| ssh 110.129.0.18 |
|
|
| docker exec -it verl bash |
| cd /mnt/models/wh/code/MindSpeed-LLM-v2.3.0 |
| conda activate ms |
|
|
| |
| |
| wandb login |
| |
| |
|
|
| source /usr/local/Ascend/cann/ascend-toolkit/set_env.sh |
| source /usr/local/Ascend/cann/nnal/atb/set_env.sh |
|
|
| bash examples/mcore/qwen3/pretrain_qwen3_0point6b_4K_ptd_new-150M-16.sh |
| bash examples/mcore/qwen3/pretrain_qwen3_0point6b_4K_ptd_new-150M-18.sh |
| bash examples/mcore/qwen3/pretrain_qwen3_0point6b_4K_ptd_new-150M-20.sh |
| bash examples/mcore/qwen3/pretrain_qwen3_0point6b_4K_ptd_new-150M-22.sh |
|
|
| export CUDA_DEVICE_MAX_CONNECTIONS=1 |
| python convert_ckpt_v2.py --load-model-type mg --save-model-type hf \ |
| --load-dir /mnt/models/wh/models/qwen_200m_0615 \ |
| --save-dir /mnt/models/wh/models/qwen_200m_0615/200M_hf/ \ |
| --model-type-hf qwen3 |
|
|
|
|
|
|
| kill -9 $(pgrep -f python) |
