54 lines
1.1 KiB
TOML
54 lines
1.1 KiB
TOML
# Foundry Swarm Worker Configuration
|
|
#
|
|
# All values can be overridden via environment variables:
|
|
# SW_COORDINATOR_HOST, SW_COORDINATOR_PORT, SW_BATCH_SIZE, etc.
|
|
|
|
[network]
|
|
# Coordinator connection
|
|
coordinator_host = "coordinator"
|
|
coordinator_port = 7100
|
|
|
|
# Reconnection settings
|
|
reconnect_attempts = 10
|
|
reconnect_delay_sec = 5
|
|
|
|
[hardware]
|
|
# Number of CPU threads (0 = auto-detect)
|
|
num_threads = 0
|
|
|
|
# Maximum memory in MB (0 = auto-detect)
|
|
max_memory_mb = 0
|
|
|
|
[training]
|
|
# Batch size (0 = auto from hardware profile)
|
|
batch_size = 0
|
|
|
|
# Gradient accumulation steps (0 = auto, coordinator can override)
|
|
accumulation_steps = 0
|
|
|
|
[data]
|
|
# Path to training data
|
|
path = "/app/data"
|
|
|
|
# Data shard index (-1 = auto-assign from coordinator)
|
|
shard_index = -1
|
|
|
|
[compression]
|
|
# Gradient compression: "none", "fp16", "topk", "qsgd"
|
|
# Note: Coordinator can override this
|
|
type = "topk"
|
|
|
|
# TopK ratio
|
|
topk_ratio = 0.1
|
|
|
|
[identity]
|
|
# Worker ID (empty = auto-generate from hostname)
|
|
worker_id = ""
|
|
|
|
[logging]
|
|
# Log level: "debug", "info", "warn", "error"
|
|
level = "info"
|
|
|
|
# Log file (empty for stdout)
|
|
file = ""
|