Skip to content

[Metax] Fix add flags (#227) #2527

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
metax666 merged 1 commit into from
Apr 27, 2026
Merged

[Metax] Fix add flags (#227) #2527

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion Paddle
Open in desktop
Submodule Paddle updated 28 files
+12 −2 cmake/cupti.cmake
+30 −0 paddle/common/flags.cc
+13 −0 paddle/fluid/pybind/eager.h
+185 −14 paddle/fluid/pybind/eager_py_layer.cc
+7 −2 paddle/phi/CMakeLists.txt
+2 −0 paddle/phi/backends/dynload/rocm_driver.cc
+1 −19 paddle/phi/kernels/cpu/elementwise.h
+29 −14 paddle/phi/kernels/funcs/distribution_helper.h
+26 −13 paddle/phi/kernels/funcs/dropout_impl.cu.h
+58 −0 paddle/phi/kernels/funcs/rng_launch_config.h
+24 −11 paddle/phi/kernels/fusion/gpu/fused_dropout_add_utils.h
+10 −4 paddle/phi/kernels/fusion/xpu/fused_rope_utils.h
+2 −1 paddle/phi/kernels/gpu/interpolate_grad_kernel.cu
+2 −1 paddle/phi/kernels/gpu/interpolate_kernel.cu
+3 −2 paddle/phi/kernels/gpu/layer_norm_kernel.cu
+144 −10 paddle/phi/kernels/gpu/rms_norm_cuda_kernel.h
+1 −1 paddle/phi/kernels/stride/matmul_stride_kernel.cu
+15 −5 python/paddle/distributed/fleet/meta_optimizers/muon_sharding_optimizer.py
+129 −31 python/paddle/distributed/fleet/meta_parallel/pipeline_parallel.py
+21 −4 python/paddle/distributed/fleet/meta_parallel/pp_utils/p2p_communication.py
+33 −84 python/paddle/distributed/fleet/recompute/recompute.py
+2 −6 python/paddle/distributed/fleet/recompute/recompute_hybrid.py
+72 −55 python/paddle/optimizer/muon.py
+66 −0 python/paddle/tensor/ops.py
+87 −3 test/collective/fleet/hybrid_parallel_sharding_muon_model.py
+46 −0 test/collective/fleet/test_parallel_dygraph_muon.py
+700 −0 test/legacy_test/test_pylayer_clear_dataptr.py
+48 −453 test/legacy_test/test_recompute_with_tuple_input.py
29 changes: 29 additions & 0 deletions backends/metax_gpu/common/flags_declare.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -116,6 +116,35 @@ PHI_DEFINE_EXPORTED_bool(use_fast_math,
false,
"Whether to use fast math GPU functions.");

/**
* GPU RNG related FLAG
* Name: FLAGS_deterministic_rng
* Since Version: 3.4
* Value Range: bool, default=false
* Example: paddle.set_flags({'FLAGS_deterministic_rng': True})
* Note: Fix RNG kernel launch config so same seed gives same results
* across GPU types.
*/
PHI_DEFINE_EXPORTED_bool(
deterministic_rng,
false,
"Enable cross-device RNG consistency by fixing GPU kernel launch "
"configuration. When true, RNG kernels use a fixed grid/block size "
"so that the same seed produces identical results across GPU types.");
/**
* GPU RNG related FLAG
* Name: FLAGS_deterministic_rng_grid
* Since Version: 3.4
* Value Range: int32, default=1024
* Example: paddle.set_flags({'FLAGS_deterministic_rng_grid': 4096})
* Note: Grid size cap used when FLAGS_deterministic_rng is enabled.
* Cross-device consistency requires the same value on all devices.
*/
PHI_DEFINE_EXPORTED_int32(
deterministic_rng_grid,
1024,
"Grid size cap when FLAGS_deterministic_rng is enabled.");

#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
/**
* FlashAttention related FLAG
Expand Down
Loading