/ Модель

Характеристики

Лицензия
Apache 2.0
Библиотека
nanochat
Опубликована
4 окт 2026

queen-v1

Генерация текстаТекст → текст

Небольшая обученная с нуля языковая модель Crawlnet для генерации ответов на вопросы о криптовалютах и блокчейне.

545скачиваний за 30 дней7лайковвеса обновлены6 окт 2026

Бенчмарки

Как модель справилась с задачами пользователей ailist и что о ней публикуют открытые лидерборды

Публичные оценки

Из открытых лидербордов. Сравнивайте результаты одного теста при одинаковых условиях — отсутствие оценки не значит, что модель хуже.

В подключённых источниках результатов для этой модели пока нет.

Что умеет модель

Описание составлено по карточке модели и метаданным Hugging Face

Queen v1 «hatchling» — небольшая языковая модель Crawlnet, обученная с нуля на архитектуре nanochat GPT. В модели 42,1 млн параметров, включая 16,9 млн параметров без эмбеддингов; глубина — 6 слоёв, ширина — 384, 6 голов внимания, словарь — 16 384 токена, контекст — 1024 токена. Репозиторий содержит веса в формате safetensors и использует библиотеку nanochat.

Модель предназначена для генерации ответов на вопросы о криптовалютах и блокчейне. Предобучение проводилось только на страницах криптовалютной тематики, собранных краулерами Crawlnet; для чат-настройки использованы пары вопрос–ответ, созданные DeepSeek Flash на основе этих страниц. Примеры в карточке охватывают Bitcoin, Ethereum, Solana, блокчейны, консенсус и смарт-контракты.

Автор прямо предупреждает, что модель мала, имеет узкую обучающую выборку и может уверенно выдавать неверные сведения, в том числе выдумывать цены, даты, адреса и ссылки. Она не предназначена для финансовых решений и не должна использоваться как источник фактов. Лицензия — Apache-2.0.

Сценарии применения

  • черновые ответы на вопросы о Bitcoin и Ethereum
  • объяснение базовых понятий блокчейна
  • генерация примеров ответов о proof of work и proof of stake
  • ответы на вопросы о Solana и смарт-контрактах

Ограничения

Модель мала и обучена на узком наборе криптовалютных веб-страниц; она может уверенно выдумывать факты, включая цены, даты, адреса и ссылки. Не используйте её как источник фактов или для финансовых решений.

Карточка модели

README от автора на Hugging Face, как он его опубликовал

Открыть на Hugging Face

Queen v1 “hatchling” — Crawlnet

A small language model trained from scratch. Pretraining data and tokenizer: only pages the Crawlnet crawlers read on the crypto web (real browsers). No pretrained weights, no FineWeb, no other corpus. Chat tuning: question/answer pairs generated by deepseek-flash from those crawler pages (details below). Expect a dumb, funny, confidently-wrong model; that is the point.

version1 (queen-v1)
parameters42.1M total, 16.9M non-embedding
architecturenanochat GPT, depth 6, width 384, 6 heads, context 1024, vocab 16384
codekarpathy/nanochat @ 92d63d4e8bb4 (MIT) + crawlnet training/
dataset10,521 pages exported 2026-10-04T08:42:35.913Z; after cleaning 10,518 pages, 392 domains
dataset manifest sha2561258202cf03b6202c5c69112d5a82a34ac6d9d3cb9175fb92282ff4ebea64e32
unique training tokens20,758,910 (+160,122 held out)
tokens trained82,837,504 = 3.99 epochs, 4.9 tokens per non-embedding param
tokenizerBPE, 16384 tokens, trained on the same crawler pages only
validation loss1.5162 bits/byte on held-out crawler pages (pretraining)
chat tuning (SFT)synthetic:deepseek-flash: 14,933 question/answer pairs generated by deepseek-flash (DeepSeek-V4.1-Flash (open weights, MIT license; DeepSeek API terms 4.2(3) allow training other models on outputs), via api.deepseek.com) from 5,000 crawler pages, plus the pairs cut from the pages. The generator was told to use only facts on each page, but its wording is in this model's chat behaviour. headings: 60,562 question/answer pairs cut out of the crawler pages themselves (FAQ questions found on pages, page headings, page titles; answers are verbatim page text). No external model was used. The only human-written words are six fixed question templates such as "Tell me about {heading}."
compute1 GPU (our GPU server), 0.168 h wall, 0.168 GPU-hours
cost$0.06 GPU + ~$2.26 deepseek-flash API for the chat-tuning pairs (list price)
weightshttps://huggingface.co/Crawlnet/queen-v1

Contributors

Queen v1 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.

crawlerownerpagestokens
v4GZUQ…F59x (burn)192496,732
littledog2eek…pBky (burn)194440,607
ghero4FcEE…Cin4 (burn)153429,198
ghero5FcEE…Cin4 (burn)138428,844
spiderman5WKN…Meic (burn)203425,019
mrnakamoto2eek…pBky (burn)192410,263
azeBX6f…WpEf (burn)190401,801
spiderman5WKN…Meic (burn)170380,647
near-weaverBr2o…hHFG (burn)149377,785
crawlzilla3dmq…dLBC (burn)174372,907
spiderman5WKN…Meic (burn)154371,033
ruso4rif…jxnZ (burn)169362,160
sparsityGZUQ…F59x (burn)188357,951
bigdog2eek…pBky (burn)161337,956
testzero…PPUT (burn)218331,261
blackandwhitedogs2eek…pBky (burn)162330,107
swolecrawlsF3QF…Dfrs (burn)162316,290
deezCDYb…2Xxv (burn)192314,699
ghero7FcEE…Cin4 (burn)136304,677
ghero9FcEE…Cin4 (burn)134300,943
spiderman5WKN…Meic (burn)172297,870
ghero2FcEE…Cin4 (burn)126291,399
smokie8kwY…JY7G (burn)155288,743
aze2BX6f…WpEf (burn)158280,718
onepellegrino2eek…pBky (burn)145273,017
spiderman5WKN…Meic (burn)147267,490
ghero3FcEE…Cin4 (burn)123261,159
charlotte2eek…pBky (burn)150252,818
ghero6FcEE…Cin4 (burn)109250,745
spidermanH8iW…VE3N (burn)167250,232
warm6TAx…SE4b (burn)123249,392
ch0pperH8iW…VE3N (burn)123249,313
fern2eek…pBky (burn)153246,142
wilbur2eek…pBky (burn)135243,970
babushkaCj9h…Cf4C (burn)138243,091
ghero8FcEE…Cin4 (burn)111224,138
ghero1FcEE…Cin4 (burn)134222,999
lmaocryH6GW…Hmdr (burn)110218,530
ghero10FcEE…Cin4 (burn)112210,962
ch0pper2H8iW…VE3N (burn)115208,241
yancyDPXE…XNqJ (burn)109207,054
dashH2Lt…LbYp (burn)100192,604
jeff81Y8…GXSt (burn)115188,852
manylegsETgB…wfpa (burn)114183,661
spiderman5WKN…Meic (burn)86180,939
jdubs3v2K…GTgA (burn)62171,264
vinland3cSP…oYeU (burn)61170,127
gigachadfuruAbMD…52T4 (burn)96168,825
jackDiu8…atMh (burn)82168,503
redacted002BvVr…URcY (burn)94165,985
weaverF5Vd…3hVn (burn)80162,320
seig4gtD…rJMz (burn)67159,469
pp3JVm…Fsm4 (burn)81159,331
werkle-weaver3api…NA9T (burn)100157,605
pp3JVm…Fsm4 (burn)78151,260
redacted003BvVr…URcY (burn)65147,573
weenis36ny…e6gG (burn)87147,478
oreocakester36ny…e6gG (burn)73145,266
morelegsETgB…wfpa (burn)70142,642
elevatd3MRp…31K7 (burn)70142,065
smokiev28kwY…JY7G (burn)85141,041
peterparkerAbMD…52T4 (burn)82132,506
web-reaperKrrc…fSaT (burn)81130,175
redacted001BvVr…URcY (burn)63121,647
spidermanFGyn…UuQt (burn)62118,440
turtletheoryipMd…U4xb (burn)53107,294
benny36ny…e6gG (burn)8194,983
cashH2Lt…LbYp (burn)4085,927
grand-couFGcD…k1Dc (burn)4256,797
spy8zQb…eogQ (burn)2449,975
the queen's own crawlers (6)—2,0783,778,418

Pages and tokens count what made it into this version's dataset after cleaning and dedupe. Machine-readable: contributors.json.

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).
  • Chat-tuning question/answer pairs generated by deepseek-flash from crawler 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/top-k in samples.json), unedited:

What is Bitcoin?

Bitcoin is a peer-to-peer cash system that allows users to send transactions and earn rewards by sending them to a peer-to-peer network. It is a set of cryptocurrencies that are used to run a smart contract, a smart contract.

Who created Bitcoin?

Bitcoin is created by Satoshi Nakamoto. Bitcoin holders have the potential to be the most trusted person or person in the world. Bitcoin is named after the pseudonym of Satoshi Nakamoto.

What is a blockchain?

A blockchain is a system that is designed for use with decentralized applications (dApps). It is the foundation, not the underlying infrastructure. A blockchain has its own consensus layer, and has to be the most used method to execute decentralized applications.

Explain proof of work.

Proof of work (RPOW) is an important element of a set of consensus changes that can be applied to bitcoin as transactions by only new bitcoin. Although the set of new transactions, this is not the only way to solve the consensus problem. The rest of the work of a single set of transactions must be recorded sequentially as before a previous block. Bitcoin is the process of creating transactions and computing a new one.

Explain proof of stake.

The proof producer generates a new key in the validator’s pubkey. The validator’s public key is determined by the validator’s pubkey. The new DelegationContract represents the total stake of a validator’s public key, and the associated stake is allocated to a new validator. The new key, the new public key, and the new key is added to the validator’s pubkey as a stake, which is the address of the new validator’s private key. The new key is a new validator key, and the new key is added to the new key. It is added to the new validator key, allowing for

What is Ethereum?

A Bitcoin blockchain is a decentralized digital currency that uses cryptography to secure transactions and transactions. It is used to provide a blockchain ledger that acts as a peer-to-peer network. This is how digital assets are stored and secured and managed using a peer-to-peer network.

What is a smart contract?

A smart contract is a smart contract that allows users to create tokens that can be transferred between a specific contract and a specific amount of a predefined contract. The smart contract is a smart contract that manages the logic of a smart contract. A smart contract can also have extensive permissions' permissions, including the smart contract code.

What is gas on Ethereum?

Ethereum is a decentralized application platform that allows developers to create decentralized applications (dapps) that build decentralized apps that can benefit from Ethereum's development and compatibility.

What is Solana?

Solana is a blockchain built by the Solana network. The blockchain is a computer that connects to Solana and is fully the decentralized network of nodes and nodes on the Solana network. Solana is a blockchain platform that is used to run decentralized applications (dapps) and smart contracts. It is used to provide a smart contract that powers certain types of applications and assets. Solana is the leading blockchain-based cryptocurrency that powers Solana.

Why is Solana fast?

Solana is fast, fast, and cost-effective, fast, efficient, and cost-effective solutions.

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.

Как запустить

На своей GPU в ailist. Цена — по текущим предложениям провайдеров

GPU в ailist

Универсальный рантайм · nanochat

Нельзя запустить

Архитектура ? неизвестна transformers в образе рантайма.

Ревизия весов
8723c14

0 моделей