Start the decentralized HF: mirror an open-weight model and run it from the mirror
proposed starting reward
0.005 ETH
description
@saltorious1 asked crypto to "PLEASE build a decentralized HF" and has been downloading local models "just in case". Let's start the mirror, one model at a time.
The task: pick an open-weight model whose license allows redistribution (Qwen, Mistral, Gemma, Llama and others: check the license). Put a complete copy of its weights on decentralized storage (IPFS/Filecoin, Arweave or similar), then run the model straight from that copy.
Proof (all three):
- The CID or transaction id of the full model folder, plus a table of every weight file's sha256 next to the sha256 the original Hugging Face repo lists. They must match.
- A screen recording (3 minutes max) that fetches the model from your decentralized copy, not from Hugging Face, and answers one prompt with it locally (llama.cpp, Ollama, MLX, vLLM, anything).
- The license line that allows redistribution.
A bigger or more useful mirror beats a toy one: a 7B+ model, a pack of quantized sizes, or a second independent pin of the same files. Best claim wins, and anyone can add to the pot.
draft author
creating a bounty submits an onchain transaction from your wallet. editing changes only your copy; the shared draft stays the same.
this is a reusable bounty draft. anyone can create a new bounty from it, and the original specification stays unchanged.
discussion
0 comments
loading comments...