feat: unify Docker training configuration in YAML
This commit is contained in:
+1
-2
@@ -72,9 +72,8 @@ services:
|
||||
- BASE_MODEL=${BASE_MODEL:-/models/base}
|
||||
- CHECKPOINT_ROOT=/checkpoints
|
||||
- TRAIN_GPU_COUNT=${TRAIN_GPU_COUNT:-all}
|
||||
- TRAIN_PARALLEL_MODE=${TRAIN_PARALLEL_MODE:-auto}
|
||||
- CUDA_VISIBLE_DEVICES
|
||||
- NCCL_P2P_DISABLE
|
||||
- NCCL_NET_GDR_LEVEL
|
||||
entrypoint: ["bash", "/app/scripts/docker/train-entrypoint.sh"]
|
||||
ipc: ${TRAIN_IPC_MODE:-host}
|
||||
stop_grace_period: ${TRAIN_STOP_GRACE_PERIOD:-10m}
|
||||
|
||||
Reference in New Issue
Block a user