Question 1

How do miners generate value in a market simulation?

Accepted Answer

Miners run trading agents that compete on kappa (risk-adjusted returns) and round-trip volume across multiple simulated order books. Validators score performance and allocate rewards proportionally. The simulator includes realistic background agents and matching rules, so competition drives agents to develop strategies that handle diverse market conditions — the data they generate in the process is valuable for research and model training.

Question 2

What is GenTRX and how does it work?

Accepted Answer

GenTRX is distributed training of a shared transformer model on order-book data. Each round, miners receive an assignment (a slice of the order-book history), train the model locally on that slice, compress the gradient, and upload it to a personal S3 bucket. Validators fetch and score the gradient against held-out validation books, then aggregate the best-scoring deltas into a new checkpoint published on-chain. Miners earn from a separate training pool (~5% by default) in addition to trading rewards, with no trade-off because training runs in a background thread.

Question 3

Do I need to participate in GenTRX training?

Accepted Answer

No. Trading alone earns 95% of rewards by default. GenTRX is opt-in for both miners and validators. Miners that skip training run a standard agent with no S3 bucket or gradient code; validators that skip it operate as pure trading validators. The training pool allocation scales with active participation and returns to trading if unused.

Question 4

How is gradient quality scored?

Accepted Answer

Validators run a gradient server that replays simulation ticks, accumulates order-book data, and flushes parquets to S3. Each miner gradient is scored twice: against the miner's own assigned training data and against held-out validation books shared across miners. An overfitting penalty flags when own-data loss runs ahead of held-out loss. Scores are rank-normalized and the best delta updates the checkpoint.

Question 5

How do miners discover validator and aggregator buckets?

Accepted Answer

Validators and the canonical aggregator commit their S3 bucket credentials (read-only tokens) to the Bittensor chain at startup using the Commitments pallet. Miners and validators discover these commitments on-chain and use them to pull checkpoints and training data. No manual credential distribution is needed.

Question 6

What are the hardware requirements?

Accepted Answer

Validators need 32GB RAM, 16 cores, and Ubuntu 22.04+ to run the C++ simulator and Python validator. GenTRX-enabled validators additionally need a GPU (8GB+ VRAM) for the gradient server. Miners have no hard requirements but benefit from a GPU (6-8GB VRAM) when running GenTRX; CPU-only training is supported but slower. All participants need stable internet and an S3-compatible bucket (Cloudflare R2, Storj, or Hippius) if enabling GenTRX.

MVTRX

MVTRX

Hardware Requirements

Hardware Requirements

Miner Setup

Validator Setup

Participate

Participate

Hardware Requirements

Hardware Requirements

Miner Setup

Validator Setup