example.env

# Example environment variables for local/dev. Do NOT commit real secrets.
# Prefer AWS Secrets Manager in production (JSON keys should mirror these names).

# Region and S3
AWS_REGION=us-east-1
S3_CHECKPOINT_URI=s3://unsloth-fsdp-training/checkpoints/job-xyz
SM_CHECKPOINT_DIR=/opt/ml/checkpoints
DATASET_S3_URI=

# Tokens/keys (use Secrets Manager for real values)
HF_TOKEN=
WANDB_API_KEY=
WANDB_PROJECT=

# NCCL and distributed defaults (orchestrator usually sets these)
NCCL_DEBUG=INFO
NCCL_IB_DISABLE=1
NCCL_P2P_LEVEL=NVL
TORCH_NCCL_ASYNC_ERROR_HANDLING=1
MASTER_ADDR=127.0.0.1
MASTER_PORT=29500
WORLD_SIZE=1
RANK=0

# Optional: Secrets Manager integration
# Set this to the name/ARN of your secret to auto-fetch on container start
AWS_SECRET_ID=dev/ProjectName/ProjectSecrets
S3_BUCKET_NAME=your-new-bucket-name
AWS_BOOTSTRAP=0
# Populate secret JSON in-memory using env var names (comma-separated), e.g. HF_TOKEN,WANDB_API_KEY
SECRET_KEYS=HF_TOKEN,WANDB_API_KEY,WANDB_PROJECT
# Or provide raw JSON directly (preferred in CI)
AWS_SECRETS_JSON=

# Entrypoint validation toggles
VALIDATE_ENV=1
WANDB_DISABLED=0