From 8cec24d36628671d20acdccc0448fe1c28f234fe Mon Sep 17 00:00:00 2001 From: Terry Kong Date: Tue, 8 Oct 2024 12:53:16 -0700 Subject: [PATCH] chore: remove deprecated flags from model xla flag configs Signed-off-by: Terry Kong --- rosetta/rosetta/projects/maxtext/xla_flags/llama2-7b-1N8G.env | 2 -- rosetta/rosetta/projects/pax/xla_flags/common.env | 2 -- rosetta/rosetta/projects/pax/xla_flags/gpt-126m.env | 2 -- rosetta/rosetta/projects/pax/xla_flags/grok-proxy.env | 2 -- 4 files changed, 8 deletions(-) diff --git a/rosetta/rosetta/projects/maxtext/xla_flags/llama2-7b-1N8G.env b/rosetta/rosetta/projects/maxtext/xla_flags/llama2-7b-1N8G.env index d999f5b5e..f6255c130 100644 --- a/rosetta/rosetta/projects/maxtext/xla_flags/llama2-7b-1N8G.env +++ b/rosetta/rosetta/projects/maxtext/xla_flags/llama2-7b-1N8G.env @@ -6,7 +6,6 @@ export XLA_FLAGS="\ --xla_gpu_enable_latency_hiding_scheduler=true \ --xla_gpu_enable_triton_gemm=false \ --xla_gpu_graph_level=0 \ - --xla_gpu_enable_highest_priority_async_stream=true \ --xla_gpu_all_reduce_combine_threshold_bytes=${THRESHOLD_BYTES} \ --xla_gpu_all_gather_combine_threshold_bytes=$((THRESHOLD_BYTES/(NUM_NODES*NUM_GPUS))) \ --xla_gpu_reduce_scatter_combine_threshold_bytes=$((THRESHOLD_BYTES/(NUM_NODES*NUM_GPUS*2))) \ @@ -14,7 +13,6 @@ export XLA_FLAGS="\ --xla_gpu_enable_pipelined_reduce_scatter=true \ --xla_gpu_enable_pipelined_all_reduce=true \ --xla_gpu_enable_while_loop_double_buffering=true \ - --xla_gpu_enable_triton_softmax_fusion=false \ --xla_gpu_enable_all_gather_combine_by_dim=false \ --xla_gpu_enable_reduce_scatter_combine_by_dim=false \ --xla_disable_hlo_passes=rematerialization \ diff --git a/rosetta/rosetta/projects/pax/xla_flags/common.env b/rosetta/rosetta/projects/pax/xla_flags/common.env index 26c819143..50e3c3e3a 100644 --- a/rosetta/rosetta/projects/pax/xla_flags/common.env +++ b/rosetta/rosetta/projects/pax/xla_flags/common.env @@ -3,8 +3,6 @@ THRESHOLD_BYTES=51200 export XLA_FLAGS="\ --xla_gpu_enable_latency_hiding_scheduler=true \ --xla_allow_excess_precision \ - --xla_gpu_enable_highest_priority_async_stream=true \ - --xla_gpu_enable_triton_softmax_fusion=false \ --xla_gpu_all_reduce_combine_threshold_bytes=${THRESHOLD_BYTES} \ --xla_gpu_graph_level=0 \ " diff --git a/rosetta/rosetta/projects/pax/xla_flags/gpt-126m.env b/rosetta/rosetta/projects/pax/xla_flags/gpt-126m.env index e5b97b466..2eb3aca6e 100644 --- a/rosetta/rosetta/projects/pax/xla_flags/gpt-126m.env +++ b/rosetta/rosetta/projects/pax/xla_flags/gpt-126m.env @@ -3,8 +3,6 @@ THRESHOLD_BYTES=33554432 export XLA_FLAGS="\ --xla_gpu_enable_latency_hiding_scheduler=true \ --xla_allow_excess_precision \ - --xla_gpu_enable_highest_priority_async_stream=true \ - --xla_gpu_enable_triton_softmax_fusion=false \ --xla_gpu_all_reduce_combine_threshold_bytes=${THRESHOLD_BYTES} \ --xla_gpu_graph_level=0 \ --xla_gpu_enable_cudnn_fmha=false \ diff --git a/rosetta/rosetta/projects/pax/xla_flags/grok-proxy.env b/rosetta/rosetta/projects/pax/xla_flags/grok-proxy.env index e48b76dcf..48166ed2f 100644 --- a/rosetta/rosetta/projects/pax/xla_flags/grok-proxy.env +++ b/rosetta/rosetta/projects/pax/xla_flags/grok-proxy.env @@ -5,8 +5,6 @@ REDUCE_SCATTER_THRESHOLD_BYTES=402653184 export XLA_FLAGS="\ --xla_gpu_enable_latency_hiding_scheduler=true \ --xla_allow_excess_precision \ - --xla_gpu_enable_highest_priority_async_stream=true \ - --xla_gpu_enable_triton_softmax_fusion=false \ --xla_gpu_all_reduce_combine_threshold_bytes=${ALL_REDUCE_THRESHOLD_BYTES} \ --xla_gpu_graph_level=0 \ --xla_gpu_all_gather_combine_threshold_bytes=${ALL_GATHER_THRESHOLD_BYTES} \