- 27 Aug, 2025 1 commit
-
-
Ming-Xu Huang authored
* 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 <mingh@nvidia.com> * Slightly refactor Signed-off-by:
Phuong Nguyen <phuonguyen@nvidia.com>
-
- 26 Aug, 2025 5 commits
-
-
Vladimir Cherepanov authored
* Pick up cuBLASMp during build Signed-off-by:
Vladimir Cherepanov <vcherepanov@nvidia.com> * Saving... Signed-off-by:
Przemyslaw Tredak <ptrendx@gmail.com> Signed-off-by:
Vladimir Cherepanov <56651474+mk-61@users.noreply.github.com> * Add license Signed-off-by:
pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by:
-
Paweł Gadziński authored
fix test on old device Signed-off-by:Pawel Gadzinski <pgadzinski@nvidia.com>
-
vcherepanov-nv authored
* Bump cuDNN FE to 1.14.0 Signed-off-by:
-
Phuong Nguyen authored
* clean up sharding Signed-off-by:
-
Xin Yao authored
expose more activation functions Signed-off-by:Xin Yao <xiny@nvidia.com>
-
- 25 Aug, 2025 1 commit
-
-
Kshitij Lakhani authored
Add Transformer Layer tests for pre_scale_bias and post_scale_bias Signed-off-by:Kshitij Lakhani <klakhani@nvidia.com>
-
- 20 Aug, 2025 3 commits
-
-
jberchtold-nvidia authored
[JAX] Error checking for mesh resource and update GemmPrimitive to use global_mesh_resource().fsdp_resource (#2088) * Enforce global MeshResource is set Signed-off-by:
Jeremy Berchtold <jberchtold@nvidia.com> * Use global_mesh_resource().fsdp_resource in gemm primitive Signed-off-by:
-
Paweł Gadziński authored
* code drop Signed-off-by:
-
Jan Bielak authored
* Add rmsnorm_bwd_add Signed-off-by:
Jan Bielak <jbielak@nvidia.com> * Add BackwardAddRMSNorm fused operation Signed-off-by:
-
- 15 Aug, 2025 2 commits
-
-
Jan Bielak authored
* Add `nvte_cublas_gemm_scaled` Signed-off-by:
Tim Moon <4406448+timmoon10@users.noreply.github.com>
-
Kshitij Lakhani authored
* Move some dist fused attn tests to L2 1. TestReorderCausalLoadBalancing: Run two (non symmetric) BSHD/SBHD data shape combination 2. TestDistributedSelfAttn: Run only one (smaller) BSHD type data shape combination 3. TestDistributedCrossAttn: Run only one (smaller) BSHD type data shape combination 4. TestDistributedContextParallelSelfAttn: Run all cp1 combinations Signed-off-by:
Kshitij Janardan Lakhani <klakhani@login-preos01.a51.clusters.nvidia.com> * Remove test code for L2 dist residing in L2 test.sh Signed-off-by:
-
- 14 Aug, 2025 3 commits
-
-
Tim Moon authored
* Unfused impl for dbias-quantize Signed-off-by:
Tim Moon <tmoon@nvidia.com> * Unfused impl for dact-dbias-quantize Signed-off-by:
-
Xin Yao authored
* reduce driver calls Signed-off-by:
-
Kshiteej K authored
* fix: update grad_output quant to avoid redundant work Signed-off-by:
kshitij12345 <kshitijkalambarkar@gmail.com> * add test Signed-off-by:
Kirthi Shankar Sivamani <ksivamani@nvidia.com> --------- Signed-off-by:
-
- 13 Aug, 2025 2 commits
-
-
Paweł Gadziński authored
* code drop Signed-off-by:
-
jberchtold-nvidia authored
* Add L2_jax_distributed_unittest Signed-off-by:
-
- 09 Aug, 2025 1 commit
-
-
Daniel Stokes authored
* fix: Add stream synchronization before destroying MPI communicator (#1979) Signed-off-by:
djns99 <40156487+djns99@users.noreply.github.com> * feat: Implement column-wise userbuffer overlap for comm+GEMM operations Add support for overlapping column-wise allgather communication with GEMM operations to improve training performance: * **Core infrastructure changes:** - Update bulk_overlap_columnwise_ag() to accept explicit stream parameter - Modify userbuffers send/recv loops to use rank-ordered iteration - Add userbuffers_send_all/recv_all function declarations * **Python integration:** - Add bulk_overlap_ag_with_external_gemm() C++ extension function - Expose new overlap function via pybind11 bindings - Update overlap method configurations to include more ring_exchange ops * **LayerNorm MLP optimization:** - Enable column-wise quantization for FC2 gradient output - Implement overlap of allgather communication with FC2 DGRAD GEMM - Use fill_userbuffers_buffer_for_all_gather for efficient buffering This optimization allows overlapping communication and computation phases more effectively, reducing training wall-clock time by hiding allgather latency behind GEMM execution. Signed-off-by:
-
- 08 Aug, 2025 2 commits
-
-
Paweł Gadziński authored
* turn on userbuffers for layers without debug Signed-off-by:
-
Phuong Nguyen authored
* rm cudaGraph compatible trait from GroupedGEMM and groupedQuantize Signed-off-by:
-
- 07 Aug, 2025 1 commit
-
-
Phuong Nguyen authored
* rm batch_dim, sequence_dim, sequence_parallel_output Signed-off-by:
-
- 06 Aug, 2025 2 commits
-
-
hx authored
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor (#1978) * fix input_quantizer in save_original_input bwd Signed-off-by:
Hongxiao Bai <hongxiaob@nvidia.com> * fix get shape of blockwise tensor with only compact colwise data Signed-off-by:
-
jberchtold-nvidia authored
* Pytest timings Signed-off-by:
-
- 04 Aug, 2025 1 commit
-
-
Tim Moon authored
* Add basic kernel for swapping first two tensor dims Signed-off-by:
-
- 01 Aug, 2025 1 commit
-
-
Autumn1998 authored
* fix bug if all values<0 Signed-off-by:
tongliu <tongliu@nvidia.com> * minor fix Signed-off-by:
-
- 29 Jul, 2025 2 commits
-
-
Tim Moon authored
* remove reciprocal op Signed-off-by:
zhongboz <zhongboz@nvidia.com> * Refactor Quantizer::create_tensor function Signed-off-by:
-
Kirthi Shankar Sivamani authored
* Add verbosity only for failing tests Signed-off-by:
-
- 25 Jul, 2025 3 commits
-
-
Tim Moon authored
* Add ops for dropout and constant scale Signed-off-by:
-
Kirthi Shankar Sivamani authored
* Remove deprecated device arg Signed-off-by:
-
Evgeny Tsykunov authored
* Support RMSNorm for QK Signed-off-by:
Evgeny <etsykunov@nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * rms -> RMSNorm, l2 -> L2Normalization (align with current pattern) Signed-off-by:
-
- 24 Jul, 2025 4 commits
-
-
Oleg Goncharov authored
* Fixed integer overflow when computing offsets Signed-off-by:
Oleg Goncharov <ogoncharov@nvidia.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Signed-off-by:
-
Jan Bielak authored
* Mark output tensors as not deletable in backward Signed-off-by:
-
Evgeny Tsykunov authored
* Increase intermediate precision and reuse tensors from fwd Signed-off-by:
-
Phuong Nguyen authored
* add manage_primitives() helper * disable GEMM primitives for non-MXFP8 recipes * implement the NVTE_JAX_CUSTOM_CALLS + deprecate NVTE_JAX_CUSTOM_CALLS_RE * replace NVTE_JAX_CUSTOM_CALLS_RE with NVTE_JAX_CUSTOM_CALLS in TE tests and examples * fix use_jax_gemm contextmanager Signed-off-by:
-
- 23 Jul, 2025 2 commits
-
-
jberchtold-nvidia authored
Fix current scaling test_helper.py and enable test_helper.py in L0 Signed-off-by: -
Charlene Yang authored
* fix current device for cuDNN/cuBLAS handles Signed-off-by:
Charlene Yang <8636796+cyanguwa@users.noreply.github.com> * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * add unit test Signed-off-by:
-
- 22 Jul, 2025 2 commits
-
-
Jan Bielak authored
* Refactor _OperationFuserAutogradFunction.forward to use less parameters Signed-off-by:
-
Oleg Goncharov authored
* Fixed conflicts Signed-off-by:
-
- 21 Jul, 2025 1 commit
-
-
Charlene Yang authored
* exclude 9.10.0/.1 for certain configs Signed-off-by:
-
- 19 Jul, 2025 1 commit
-
-
jberchtold-nvidia authored
Update tolerance of distributed layernorm MLP for FP8 Signed-off-by:
-
