已合并
[mindspore][sh] add --master_addr and --node_rank for qwen3 tune sh #3666
wanglijun55创建于 2025年11月11日
[mindspore][sh] add --master_addr and --node_rank for qwen3 tune sh #3666
已合并
共 5 个文件变更+25-15
| @@ -25,12 +25,14 @@ SEQ_LENGTH=4096 | |||
| 25 | TRAIN_ITERS=2000 | 25 | TRAIN_ITERS=2000 |
| 26 | 26 | ||
| 27 | DISTRIBUTED_ARGS=" | 27 | DISTRIBUTED_ARGS=" |
| 28 | - --worker_num $WORLD_SIZE \ | ||
| 29 | --local_worker_num $NPUS_PER_NODE \ | 28 | --local_worker_num $NPUS_PER_NODE \ |
| 29 | + --worker_num $WORLD_SIZE \ | ||
| 30 | + --node_rank $NODE_RANK \ | ||
| 31 | + --master_addr $MASTER_ADDR \ | ||
| 32 | + --master_port $MASTER_PORT \ | ||
| 30 | --log_dir="msrun_log" \ | 33 | --log_dir="msrun_log" \ |
| 31 | --join=True \ | 34 | --join=True \ |
| 32 | - --cluster_time_out=300 \ | 35 | + --cluster_time_out=300 |
| 33 | - --master_port $MASTER_PORT | ||
| 34 | " | 36 | " |
| 35 | 37 | ||
| 36 | GPT_ARGS=" | 38 | GPT_ARGS=" |
| @@ -25,12 +25,14 @@ SEQ_LENGTH=4096 | |||
| 25 | TRAIN_ITERS=2000 | 25 | TRAIN_ITERS=2000 |
| 26 | 26 | ||
| 27 | DISTRIBUTED_ARGS=" | 27 | DISTRIBUTED_ARGS=" |
| 28 | - --worker_num $WORLD_SIZE \ | ||
| 29 | --local_worker_num $NPUS_PER_NODE \ | 28 | --local_worker_num $NPUS_PER_NODE \ |
| 29 | + --worker_num $WORLD_SIZE \ | ||
| 30 | + --node_rank $NODE_RANK \ | ||
| 31 | + --master_addr $MASTER_ADDR \ | ||
| 32 | + --master_port $MASTER_PORT \ | ||
| 30 | --log_dir="msrun_log" \ | 33 | --log_dir="msrun_log" \ |
| 31 | --join=True \ | 34 | --join=True \ |
| 32 | - --cluster_time_out=300 \ | 35 | + --cluster_time_out=300 |
| 33 | - --master_port $MASTER_PORT | ||
| 34 | " | 36 | " |
| 35 | 37 | ||
| 36 | GPT_ARGS=" | 38 | GPT_ARGS=" |
| @@ -21,12 +21,14 @@ SEQ_LENGTH=4096 | |||
| 21 | TRAIN_ITERS=2000 | 21 | TRAIN_ITERS=2000 |
| 22 | 22 | ||
| 23 | DISTRIBUTED_ARGS=" | 23 | DISTRIBUTED_ARGS=" |
| 24 | - --worker_num $WORLD_SIZE \ | ||
| 25 | --local_worker_num $NPUS_PER_NODE \ | 24 | --local_worker_num $NPUS_PER_NODE \ |
| 25 | + --worker_num $WORLD_SIZE \ | ||
| 26 | + --node_rank $NODE_RANK \ | ||
| 27 | + --master_addr $MASTER_ADDR \ | ||
| 28 | + --master_port $MASTER_PORT \ | ||
| 26 | --log_dir="msrun_log" \ | 29 | --log_dir="msrun_log" \ |
| 27 | --join=True \ | 30 | --join=True \ |
| 28 | - --cluster_time_out=300 \ | 31 | + --cluster_time_out=300 |
| 29 | - --master_port $MASTER_PORT | ||
| 30 | " | 32 | " |
| 31 | 33 | ||
| 32 | OPTIMIZE_ARGS=" | 34 | OPTIMIZE_ARGS=" |
| @@ -21,12 +21,14 @@ MBS=1 | |||
| 21 | GBS=16 | 21 | GBS=16 |
| 22 | 22 | ||
| 23 | DISTRIBUTED_ARGS=" | 23 | DISTRIBUTED_ARGS=" |
| 24 | - --worker_num $WORLD_SIZE \ | ||
| 25 | --local_worker_num $NPUS_PER_NODE \ | 24 | --local_worker_num $NPUS_PER_NODE \ |
| 25 | + --worker_num $WORLD_SIZE \ | ||
| 26 | + --node_rank $NODE_RANK \ | ||
| 27 | + --master_addr $MASTER_ADDR \ | ||
| 28 | + --master_port $MASTER_PORT \ | ||
| 26 | --log_dir="msrun_log" \ | 29 | --log_dir="msrun_log" \ |
| 27 | --join=True \ | 30 | --join=True \ |
| 28 | - --cluster_time_out=300 \ | 31 | + --cluster_time_out=300 |
| 29 | - --master_port $MASTER_PORT | ||
| 30 | " | 32 | " |
| 31 | 33 | ||
| 32 | GPT_ARGS=" | 34 | GPT_ARGS=" |
| @@ -21,12 +21,14 @@ MBS=1 | |||
| 21 | GBS=16 | 21 | GBS=16 |
| 22 | 22 | ||
| 23 | DISTRIBUTED_ARGS=" | 23 | DISTRIBUTED_ARGS=" |
| 24 | - --worker_num $WORLD_SIZE \ | ||
| 25 | --local_worker_num $NPUS_PER_NODE \ | 24 | --local_worker_num $NPUS_PER_NODE \ |
| 25 | + --worker_num $WORLD_SIZE \ | ||
| 26 | + --node_rank $NODE_RANK \ | ||
| 27 | + --master_addr $MASTER_ADDR \ | ||
| 28 | + --master_port $MASTER_PORT \ | ||
| 26 | --log_dir="msrun_log" \ | 29 | --log_dir="msrun_log" \ |
| 27 | --join=True \ | 30 | --join=True \ |
| 28 | - --cluster_time_out=300 \ | 31 | + --cluster_time_out=300 |
| 29 | - --master_port $MASTER_PORT | ||
| 30 | " | 32 | " |
| 31 | 33 | ||
| 32 | GPT_ARGS=" | 34 | GPT_ARGS=" |