qwen 7b benchmark

wtmlon · Jan 25, 2024 · 8b6d037 · 8b6d037
1 parent 86b00d7
commit 8b6d037
Show file tree

Hide file tree

Showing 3 changed files with 6 additions and 6 deletions.
diff --git a/...bf16_MP4-PP2-vpp1-mbs1-acc16-recompute.sh → ...bf16_MP4-PP2-vpp1-mbs1-acc16-recompute.sh b/...bf16_MP4-PP2-vpp1-mbs1-acc16-recompute.sh → ...bf16_MP4-PP2-vpp1-mbs1-acc16-recompute.sh
@@ -13,7 +13,7 @@
 # limitations under the License.
 
 
-param="model_name_or_path=qwen/qwen-14b "
+param="model_name_or_path=qwen/qwen-7b "
 param+="per_device_train_batch_size=1 "
 param+="data_parallel_degree=1 "
 param+="tensor_parallel_degree=4 "
@@ -26,7 +26,7 @@ param+="recompute=1 "
 param+="run_mode=MP4-PP2-vpp1-mbs1-acc16-recompute "
 param+="device_num=N1C8 "
 param+="global_batch_size=16 "
-param+="model_item=qwen-qwen-14b_seqlen2048_pretrain "
+param+="model_item=qwen-qwen-7b_seqlen2048_pretrain "
 param+="max_steps=150 "
 param+="gradient_accumulation_steps=16 "
 param+="pp_recompute_interval=1 "

diff --git a/...bs16_bf16_MP8-PP1-mbs1-acc16-recompute.sh → ...bs16_bf16_MP8-PP1-mbs1-acc16-recompute.sh b/...bs16_bf16_MP8-PP1-mbs1-acc16-recompute.sh → ...bs16_bf16_MP8-PP1-mbs1-acc16-recompute.sh
@@ -13,7 +13,7 @@
 # limitations under the License.
 
 
-param="model_name_or_path=qwen/qwen-14b "
+param="model_name_or_path=qwen/qwen-7b "
 param+="per_device_train_batch_size=1 "
 param+="data_parallel_degree=1 "
 param+="tensor_parallel_degree=8 "
@@ -26,7 +26,7 @@ param+="recompute=1 "
 param+="run_mode=MP8-PP1-mbs1-acc16 "
 param+="device_num=N1C8 "
 param+="global_batch_size=16 "
-param+="model_item=qwen-qwen-14b_seqlen2048_pretrain "
+param+="model_item=qwen-qwen-7b_seqlen2048_pretrain "
 param+="max_steps=150 "
 param+="gradient_accumulation_steps=16 "
 param+="pp_recompute_interval=1 "

diff --git a/tests/test_tipc/dygraph/hybrid_parallelism/qwen/benchmark_common/run_benchmark.sh b/tests/test_tipc/dygraph/hybrid_parallelism/qwen/benchmark_common/run_benchmark.sh
@@ -17,7 +17,7 @@
 # Test training benchmark for a model.
 # Usage：bash benchmark/run_benchmark.sh ${model_name_or_path} ${per_device_train_batch_size} ${tensor_parallel_degree} ${pipeline_parallel_degree} ${virtual_pp_degree} ${sequence_parallel} ${sharding_parallel_degree} ${sharding} ${recompute} ${run_mode} ${device_num}
 function _set_params(){
-    model_name_or_path=${model_name_or_path:-"qwen/qwen-14b"}
+    model_name_or_path=${model_name_or_path:-"qwen/qwen-7b"}
     per_device_train_batch_size=${per_device_train_batch_size:-1}
     tensor_parallel_degree=${tensor_parallel_degree:-1}
     data_parallel_degree=${data_parallel_degree:-1}
@@ -30,7 +30,7 @@ function _set_params(){
     run_mode=${run_mode:-"DP1-MP1-PP4-mbs1-acc8-recompute"}
     device_num=${device_num:-"N1C8"}
     global_batch_size=${global_batch_size:-16}
-    model_item=${model_item:-"qwen-qwen-14b_seqlen2048_pretrain"}
+    model_item=${model_item:-"qwen-qwen-7b_seqlen2048_pretrain"}
     max_steps=${max_steps:-150}
     gradient_accumulation_steps=${gradient_accumulation_steps:-8}
     pp_recompute_interval=${pp_recompute_interval:-1}