DaisyChain-Train / config /cluster.example.env
Quazim0t0's picture
Old-hardware training through emulated GPU logic
309b968 verified
Raw
History Blame Contribute Delete
685 Bytes
# Copy to cluster.env and set on EACH machine (change only RANK per machine).
# Then: source it and run `daisychain-train` (or use the scripts/ helpers).
MASTER_ADDR=100.101.102.10 # the coordinator's IP (Tailscale 100.x recommended)
MASTER_PORT=29560
WORLD_SIZE=3
RANK=0 # 0 on the coordinator, 1 / 2 / ... on the others
GLOO_SOCKET_IFNAME=tailscale0 # the NIC to use (tailscale0, or eth0 / your LAN NIC)
USE_LIBUV=0
# Task + training
DAISY_TASK=daisychain.example_task:ExampleTask # swap for "your_module:YourTask"
DAISY_STEPS=300
DAISY_LR=0.05
DAISY_OPTIMIZER=sgd
DAISY_BASE_BATCH=32
DAISY_STATUS_FILE=status.json
DAISY_SAVE=daisychain_model.pt