Skip to content
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

Jump to bottom

Remove async XLA_FLAGS from A3 configs. #645

Merged
merged 2 commits into from
May 15, 2024
Merged

Remove async XLA_FLAGS from A3 configs. #645

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e84114d
Remove async XLA_FLAGS from A3 configs.
reedwm May 10, 2024
2d2c3c4
Merge branch 'main' into xla_flags_fix
reedwm May 15, 2024
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
Next Next commit
Remove async XLA_FLAGS from A3 configs. 
XLA PR openxla/xla#11422 removed some XLA flags relating to async collectives. This caused the A3 configs to fail to run, so this change removes such flags from the A3 configs. The flags removed are:

--xla_gpu_enable_async_all_gather=true
--xla_gpu_enable_async_reduce_scatter=true
--xla_gpu_enable_async_all_reduce=true

Such flags had no impact before the XLA PR as the async collectives were already enabled by default.
  • Loading branch information
@reedwm
reedwm committed May 10, 2024
commit e84114dfe449de7ec4516c13ec40b9e8fadc16a3
4 changes: 2 additions & 2 deletions MaxText/configs/a3/llama_2_7b/16vm.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,9 +25,9 @@ export CUDA_DEVICE_MAX_CONNECTIONS=1
export NVTE_FUSED_ATTN=1
export NCCL_DEBUG=VERSION
export XLA_FLAGS="--xla_dump_to=$OUTPUT_PATH/$RUN_NAME/HLO_dumps/
--xla_gpu_enable_latency_hiding_scheduler=true --xla_gpu_enable_async_all_gather=true --xla_gpu_enable_async_reduce_scatter=true
--xla_gpu_enable_latency_hiding_scheduler=true
--xla_gpu_enable_triton_gemm=false --xla_gpu_simplify_all_fp_conversions --xla_gpu_graph_level=0
--xla_gpu_enable_async_all_reduce=true --xla_gpu_enable_highest_priority_async_stream=true
--xla_gpu_enable_highest_priority_async_stream=true
--xla_gpu_all_reduce_combine_threshold_bytes=1073741824 --xla_gpu_all_gather_combine_threshold_bytes=134217728
--xla_gpu_reduce_scatter_combine_threshold_bytes=134217728 --xla_gpu_enable_pipelined_all_gather=true
--xla_gpu_enable_pipelined_reduce_scatter=true --xla_gpu_enable_pipelined_all_reduce=true
Expand Down
6 changes: 3 additions & 3 deletions MaxText/configs/a3/llama_2_7b/1vm.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,9 +25,9 @@ export CUDA_DEVICE_MAX_CONNECTIONS=1
export NVTE_FUSED_ATTN=1
export NCCL_DEBUG=VERSION
export XLA_FLAGS="--xla_dump_to=$OUTPUT_PATH/$RUN_NAME/HLO_dumps/
--xla_gpu_enable_latency_hiding_scheduler=true --xla_gpu_enable_async_all_gather=true
--xla_gpu_enable_async_reduce_scatter=true --xla_gpu_enable_triton_gemm=false --xla_gpu_simplify_all_fp_conversions
--xla_gpu_graph_level=0 --xla_gpu_enable_async_all_reduce=true --xla_gpu_enable_highest_priority_async_stream=true
--xla_gpu_enable_latency_hiding_scheduler=true
--xla_gpu_enable_triton_gemm=false --xla_gpu_simplify_all_fp_conversions
--xla_gpu_graph_level=0 --xla_gpu_enable_highest_priority_async_stream=true
--xla_gpu_all_reduce_combine_threshold_bytes=134217728 --xla_gpu_all_gather_combine_threshold_bytes=134217728
--xla_gpu_reduce_scatter_combine_threshold_bytes=67108864 --xla_gpu_enable_pipelined_all_gather=true
--xla_gpu_enable_pipelined_reduce_scatter=true --xla_gpu_enable_pipelined_all_reduce=true
Expand Down
4 changes: 2 additions & 2 deletions MaxText/configs/a3/llama_2_7b/2vm.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,9 +25,9 @@ export CUDA_DEVICE_MAX_CONNECTIONS=1
export NVTE_FUSED_ATTN=1
export NCCL_DEBUG=VERSION
export XLA_FLAGS="--xla_dump_to=$OUTPUT_PATH/$RUN_NAME/HLO_dumps/
--xla_gpu_enable_latency_hiding_scheduler=true --xla_gpu_enable_async_all_gather=true --xla_gpu_enable_async_reduce_scatter=true
--xla_gpu_enable_latency_hiding_scheduler=true
--xla_gpu_enable_triton_gemm=false --xla_gpu_simplify_all_fp_conversions --xla_gpu_graph_level=0
--xla_gpu_enable_async_all_reduce=true --xla_gpu_enable_highest_priority_async_stream=true
--xla_gpu_enable_highest_priority_async_stream=true
--xla_gpu_all_reduce_combine_threshold_bytes=67108864 --xla_gpu_all_gather_combine_threshold_bytes=134217728
--xla_gpu_reduce_scatter_combine_threshold_bytes=67108864 --xla_gpu_enable_pipelined_all_gather=true
--xla_gpu_enable_pipelined_reduce_scatter=true --xla_gpu_enable_pipelined_all_reduce=true
Expand Down
6 changes: 3 additions & 3 deletions MaxText/configs/a3/llama_2_7b/4vm.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,9 +25,9 @@ export CUDA_DEVICE_MAX_CONNECTIONS=1
export NVTE_FUSED_ATTN=1
export NCCL_DEBUG=VERSION
export XLA_FLAGS="--xla_dump_to=$OUTPUT_PATH/$RUN_NAME/HLO_dumps/
--xla_gpu_enable_latency_hiding_scheduler=true --xla_gpu_enable_async_all_gather=true
--xla_gpu_enable_async_reduce_scatter=true --xla_gpu_enable_triton_gemm=false --xla_gpu_simplify_all_fp_conversions
--xla_gpu_graph_level=0 --xla_gpu_enable_async_all_reduce=true --xla_gpu_enable_highest_priority_async_stream=true
--xla_gpu_enable_latency_hiding_scheduler=true
--xla_gpu_enable_triton_gemm=false --xla_gpu_simplify_all_fp_conversions
--xla_gpu_graph_level=0 --xla_gpu_enable_highest_priority_async_stream=true
--xla_gpu_all_reduce_combine_threshold_bytes=536870912 --xla_gpu_all_gather_combine_threshold_bytes=134217728
--xla_gpu_reduce_scatter_combine_threshold_bytes=67108864 --xla_gpu_enable_pipelined_all_gather=true
--xla_gpu_enable_pipelined_reduce_scatter=true --xla_gpu_enable_pipelined_all_reduce=true
Expand Down
4 changes: 2 additions & 2 deletions MaxText/configs/a3/llama_2_7b/8vm.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,9 +25,9 @@ export CUDA_DEVICE_MAX_CONNECTIONS=1
export NVTE_FUSED_ATTN=1
export NCCL_DEBUG=VERSION
export XLA_FLAGS="--xla_dump_to=$OUTPUT_PATH/$RUN_NAME/HLO_dumps/
--xla_gpu_enable_latency_hiding_scheduler=true --xla_gpu_enable_async_all_gather=true --xla_gpu_enable_async_reduce_scatter=true
--xla_gpu_enable_latency_hiding_scheduler=true
--xla_gpu_enable_triton_gemm=false --xla_gpu_simplify_all_fp_conversions --xla_gpu_graph_level=0
--xla_gpu_enable_async_all_reduce=true --xla_gpu_enable_highest_priority_async_stream=true
--xla_gpu_enable_highest_priority_async_stream=true
--xla_gpu_all_reduce_combine_threshold_bytes=1073741824 --xla_gpu_all_gather_combine_threshold_bytes=134217728
--xla_gpu_reduce_scatter_combine_threshold_bytes=67108864 --xla_gpu_enable_pipelined_all_gather=true
--xla_gpu_enable_pipelined_reduce_scatter=true --xla_gpu_enable_pipelined_all_reduce=true
Expand Down
Loading