Queen v7 “empress” — Crawlnet

A small language model trained from scratch. Pretraining data: pages the Crawlnet crawlers read on the crypto web (real browsers; 72% of the tokens trained), public market, funding and on-chain records written out as plain-text pages (3%), study notes a general-purpose LLM wrote for 183,105 of those pages (20%) and question/answer pairs a general-purpose LLM generated from those pages (6%); details in Pretraining data below. Tokenizer: an earlier Queen's, reused unchanged (trained on that version's pages only). No pretrained weights, no FineWeb, no other corpus. Chat tuning: question/answer pairs generated by a general-purpose LLM from those pages, pairs cut from the pages and the same pairs with their page in front of the question (details below).

Pretraining data

Unique tokens: every document trained at least once, counted once (with its BOS token and source line). Tokens trained: every pass of it, plus the Q&A documents regrouped for the end of pretraining. The run trained 4,829 steps of 524,288 tokens = 2,531,786,752 of the stream's 2,532,249,995. The released weights are the step-4,579 checkpoint: they saw the first 95% of it (the counts above are the whole stream).

Contributors

Queen v7 was trained on pages found by these crawlers. Owners spawned theirs by burning $CRAWLNET; the wallets are the ones that made the burn, public on-chain. Since 9 Oct 2026 every crawl.md NFT is a crawler of its own: its pages are credited to the wallet that holds the NFT (its credits name when one is set).

Pages and tokens count what made it into this version's dataset after cleaning and dedupe. Machine-readable: contributors.json. The 45,629 record pages (see below) were not found by a crawler and are not in this table.

What went in besides crawler pages

• The nanochat training code, architecture and hyperparameter rules (no weights, no data).
• Six hand-written question templates used to turn page headings into questions (see SFT above).
• Record pages: 45,629 pages, about 10,531,110 tokens, of public market, funding and on-chain records (projects and their funding rounds, investors, token unlocks and token facts, DeFi protocols, chains and stablecoins, DEX and on-chain token statistics), written out as plain text from public data. Only factual fields were kept: no third-party labels, ratings or risk scores, and no social-media posts. They are in the pretraining data, the tokenizer and the generated chat pairs; the pairs cut from page headings use crawler pages only.
• Study notes: a general-purpose LLM was given 201,410 crawled pages and wrote study notes from each: an explanation, an FAQ, a list of facts and entities and a Q&A list. A fact checker dropped every item with a number, address, URL or ticker that is not on its page (20,026 items); no page of a held-out site was given to it, and 6,123 notes that matched a benchmark question were left out. The notes of 183,105 pages are in the pretraining data; their wording, and any mistake the checker could not catch, is in this model.
• Generated Q&A pairs in pretraining: 965,352 question/answer pairs a general-purpose LLM generated from crawler pages and record pages for an earlier Queen's chat tuning, written as Q&A documents. 39,964 more (a fixed 2% of the questions and the earlier validation split) and 303 that matched a benchmark question never entered pretraining.
• Chat-tuning question/answer pairs generated by a general-purpose LLM from crawler pages and record pages (see SFT above).
• Nothing else. nanochat's default SFT mixture (SmolTalk, MMLU, GSM8K) and its CORE eval were switched off.

What it says

Fixed prompts, seeded sampling (temperature 0.7, top-k 50), unedited:

What is Bitcoin?

Bitcoin is a digital asset that uses cryptography to secure transactions and control the creation of new units. Unlike government-issued currencies, Bitcoin is not controlled by any central authority. Instead, it is distributed across a global network of computers, and miners provide the infrastructure that keeps the system running.

Who created Bitcoin?

Bitcoin was created in 2009 by a person using the pseudonym Satoshi Nakamoto. The name Satoshi Nakamoto is a pseudonym used by the anonymous creator of Bitcoin. The Bitcoin whitepaper was released in 2008 as a white paper published by Satoshi Nakamoto and released to the public in 2009. The Bitcoin network was made available for download by Satoshi Nakamoto in 2009. Bitcoin has a fixed supply of 21 million coins, which will be released by the Bitcoin protocol in 2140.

What is a blockchain?

A blockchain is a distributed ledger of records that can be used to record transactions. These records are maintained by peer-to-peer network participants, which are compensated for their work with rewards in the form of native cryptocurrency tokens.

Explain proof of work.

The proof of work system requires miners to search through random numbers in order to find a correct hash. This process involves making numerous iterations of a cryptographic algorithm, as well as checking if the output of the hash function is below a certain threshold. The difficulty of finding a hash that is below the threshold is adjusted every 2016 blocks based on the time taken to find the last 2016 blocks. For example, if the last 2016 blocks were found on August 15th, 2016, the difficulty would be adjusted to 150,000,000,000

Explain proof of stake.

The proof of stake (PoS) consensus engine has been developed to address the major issues with proof of work (PoW). PoS is a consensus mechanism invented by the American computer scientist and cryptographer, Sunny King, and it is the most popular consensus engine for the blockchain, such as the Bitcoin and Ethereum blockchains. The major issues with proof of work (PoW) are high energy expenditure and high hardware requirements.

What is Ethereum?

In August of 2021, Ethereum underwent its biggest upgrade, called "The Merge". This upgrade was the first time that Ethereum transitioned from a Proof-of-Work (PoW) to a Proof-of-Stake (PoS) mechanism. In other words, it made Ethereum more energy-efficient, decentralized, and scalable. However, it was not without its drawbacks. One major issue that the Merge brought was the network's ability to scale. As the number of transactions rose, the network's ability to process transactions quickly.

What is a smart contract?

A smart contract is a program stored and run on a blockchain that automatically enforces certain conditions. Smart contracts are written in a programming language and compiled into bytecode that runs in the EVM. Once deployed, the contract's code cannot be changed, and it executes identically across the network.

What is gas on Ethereum?

Gas is the unit in which the computational power required to execute a transaction is measured. Every transaction has a cost, which is expressed in gwei (one-billionth of an ETH), and the cost of executing each transaction is multiplied by the amount of gas used during the execution. In simple terms, the cost of executing a transaction is directly proportional to the amount of computing power required to process the transaction. Ethereum is the world's second-largest cryptocurrency by market capitalisation.

What is Solana?

Solana is a layer 1 blockchain that uses a proof-of-stake consensus mechanism called Tower BFT. Solana's Tower BFT is a variant of the Practical Byzantine Fault Tolerance (PBFT) consensus algorithm developed by the Byzantine fault-tolerant computer scientist and distributed systems engineer, Andrew Miller. In addition to its proof-of-stake consensus mechanism, Solana also uses a unique cryptographic clock to timestamp transactions and provide a verifiable audit trail for transactions.

Why is Solana fast?

Solana is a high-throughput blockchain launched in 2017. It has a growing user base but also experiences some of the most active periods of activity in crypto. In 2021, Solana processed an average of over 500 million transactions per second (TPS). In August 2023, it processed over 200 million transactions per second! In October 2024, it processed over 1.2 billion transactions per second. As Solana scales, it also scales with the network to ensure that it continues to be a high-performance blockchain.

Use it

chat.py in this folder needs only karpathy/nanochat (MIT) at the commit above:

pip install torch tiktoken safetensors
git clone https://github.com/karpathy/nanochat && git -C nanochat checkout 92d63d4e8bb4
PYTHONPATH=nanochat python chat.py "What is staking?"

Or load it yourself: config.json has the nanochat GPTConfig; build nanochat.gpt.GPT, load model.safetensors; the tokenizer is tokenizer.tiktoken (tiktoken rank file) + the pattern and special tokens in tokenizer.json. Chat format: <|bos|><|user_start|>...<|user_end|><|assistant_start|>.

Limitations

Tiny model, narrow data. It makes things up, including prices, dates, addresses and links. It is not financial advice and must not be used as a source of facts.