From d0f2e1a81a482afe02222d00eb65fc102e2ed80d Mon Sep 17 00:00:00 2001 From: qu_yueze Date: Thu, 26 Jun 2025 19:02:24 +0800 Subject: [PATCH] fix scripts of qwen3 --- examples/mcore/qwen3/evaluate_qwen3_0point6b_ptd.sh | 2 +- examples/mcore/qwen3/evaluate_qwen3_14b_ptd.sh | 2 +- examples/mcore/qwen3/tune_qwen3_8b_4K_lora_ptd.sh | 2 +- .../mcore/qwen3_moe/pretrain_qwen3_235b_a22b_4k_A3_ptd.sh | 5 ++--- tests/0day/qwen3/qwen3-0.6b/evaluate_qwen3_0point6b_ptd.sh | 2 +- tests/0day/qwen3/qwen3-14b/evaluate_qwen3_14b_ptd.sh | 2 +- .../qwen3-235b-a22b/pretrain_qwen3_235b_a22b_4k_A3_ptd.sh | 2 +- tests/0day/qwen3/qwen3-32b/tune_qwen3_32b_4K_lora_ptd.sh | 2 +- tests/0day/qwen3/qwen3-8b/tune_qwen3_8b_4K_lora_ptd.sh | 2 +- tests/pipeline/llama2/llama2_tp1_pp8_patch_gloo_ptd.sh | 2 +- 10 files changed, 11 insertions(+), 12 deletions(-) diff --git a/examples/mcore/qwen3/evaluate_qwen3_0point6b_ptd.sh b/examples/mcore/qwen3/evaluate_qwen3_0point6b_ptd.sh index e41d2f13b..0148eb498 100644 --- a/examples/mcore/qwen3/evaluate_qwen3_0point6b_ptd.sh +++ b/examples/mcore/qwen3/evaluate_qwen3_0point6b_ptd.sh @@ -11,7 +11,7 @@ TASK="mmlu" # Change for multinode config MASTER_ADDR=localhost -MASTER_PORT=60014 +MASTER_PORT=6000 NNODES=1 NODE_RANK=0 NPUS_PER_NODE=8 diff --git a/examples/mcore/qwen3/evaluate_qwen3_14b_ptd.sh b/examples/mcore/qwen3/evaluate_qwen3_14b_ptd.sh index fc9174908..8d6be3f6d 100644 --- a/examples/mcore/qwen3/evaluate_qwen3_14b_ptd.sh +++ b/examples/mcore/qwen3/evaluate_qwen3_14b_ptd.sh @@ -11,7 +11,7 @@ TASK="mmlu" # Change for multinode config MASTER_ADDR=localhost -MASTER_PORT=60014 +MASTER_PORT=6000 NNODES=1 NODE_RANK=0 NPUS_PER_NODE=8 diff --git a/examples/mcore/qwen3/tune_qwen3_8b_4K_lora_ptd.sh b/examples/mcore/qwen3/tune_qwen3_8b_4K_lora_ptd.sh index 825addbd7..4093fea5b 100644 --- a/examples/mcore/qwen3/tune_qwen3_8b_4K_lora_ptd.sh +++ b/examples/mcore/qwen3/tune_qwen3_8b_4K_lora_ptd.sh @@ -4,7 +4,7 @@ export CUDA_DEVICE_MAX_CONNECTIONS=1 NPUS_PER_NODE=8 MASTER_ADDR=localhost -MASTER_PORT=60014 +MASTER_PORT=6000 NNODES=1 NODE_RANK=0 WORLD_SIZE=$(($NPUS_PER_NODE*$NNODES)) diff --git a/examples/mcore/qwen3_moe/pretrain_qwen3_235b_a22b_4k_A3_ptd.sh b/examples/mcore/qwen3_moe/pretrain_qwen3_235b_a22b_4k_A3_ptd.sh index 203c0f6a0..bedccb587 100644 --- a/examples/mcore/qwen3_moe/pretrain_qwen3_235b_a22b_4k_A3_ptd.sh +++ b/examples/mcore/qwen3_moe/pretrain_qwen3_235b_a22b_4k_A3_ptd.sh @@ -9,7 +9,7 @@ export TASK_QUEUE_ENABLE=2 NPUS_PER_NODE=16 MASTER_ADDR=localhost -MASTER_PORT=60014 +MASTER_PORT=6000 NNODES=16 NODE_RANK=0 WORLD_SIZE=$(($NPUS_PER_NODE*$NNODES)) @@ -20,7 +20,7 @@ DATA_PATH="your data path" TOKENIZER_PATH="your tokenizer path" CKPT_LOAD_DIR="your model ckpt path" -TP=2 +TP=1 PP=4 EP=32 CP=1 @@ -58,7 +58,6 @@ MOE_ARGS=" --moe-aux-loss-coeff 0.001 \ --moe-permutation-async-comm \ --moe-alltoall-overlap-comm \ - --moe-tp-extend-ep \ " OPTIMIZE_ARGS=" diff --git a/tests/0day/qwen3/qwen3-0.6b/evaluate_qwen3_0point6b_ptd.sh b/tests/0day/qwen3/qwen3-0.6b/evaluate_qwen3_0point6b_ptd.sh index 03c7a0dfe..0b7476e52 100644 --- a/tests/0day/qwen3/qwen3-0.6b/evaluate_qwen3_0point6b_ptd.sh +++ b/tests/0day/qwen3/qwen3-0.6b/evaluate_qwen3_0point6b_ptd.sh @@ -11,7 +11,7 @@ TASK="mmlu" # Change for multinode config MASTER_ADDR=localhost -MASTER_PORT=60014 +MASTER_PORT=6000 NNODES=1 NODE_RANK=0 NPUS_PER_NODE=8 diff --git a/tests/0day/qwen3/qwen3-14b/evaluate_qwen3_14b_ptd.sh b/tests/0day/qwen3/qwen3-14b/evaluate_qwen3_14b_ptd.sh index fc9174908..8d6be3f6d 100644 --- a/tests/0day/qwen3/qwen3-14b/evaluate_qwen3_14b_ptd.sh +++ b/tests/0day/qwen3/qwen3-14b/evaluate_qwen3_14b_ptd.sh @@ -11,7 +11,7 @@ TASK="mmlu" # Change for multinode config MASTER_ADDR=localhost -MASTER_PORT=60014 +MASTER_PORT=6000 NNODES=1 NODE_RANK=0 NPUS_PER_NODE=8 diff --git a/tests/0day/qwen3/qwen3-235b-a22b/pretrain_qwen3_235b_a22b_4k_A3_ptd.sh b/tests/0day/qwen3/qwen3-235b-a22b/pretrain_qwen3_235b_a22b_4k_A3_ptd.sh index 140b12696..d20d167f1 100644 --- a/tests/0day/qwen3/qwen3-235b-a22b/pretrain_qwen3_235b_a22b_4k_A3_ptd.sh +++ b/tests/0day/qwen3/qwen3-235b-a22b/pretrain_qwen3_235b_a22b_4k_A3_ptd.sh @@ -9,7 +9,7 @@ export TASK_QUEUE_ENABLE=2 NPUS_PER_NODE=16 MASTER_ADDR=localhost -MASTER_PORT=60014 +MASTER_PORT=6000 NNODES=16 NODE_RANK=0 WORLD_SIZE=$(($NPUS_PER_NODE*$NNODES)) diff --git a/tests/0day/qwen3/qwen3-32b/tune_qwen3_32b_4K_lora_ptd.sh b/tests/0day/qwen3/qwen3-32b/tune_qwen3_32b_4K_lora_ptd.sh index 81f30468b..9de182f49 100644 --- a/tests/0day/qwen3/qwen3-32b/tune_qwen3_32b_4K_lora_ptd.sh +++ b/tests/0day/qwen3/qwen3-32b/tune_qwen3_32b_4K_lora_ptd.sh @@ -4,7 +4,7 @@ export CUDA_DEVICE_MAX_CONNECTIONS=1 NPUS_PER_NODE=8 MASTER_ADDR=localhost -MASTER_PORT=60014 +MASTER_PORT=6000 NNODES=1 NODE_RANK=0 WORLD_SIZE=$(($NPUS_PER_NODE*$NNODES)) diff --git a/tests/0day/qwen3/qwen3-8b/tune_qwen3_8b_4K_lora_ptd.sh b/tests/0day/qwen3/qwen3-8b/tune_qwen3_8b_4K_lora_ptd.sh index 825addbd7..4093fea5b 100644 --- a/tests/0day/qwen3/qwen3-8b/tune_qwen3_8b_4K_lora_ptd.sh +++ b/tests/0day/qwen3/qwen3-8b/tune_qwen3_8b_4K_lora_ptd.sh @@ -4,7 +4,7 @@ export CUDA_DEVICE_MAX_CONNECTIONS=1 NPUS_PER_NODE=8 MASTER_ADDR=localhost -MASTER_PORT=60014 +MASTER_PORT=6000 NNODES=1 NODE_RANK=0 WORLD_SIZE=$(($NPUS_PER_NODE*$NNODES)) diff --git a/tests/pipeline/llama2/llama2_tp1_pp8_patch_gloo_ptd.sh b/tests/pipeline/llama2/llama2_tp1_pp8_patch_gloo_ptd.sh index 1a2593da3..c3f0cbaa5 100644 --- a/tests/pipeline/llama2/llama2_tp1_pp8_patch_gloo_ptd.sh +++ b/tests/pipeline/llama2/llama2_tp1_pp8_patch_gloo_ptd.sh @@ -7,7 +7,7 @@ export CUDA_DEVICE_MAX_CONNECTIONS=1 # config for multinode NPUS_PER_NODE=8 MASTER_ADDR=localhost -MASTER_PORT=60014 +MASTER_PORT=6000 NNODES=1 NODE_RANK=0 WORLD_SIZE=$(($NPUS_PER_NODE*$NNODES)) -- Gitee