mirror of
https://github.com/NVIDIA/dgx-spark-playbooks.git
synced 2026-04-22 18:13:52 +00:00
48 lines
1.1 KiB
YAML
48 lines
1.1 KiB
YAML
version: '3.8'
|
|
|
|
services:
|
|
finetunine:
|
|
image: nvcr.io/nvidia/pytorch:25.11-py3
|
|
deploy:
|
|
replicas: 2
|
|
restart_policy:
|
|
condition: any
|
|
delay: 5s
|
|
max_attempts: 3
|
|
window: 120s
|
|
resources:
|
|
reservations:
|
|
generic_resources:
|
|
- discrete_resource_spec:
|
|
kind: 'NVIDIA_GPU'
|
|
value: 1
|
|
environment:
|
|
- UCX_NET_DEVICES=enp1s0f1np1
|
|
- NCCL_SOCKET_IFNAME=enp1s0f1np1
|
|
- NCCL_DEBUG=INFO
|
|
- TORCH_NCCL_ASYNC_ERROR_HANDLING=1
|
|
- GLOO_SOCKET_IFNAME=enp1s0f1np1
|
|
- CUDA_DEVICE_MAX_CONNECTIONS=1
|
|
- CUDA_VISIBLE_DEVICES=0
|
|
entrypoint: /opt/pytorch-ft-entrypoint.sh
|
|
volumes:
|
|
- ${PWD}:/workspace
|
|
- ${PWD}/pytorch-ft-entrypoint.sh:/opt/pytorch-ft-entrypoint.sh
|
|
- ~/.cache/huggingface/:/root/.cache/huggingface/
|
|
- ~/.ssh:/tmp/.ssh:ro
|
|
ulimits:
|
|
memlock: -1
|
|
stack: 67108864
|
|
networks:
|
|
- host
|
|
healthcheck:
|
|
test: ["CMD", "service", "ssh", "status"]
|
|
interval: 30s
|
|
timeout: 10s
|
|
retries: 10
|
|
|
|
networks:
|
|
host:
|
|
name: host
|
|
external: true
|