mirror of
https://github.com/k2-fsa/icefall.git
synced 2025-08-09 01:52:41 +00:00
add multi-node
This commit is contained in:
parent
4a29430349
commit
50fc1aba60
@ -295,7 +295,7 @@ if [ $stage -le 16 ] && [ $stop_stage -ge 16 ]; then
|
|||||||
done
|
done
|
||||||
fi
|
fi
|
||||||
|
|
||||||
train_cmd_args="--max-duration 1200 \
|
train_cmd_args="--max-duration 800 \
|
||||||
--enable-musan False \
|
--enable-musan False \
|
||||||
--audio-key audio --text-key continuation \
|
--audio-key audio --text-key continuation \
|
||||||
--exp-dir $exp_dir \
|
--exp-dir $exp_dir \
|
||||||
@ -318,7 +318,7 @@ if [ $stage -le 16 ] && [ $stop_stage -ge 16 ]; then
|
|||||||
# No pretrained model or sampler state dict needed for the first run
|
# No pretrained model or sampler state dict needed for the first run
|
||||||
fi
|
fi
|
||||||
|
|
||||||
torchrun --nproc_per_node $ngpu ./qwen_omni/train.py \
|
torchrun --nproc_per_node $ngpu --nnodes $SLURM_JOB_NUM_NODES --rdzv_endpoint $MASTER_ADDR:$MASTER_PORT --rdzv_backend c10d --rdzv_id $SLURM_JOBID ./qwen_omni/train.py \
|
||||||
$train_cmd_args
|
$train_cmd_args
|
||||||
fi
|
fi
|
||||||
|
|
||||||
|
Loading…
x
Reference in New Issue
Block a user