Skip to content

GitLab

Unverified Commit 52450b5f authored by one's avatar one Committed by GitHub
Browse files

Config: Update dist-inference params (#17)

parent b015ceeb
Hide whitespace changes
Inline Side-by-side
Showing with 32 additions and 42 deletions
superbench/config/hygon_bw1000.yaml
View file @ 52450b5f
...@@ -507,24 +507,19 @@ superbench: ...@@ -507,24 +507,19 @@ superbench:
ib_dev: '"$(case $LOCAL_RANK in 0) echo mlx5_1 ;; 1) echo mlx5_2 ;; 2) echo mlx5_3 ;; 3) echo mlx5_4 ;; 4) echo mlx5_7 ;; 5) echo mlx5_8 ;; 6) echo mlx5_9 ;; 7) echo mlx5_10 ;; esac)"' ib_dev: '"$(case $LOCAL_RANK in 0) echo mlx5_1 ;; 1) echo mlx5_2 ;; 2) echo mlx5_3 ;; 3) echo mlx5_4 ;; 4) echo mlx5_7 ;; 5) echo mlx5_8 ;; 6) echo mlx5_9 ;; 7) echo mlx5_10 ;; esac)"'
numa_dev: '"$(case $LOCAL_RANK in 0) echo 3 ;; 1) echo 1 ;; 2) echo 1 ;; 3) echo 0 ;; 4) echo 7 ;; 5) echo 5 ;; 6) echo 5 ;; 7) echo 4 ;; esac)"' numa_dev: '"$(case $LOCAL_RANK in 0) echo 3 ;; 1) echo 1 ;; 2) echo 1 ;; 3) echo 0 ;; 4) echo 7 ;; 5) echo 5 ;; 6) echo 5 ;; 7) echo 4 ;; esac)"'
bidirectional: false bidirectional: false
# dist-inference: dist-inference:
# modes: modes:
# - name: mpi - name: mpi
# proc_num: 8 proc_num: 8
# node_num: 1 node_num: 1
# mca: frameworks:
# pml: ob1 - pytorch
# btl: ^openib parameters:
# btl_tcp_if_exclude: lo,docker0 num_layers: 20
# coll_hcoll_enable: 0 num_warmup: 20
# frameworks: num_steps: 100
# - pytorch use_cuda_graph: true
# parameters: precision: float16
# num_layers: 50 hidden_size: 4096
# num_warmup: 20 input_size: 4096
# num_steps: 100 batch_size: 1024
# use_cuda_graph: true
# precision: float16
# hidden_size: 128
# input_size: 128
# batch_size: 1024
superbench/config/hygon_bw150.yaml
View file @ 52450b5f
...@@ -396,24 +396,19 @@ superbench: ...@@ -396,24 +396,19 @@ superbench:
parameters: parameters:
mem_type: [htod, dtoh, dtod, one_to_all, all_to_one, all_to_all] mem_type: [htod, dtoh, dtod, one_to_all, all_to_one, all_to_all]
copy_type: [sm, dma] copy_type: [sm, dma]
# dist-inference: dist-inference:
# modes: modes:
# - name: mpi - name: mpi
# proc_num: 8 proc_num: 8
# node_num: 1 node_num: 1
# mca: frameworks:
# pml: ob1 - pytorch
# btl: ^openib parameters:
# btl_tcp_if_exclude: lo,docker0 num_layers: 20
# coll_hcoll_enable: 0 num_warmup: 20
# frameworks: num_steps: 100
# - pytorch use_cuda_graph: true
# parameters: precision: float16
# num_layers: 50 hidden_size: 4096
# num_warmup: 20 input_size: 4096
# num_steps: 100 batch_size: 1024
# use_cuda_graph: true
# precision: float16
# hidden_size: 128
# input_size: 128
# batch_size: 1024
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment