Moon <[email protected]> * fix weired dispatch in ln/rmsnorm Signed-off-by: Xin Yao <[email protected]> --------- Signed-off-by: Xin Yao <[email protected]> Signed-off-by: Tim Moon <[email protected]> Co-authored-by: Tim Moon <[email protected]> Co-authored-by: Tim Moon <[email protected]> [PyTorch] Support bf16+fp8 cudagraph (NVIDIA#2098) * support bf16+fp8 model Signed-off-by: Robin Zhang <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * update Signed-off-by: Robin Zhang <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * update Signed-off-by: Robin Zhang <[email protected]> --------- Signed-off-by: Robin Zhang <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Tim Moon <[email protected]> Dropout with 8-bit RNG (NVIDIA#2014) * Add dropout kernel with 8-bit RNG Co-authored-by: Vasudevan Rengasamy <[email protected]> Co-authored-by: Tim Moon <[email protected]> Signed-off-by: Tim Moon <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Fix license Signed-off-by: Tim Moon <[email protected]> * Avoid ambiguous types Signed-off-by: Tim Moon <[email protected]> * Do not enforce dropout prob is representable in 8 bits Signed-off-by: Tim Moon <[email protected]> * Expand error message Signed-off-by: Tim Moon <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Fix small statistical bug from using less-equal instead of less-than Refactor kernel implementations and add comments. Interpret masks as bytes rather than 16-bit uints. Signed-off-by: Tim Moon <[email protected]> * Fix linter warning Signed-off-by: Tim Moon <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Remove unnecessary helper function in PyTorch extensions Signed-off-by: Tim Moon <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Signed-off-by: Tim Moon <[email protected]> Co-authored-by: Tim Moon <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Create GPU reload buffers on main stream (NVIDIA#2131) * Create GPU relaod buffers on main stream Signed-off-by: Selvaraj Anandaraj <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Fixed typo Signed-off-by: Selvaraj Anandaraj <[email protected]> * Fixed typo Signed-off-by: Selvaraj Anandaraj <[email protected]> --------- Signed-off-by: Selvaraj Anandaraj <[email protected]> Signed-off-by: Selvaraj Anandaraj <[email protected]> Co-authored-by: Selvaraj Anandaraj <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Selvaraj Anandaraj <[email protected]> Co-authored-by: Paweł Gadziński <[email protected]> Fix CI failures for UB overlap changes (NVIDIA#2149) Signed-off-by: djns99 <[email protected]> [JAX] Fix failing fused attn tests for dropout=0.1 and bias for sm100 (NVIDIA#2135) * Fix failing tests for dropout=0.1 and bias for fused attn for blackwell Signed-off-by: Kshitij Lakhani <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Fix the skip message Signed-off-by: Kshitij Lakhani <[email protected]> * Assert in fused attn bwd pass for sm100 Signed-off-by: Kshitij Lakhani <[email protected]> Add check for sm100 Signed-off-by: Kshitij Lakhani <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Add support to get all devs in the process for jax Signed-off-by: Kshitij Lakhani <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Code clean up Signed-off-by: Kshitij Lakhani <[email protected]> * Make get_all_device_compute_capability more pythonic, thereby avoiding unnecessary type conversion Signed-off-by: Kshitij Lakhani <[email protected]> * Represent attn bias using enum instead of string Signed-off-by: Kshitij Lakhani <[email protected]> --------- Signed-off-by: Kshitij Lakhani <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> [PyTorch][CUDA Graph] Fix FP8 Weight Quantization Cache under CUDA Graph (NVIDIA#2119) * add noop to comp amax Signed-off-by: zhongboz <[email protected]> * fix for fp8 blockwise recipe Signed-off-by: zhongboz <[email protected]> * resolve comments Signed-off-by: zhongboz <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Signed-off-by: zhongboz <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Tim Moon <[email protected]> [PyTorch] fix cross entropy vanishing gradients (NVIDIA#2139) * fix cross entropy Signed-off-by: Casper <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci Signed-off-by: Casper <[email protected]> * fix comments Signed-off-by: Casper <[email protected]> * fix: few more style issues Signed-off-by: Casper <[email protected]> * fix: remove grad_output_stride (unnecessary) Signed-off-by: Casper <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix: only backward was broken Signed-off-by: Casper <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Generalize cross entropy backward kernel to handle reduced and unreduced loss Signed-off-by: Tim Moon <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Signed-off-by: Casper <[email protected]> Signed-off-by: Tim Moon <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Tim Moon <[email protected]> Co-authored-by: Tim Moon <[email protected]> Fix bug when enabling --overlap-grad-reduce in mcore (NVIDIA#2142) * fix bugs when enabling --overlap-grad-reduce in mcore Signed-off-by: Hongbin Liu <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * fix CI Signed-off-by: Hongbin Liu <[email protected]> * format Signed-off-by: Hongbin Liu <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Signed-off-by: Hongbin Liu <[email protected]> Co-authored-by: Hongbin Liu <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Fix CUDA version in setup.py (NVIDIA#2132) * Fix CUDA version in setup.py Signed-off-by: Vladimir Cherepanov <[email protected]> * Re-enable building comm-gemm tests Signed-off-by: Vladimir Cherepanov <[email protected]> * WAR for nvidia-nvshmem package Signed-off-by: Vladimir Cherepanov <[email protected]> --------- Signed-off-by: Vladimir Cherepanov <[email protected]> Co-authored-by: Tim Moon <[email protected]> [JAX] NoScaleTensor wrapper for non-quantized data (NVIDIA#2136) * Custom call tests passing Signed-off-by: Jeremy Berchtold <[email protected]> * Fix test_layer.py Signed-off-by: Jeremy Berchtold <[email protected]> * Lint Signed-off-by: Jeremy Berchtold <[email protected]> * Fix comments Signed-off-by: Jeremy Berchtold <[email protected]> * Support using amax on HighPrecision tensor if it exists instead of recomputing for current scaling Signed-off-by: Jeremy Berchtold <[email protected]> * Fix shardy issue with amax being shape 1,1,1 instead of shape (1,) Signed-off-by: Jeremy Berchtold <[email protected]> * Add higher-precision VJP tests to test_distributed_layernorm_mlp Signed-off-by: Jeremy Berchtold <[email protected]> * Cast non-quantized kernels to input dtype in VJPs Signed-off-by: Jeremy Berchtold <[email protected]> * Rename HighPrecisionTensor to NoScaleTensor Signed-off-by: Jeremy Berchtold <[email protected]> * Use NoScaleTensor in pure JAX impls where it was missing Signed-off-by: Jeremy Berchtold <[email protected]> * Fix tests Signed-off-by: Jeremy Berchtold <[email protected]> --------- Signed-off-by: Jeremy Berchtold <[email protected]> [JAX] Fix GroupedScaledTensor creation with keyword arg (NVIDIA#2154) Fix GroupedScaledTensor creation Signed-off-by: Phuong Nguyen <[email protected]> Fixing few issues with multi-process launching. (NVIDIA#2155) * Fixing few issues with multi-process launching. Signed-off-by: Ming Huang <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Signed-off-by: Ming Huang <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Phuong Nguyen <[email protected]> Update list of authorized CI users (NVIDIA#2152) Signed-off-by: Tim Moon <[email protected]> Fused RoPE with combined QKV input. (NVIDIA#2122) * Fused RoPE with combined QKV input. Initial commit for Dropout with 8-bit RNG Fix documentation Initial commit for Fused QKV RoPE WIP Initial tests passing Enable rotary percent and margin Enable CP2, start_positions, interleaved Cleanup test Revert "Fix documentation" This reverts commit 53df100. Revert "Initial commit for Dropout with 8-bit RNG" This reverts commit 301505e. Cleanup. Minor cleanup Signed-off-by: Vasudevan Rengasamy <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci Signed-off-by: Vasudevan Rengasamy <[email protected]> * Optimize kernels Signed-off-by: Vasudevan Rengasamy <[email protected]> * Misc. Cleanup Signed-off-by: Vasudevan Rengasamy <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci Signed-off-by: Vasudevan Rengasamy <[email protected]> * Optimize kernel performance Signed-off-by: Vasudevan Rengasamy <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci Signed-off-by: Vasudevan Rengasamy <[email protected]> * Move fused_qkv_rope test to test_fused_rope.py Signed-off-by: Vasudevan Rengasamy <[email protected]> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * apply shared memory optimization to separate fused rope kernels Signed-off-by: Xin Yao <[email protected]> * fix lint Signed-off-by: Xin Yao <[email protected]> --------- Signed-off-by: Vasudevan Rengasamy <[email protected]> Signed-off-by: Xin Yao <[email protected]> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Xin Yao <[email protected]> Co-authored-by: Tim Moon <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
Update list of authorized CI users
Type of change
Changes
Checklist: