SmolLM2 360M, written into Solana.
360M parameters, Q4_K_M quantized, 258.1 MB. Split into 0 chunks of 3600 bytes. Each chunk is one transaction.
Inscription map
Every cell is a group of chunks. It fills as transactions confirm.
0.00%
- chunks inscribed
- 0
- bytes on-chain
- 0 B
- transactions
- 0
- cost so far
- 0.0000 SOL
Recorded on-chain
SHA-256 of the file
—
Merkle root of the chunks
—
Encoding: SPL Noop instruction data, header LLMSOL 0x01 sha256(id)[0..8] u32le(index) then 3,600 raw bytes. Anyone can rebuild the file from an archive node.
Verify what your browser runs
Load the model, then hash the exact bytes in your browser cache and compare them with the SHA-256 recorded on-chain. No trust in this website required.
- Model loaded in this tab
- Local SHA-256 matches the ledger
Engine: llama.cpp in WebAssembly · not loaded.
Timeline
- Sept 15, 2026Transaction v1 goes liveSolana raises the transaction limit to 4,096 bytes. Big enough for 3.6 KB of weights per signature.
- Sept 2026Manifest inscribedSHA-256 and Merkle root written as a memo. Anyone can verify the file.
- NowChunks being inscribed0 of 0 chunks confirmed. The map above fills live.
- Nov 2026Genesis upload to accountsRent drops 90%. The weights move into Solana accounts, loadable straight from any RPC.
Talk to the base model.
No personality on top. This is the raw thing every agent is built from.
SmolLM2 360M (base)
No persona, default sampling
Not loaded
Nothing you type leaves this tab.