diff --git a/examples/mcore/qwen3/pretrain_qwen3_32b_4K_ptd.sh b/examples/mcore/qwen3/pretrain_qwen3_32b_4K_ptd.sh
index f58f0a23bafd9a63e9160ca7d8a6cb9eb5565b01..d83c0735da79df330a3f0c90887459c0041c62ee 100644
--- a/examples/mcore/qwen3/pretrain_qwen3_32b_4K_ptd.sh
+++ b/examples/mcore/qwen3/pretrain_qwen3_32b_4K_ptd.sh
@@ -1,6 +1,5 @@
 export HCCL_CONNECT_TIMEOUT=1800
 export CUDA_DEVICE_MAX_CONNECTIONS=1
-export NPU_ASD_ENABLE=0
 
 NPUS_PER_NODE=8
 MASTER_ADDR=localhost
@@ -17,14 +16,12 @@ CKPT_LOAD_DIR="your model ckpt path"
 
 TP=8
 PP=2
-CP=1
-VPP=4
+VPP=8
 
-MBS=1
-GBS=128
+MBS=2
+GBS=256
 SEQ_LENGTH=4096
 TRAIN_ITERS=2000
-CP_TYPE='ulysses_cp_algo'
 
 DISTRIBUTED_ARGS="
     --nproc_per_node $NPUS_PER_NODE \
@@ -47,7 +44,7 @@ OPTIMIZE_ARGS="
 "
 
 TRAIN_ARGS="
-    --micro-batch-size 1 \
+    --micro-batch-size ${MBS} \
     --global-batch-size ${GBS} \
     --lr 1.25e-6 \
     --lr-decay-style cosine \
@@ -72,12 +69,11 @@ MODEL_PARALLEL_ARGS="
     --tensor-model-parallel-size ${TP} \
     --pipeline-model-parallel-size ${PP} \
     --num-layers-per-virtual-pipeline-stage ${VPP} \
-    --context-parallel-size ${CP} \
-    --context-parallel-algo ${CP_TYPE} \
 "
 
 GPT_ARGS="
     --use-mcore-models \
+    --sequence-parallel \
     --spec mindspeed_llm.tasks.models.spec.qwen3_spec layer_spec \
     --kv-channels 128 \
     --qk-layernorm \
@@ -99,7 +95,8 @@ GPT_ARGS="
     --attention-softmax-in-fp32 \
     --no-gradient-accumulation-fusion \
     --group-query-attention \
-    --num-query-groups 8
+    --num-query-groups 8 \
+    --reset-position-ids
 "
 
 DATA_ARGS="
@@ -113,7 +110,8 @@ OUTPUT_ARGS="
     --eval-interval ${TRAIN_ITERS} \
     --eval-iters 0 \
     --no-load-optim \
-    --no-load-rng
+    --no-load-rng \
+    --log-throughput
 "
 
 torchrun $DISTRIBUTED_ARGS pretrain_gpt.py \
@@ -126,4 +124,4 @@ torchrun $DISTRIBUTED_ARGS pretrain_gpt.py \
     --distributed-backend nccl \
     --load ${CKPT_LOAD_DIR} \
     --save ${CKPT_SAVE_DIR} \
-    | tee logs/train_mcore_qwen3_32b.log
\ No newline at end of file
+    | tee logs/train_mcore_qwen3_32b.log
diff --git a/examples/mcore/qwen3/pretrain_qwen3_32b_4K_ptd_A3.sh b/examples/mcore/qwen3/pretrain_qwen3_32b_4K_ptd_A3.sh
new file mode 100644
index 0000000000000000000000000000000000000000..1cfd82b548f19a962b0867907f4e8efd07f4b9aa
--- /dev/null
+++ b/examples/mcore/qwen3/pretrain_qwen3_32b_4K_ptd_A3.sh
@@ -0,0 +1,129 @@
+export HCCL_CONNECT_TIMEOUT=1800
+export CUDA_DEVICE_MAX_CONNECTIONS=1
+
+NPUS_PER_NODE=16
+MASTER_ADDR=localhost
+MASTER_PORT=6000
+NNODES=1
+NODE_RANK=0
+WORLD_SIZE=$(($NPUS_PER_NODE*$NNODES))
+
+# please fill these path configurations
+CKPT_SAVE_DIR="your model save ckpt path"
+DATA_PATH="your data path"
+TOKENIZER_PATH="your tokenizer path"
+CKPT_LOAD_DIR="your model ckpt path"
+
+TP=8
+PP=2
+VPP=4
+
+MBS=2
+GBS=256
+SEQ_LENGTH=4096
+TRAIN_ITERS=2000
+
+DISTRIBUTED_ARGS="
+    --nproc_per_node $NPUS_PER_NODE \
+    --nnodes $NNODES \
+    --node_rank $NODE_RANK \
+    --master_addr $MASTER_ADDR \
+    --master_port $MASTER_PORT
+"
+
+OPTIMIZE_ARGS="
+    --use-flash-attn \
+    --use-fused-rotary-pos-emb \
+    --use-rotary-position-embeddings \
+    --use-fused-swiglu \
+    --use-fused-rmsnorm \
+    --no-masked-softmax-fusion \
+    --use-distributed-optimizer \
+    --overlap-grad-reduce \
+    --overlap-param-gather
+"
+
+TRAIN_ARGS="
+    --micro-batch-size ${MBS} \
+    --global-batch-size ${GBS} \
+    --lr 1.25e-6 \
+    --lr-decay-style cosine \
+    --min-lr 1.25e-7 \
+    --weight-decay 1e-1 \
+    --lr-warmup-fraction 0.01 \
+    --attention-dropout 0.0 \
+    --init-method-std 0.01 \
+    --hidden-dropout 0.0 \
+    --clip-grad 1.0 \
+    --adam-beta1 0.9 \
+    --adam-beta2 0.95 \
+    --initial-loss-scale 4096 \
+    --seed 42 \
+    --bf16 \
+    --train-iters ${TRAIN_ITERS} \
+    --seq-length ${SEQ_LENGTH} \
+    --no-shared-storage
+"
+
+MODEL_PARALLEL_ARGS="
+    --tensor-model-parallel-size ${TP} \
+    --pipeline-model-parallel-size ${PP} \
+    --num-layers-per-virtual-pipeline-stage ${VPP} \
+"
+
+GPT_ARGS="
+    --use-mcore-models \
+    --sequence-parallel \
+    --spec mindspeed_llm.tasks.models.spec.qwen3_spec layer_spec \
+    --kv-channels 128 \
+    --qk-layernorm \
+    --tokenizer-name-or-path ${TOKENIZER_PATH} \
+    --max-position-embeddings ${SEQ_LENGTH} \
+    --num-layers 64 \
+    --hidden-size 5120 \
+    --ffn-hidden-size 25600 \
+    --num-attention-heads 64 \
+    --tokenizer-type PretrainedFromHF \
+    --make-vocab-size-divisible-by 1 \
+    --padded-vocab-size 151936 \
+    --rotary-base 1000000 \
+    --untie-embeddings-and-output-weights \
+    --disable-bias-linear \
+    --position-embedding-type rope \
+    --normalization RMSNorm \
+    --swiglu \
+    --attention-softmax-in-fp32 \
+    --no-gradient-accumulation-fusion \
+    --group-query-attention \
+    --num-query-groups 8 \
+    --reset-position-ids \
+    --use-ascend-coc \
+    --coc-fused-kernel
+"
+
+DATA_ARGS="
+    --data-path $DATA_PATH \
+    --split 100,0,0
+"
+
+OUTPUT_ARGS="
+    --log-interval 1 \
+    --save-interval ${TRAIN_ITERS} \
+    --eval-interval ${TRAIN_ITERS} \
+    --eval-iters 0 \
+    --no-load-optim \
+    --no-load-rng \
+    --log-throughput
+"
+
+torchrun $DISTRIBUTED_ARGS pretrain_gpt.py \
+    $GPT_ARGS \
+    $DATA_ARGS \
+    $OUTPUT_ARGS \
+    $OPTIMIZE_ARGS \
+    $TRAIN_ARGS \
+    $MODEL_PARALLEL_ARGS \
+    --distributed-backend nccl \
+    --load ${CKPT_LOAD_DIR} \
+    --save ${CKPT_SAVE_DIR} \
+    | tee logs/train_mcore_qwen3_32b_4k_A3.log