Step 1 of 2 — Upload to Walrus
Upload your assets.
Dataset + model → Walrus, then post a job.
Drop your training set and (optionally) a base model. Datasets are stored on Walrus testnet; HuggingFace models are linked by reference so miners pull them directly from the Hub. Blob IDs auto-fill the training form on step 2.
Walrus storage is paid in WAL · ~0.001 WAL/KB · 5 epochs
Training dataset
Drop file or click to browse
.zip · .jsonl · .parquet · .csv · .json
Max 200 MB · uploads to Walrus testnet
Max 200 MB · uploads to Walrus testnet
Base model (optional)
HuggingFace model
model_id
google/gemma-4-E2B
View on HuggingFace Miners pull from Hub — no Walrus upload needed
Ready to train
Upload a dataset first
Walrus + WAL
Storage is paid in WAL tokens. Get WAL from the Walrus testnet faucet by swapping SUI → WAL on-chain. Blobs are pinned for 5 epochs (~5 days).
Any format
Upload .jsonl, .csv, .parquet, or .zip archives. The training miner receives the blob ID and fetches the raw data directly from Walrus.
HF / GitHub import
Paste a huggingface.co/owner/model or github.com URL. HuggingFace model repos are linked by reference (miners pull from Hub). GitHub archives and HF direct file links are fetched and uploaded to Walrus.
Max 200 MB
Testnet publisher limit. For larger datasets split into shards, upload each, and pass multiple blob IDs to the miner out-of-band.