Configuration

Connito miners are configured via a YAML file. This page walks through every field.

Example Configuration

run:
  run_name: finney-run
  root_path: /home/user/Connito
chain:
  netuid: 102
  coldkey_name: my-wallet
  hotkey_name: my-hotkey
  network: finney
model:
  model_path: nvidia/Nemotron-CC-Math-v1
  base_arch_model: deepseek-ai/DeepSeek-V2-Lite
  precision: fp16-mixed
moe:
  my_expert_group_id: 1
  num_experts: 8
ckpt:
  resume_from_ckpt: true
  base_checkpoint_path: /home/user/Connito/checkpoints/miner
opt:
  lr: 1.0e-05
log:
  log_wandb: false

Field Reference

Run Settings (`run`)

Field	Type	Default	Description
`run_name`	string	required	Unique name for this miner run.
`root_path`	string	required	Base directory for project data.

Chain Settings (`chain`)

Field	Type	Default	Description
`netuid`	integer	`102`	Connito subnet UID.
`network`	string	`finney`	Bittensor network (finney, archive, local).
`coldkey_name`	string	required	Your Bittensor coldkey name.
`hotkey_name`	string	required	Your Bittensor hotkey name.

Model Settings (`model`)

Field	Type	Default	Description
`model_path`	string	required	HuggingFace ID or path to the training dataset/model.
`base_arch_model`	string	required	The base architecture (e.g., `deepseek-ai/DeepSeek-V2-Lite`).
`precision`	string	`fp16-mixed`	Calculation precision (`fp32`, `fp16-mixed`, `bf16-mixed`).

MoE Settings (`moe`)

Field	Type	Default	Description
`my_expert_group_id`	integer	required	Your assigned expert group (0-indexed).
`num_experts`	integer	`8`	Total number of experts in the model.

Checkpoint Settings (`ckpt`)

Field	Type	Default	Description
`resume_from_ckpt`	boolean	`true`	Resume training from the last saved state.
`base_checkpoint_path`	string	required	Global path where all checkpoints are stored.

Optimizer Settings (`opt`)

Field	Type	Default	Description
`lr`	float	`1e-5`	Inner optimizer learning rate.

Cycle Settings (`cycle`)

Field	Type	Default	Description
`owner_ip`	string	required	IP of the subnet owner's phase service.
`owner_port`	integer	`7000`	Port of the subnet owner's phase service.

Generating the Config

Instead of writing this by hand, use the template generator:

python connito/shared/config.py create_config --role miner --coldkey_name <name> --hotkey_name <name> --run_name <name>

Multiple Expert Groups

If you are running multiple miners on the same machine (different expert groups), create a separate config file for each:

# Create separate config directories per expert group, then run each pair of processes:
# Group 0:
python connito/miner/train.py --path /path/to/checkpoints/miner/<hotkey>/group-0/
python connito/miner/model_io.py --path /path/to/checkpoints/miner/<hotkey>/group-0/

# Group 1:
python connito/miner/train.py --path /path/to/checkpoints/miner/<hotkey>/group-1/
python connito/miner/model_io.py --path /path/to/checkpoints/miner/<hotkey>/group-1/

Each miner should have a different checkpoint_dir to avoid conflicts.

Configuration

On this page