/ Модель

Характеристики

Лицензия
Apache 2.0
Библиотека
nanochat
Опубликована
5 окт 2026

queen-v2

Генерация текстаТекст → текст

Небольшая англоязычная модель Crawlnet для генерации ответов на вопросы о криптовалютах и блокчейнах, обученная с нуля на тематических веб- и ончейн-данных.

334скачиваний за 30 дней5лайковвеса 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 v2 «crawler» — небольшая языковая модель Crawlnet, обученная с нуля на архитектуре nanochat GPT. В ней 75,5 млн параметров, контекст составляет 1024 токена, словарь — 16 384 токена. Предобучение использовало страницы криптовалютного веба, собранные краулерами, и открытые рыночные, инвестиционные и ончейн-данные в текстовом виде. Для настройки диалога применялись пары вопросов и ответов, сгенерированные deepseek-flash на основе этих материалов.

Модель предназначена для генерации англоязычных ответов на вопросы о криптовалютах, блокчейнах, криптопроектах и связанных рыночных данных. Узкая тематика и малый размер позволяют использовать её как экспериментальную демонстрацию языковой модели, обученной на специализированном корпусе. Примеры в карточке показывают, что ответы могут быть бессвязными или фактически неверными.

Репозиторий распространяется по лицензии Apache-2.0. Модель может выдумывать цены, даты, адреса и ссылки; её ответы не подходят как источник фактов или финансовая рекомендация.

Сценарии применения

  • черновые ответы на вопросы о Bitcoin, Ethereum и других криптовалютах
  • объяснение базовых понятий блокчейна и криптографии
  • генерация текста о DeFi-протоколах и блокчейн-сетях
  • эксперименты с небольшой языковой моделью, обученной на криптотематическом корпусе

Ограничения

Модель мала и обучена на узком корпусе; она может уверенно выдавать бессвязные или ложные сведения, в том числе цены, даты, адреса и ссылки. Не используйте её ответы как источник фактов или финансовую рекомендацию.

Карточка модели

README от автора на Hugging Face, как он его опубликовал

Открыть на Hugging Face

Queen v2 “crawler” — Crawlnet

A small language model trained from scratch. Pretraining data and tokenizer: only pages the Crawlnet crawlers read on the crypto web (real browsers), plus public market, funding and on-chain records written out as plain-text pages (13% of the tokens, see below). No pretrained weights, no FineWeb, no other corpus. Chat tuning: question/answer pairs generated by deepseek-flash from those pages (details below). Expect a dumb, funny, confidently-wrong model; that is the point.

version2 (queen-v2)
parameters75.5M total, 33.6M non-embedding
architecturenanochat GPT, depth 8, width 512, 8 heads, context 1024, vocab 16384
codekarpathy/nanochat @ 92d63d4e8bb4 (MIT) + crawlnet training/
dataset79,250 pages exported 2026-10-05T10:11:17+00:00; after cleaning 79,202 pages: 34,234 crawler pages from 588 domains + 44,968 record pages
dataset manifest sha2564ba917bb50964b231dcca9e641822941975260afb51ef87f34101958cd50e042
unique training tokens75,776,124 (+801,406 held out); the record pages are about 9,707,932 tokens of it (13%)
tokens trained352,321,536 = 4.65 epochs, 10.5 tokens per non-embedding param
tokenizerBPE, 16384 tokens, trained on this dataset's pages only (crawler pages and record pages)
validation loss1.0196 bits/byte on held-out pages (pretraining)
chat tuning (SFT)synthetic+openbook:deepseek-flash: 335,331 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 111,941 page chunks (crawler pages and record 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: 189,567 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}." Open book: the generated pairs were also trained once more with the page text they were written from in front of the question (270,005 conversations, 47% with a second page the search ranked high for the same question), so she can answer from the pages retrieved for a question.
compute1 GPU (our GPU server), 0.725 h wall, 0.725 GPU-hours
cost$0.25 GPU + ~$49.86 deepseek-flash API for the chat-tuning pairs (list price)
weightshttps://huggingface.co/Crawlnet/queen-v2

Contributors

Queen v2 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
ghero4FcEE…Cin4 (burn)390896,970
v4GZUQ…F59x (burn)432885,864
near-weaverBr2o…hHFG (burn)397881,338
testzero…PPUT (burn)445867,909
littledog2eek…pBky (burn)447858,649
spiderman5WKN…Meic (burn)451857,045
azeBX6f…WpEf (burn)430853,293
ghero5FcEE…Cin4 (burn)362848,798
spiderman5WKN…Meic (burn)423844,824
crawlzilla3dmq…dLBC (burn)413841,677
spiderman5WKN…Meic (burn)401819,782
blackandwhitedogs2eek…pBky (burn)414815,778
onepellegrino2eek…pBky (burn)392796,229
mrnakamoto2eek…pBky (burn)403796,169
spiderman5WKN…Meic (burn)423789,360
werkle-weaver3api…NA9T (burn)340776,187
ch0pperH8iW…VE3N (burn)363770,133
swolecrawlsF3QF…Dfrs (burn)382755,329
manylegsETgB…wfpa (burn)335745,394
aze2BX6f…WpEf (burn)413733,650
deezCDYb…2Xxv (burn)418731,996
ruso4rif…jxnZ (burn)394730,877
ghero3FcEE…Cin4 (burn)353730,650
bigdog2eek…pBky (burn)392727,568
sparsityGZUQ…F59x (burn)398722,394
jackDiu8…atMh (burn)343707,366
pp3JVm…Fsm4 (burn)334704,112
ghero7FcEE…Cin4 (burn)352703,073
charlotte2eek…pBky (burn)405702,661
pp3JVm…Fsm4 (burn)315701,753
warm6TAx…SE4b (burn)349699,534
ghero9FcEE…Cin4 (burn)366693,666
smokie8kwY…JY7G (burn)393686,291
spiderman5WKN…Meic (burn)381676,380
redacted003BvVr…URcY (burn)330671,556
dashH2Lt…LbYp (burn)317667,874
ghero10FcEE…Cin4 (burn)357665,652
babushkaCj9h…Cf4C (burn)366660,752
ghero2FcEE…Cin4 (burn)330654,989
fern2eek…pBky (burn)368649,628
wilbur2eek…pBky (burn)383648,998
lmaocryH6GW…Hmdr (burn)335640,231
ghero6FcEE…Cin4 (burn)326640,198
spidermanH8iW…VE3N (burn)413628,314
ghero8FcEE…Cin4 (burn)354625,402
seig4gtD…rJMz (burn)305625,070
ghero1FcEE…Cin4 (burn)375621,254
ch0pper2H8iW…VE3N (burn)323617,075
spiderman5WKN…Meic (burn)312612,264
weaverF5Vd…3hVn (burn)307612,229
turtletheoryipMd…U4xb (burn)290602,025
gigachadfuruAbMD…52T4 (burn)310596,308
yancyDPXE…XNqJ (burn)322594,881
jeff81Y8…GXSt (burn)339590,258
peterparkerAbMD…52T4 (burn)310586,225
vinland3cSP…oYeU (burn)276573,573
weenis36ny…e6gG (burn)315571,618
redacted001BvVr…URcY (burn)301569,906
web-reaperKrrc…fSaT (burn)334566,293
spidermanFGyn…UuQt (burn)291563,018
redacted002BvVr…URcY (burn)321562,490
morelegsETgB…wfpa (burn)297558,437
cashH2Lt…LbYp (burn)263553,856
smokiev28kwY…JY7G (burn)334546,490
jdubs3v2K…GTgA (burn)278544,089
oreocakester36ny…e6gG (burn)303538,592
benny36ny…e6gG (burn)293526,076
spy8zQb…eogQ (burn)260523,524
e04AcUm…vLRa (burn)213442,310
octangone01G64q…5PiU (burn)210438,727
sneaky6yNt…iikJ (burn)202437,334
007ipMd…U4xb (burn)184429,432
ruso-24rif…jxnZ (burn)181426,382
needmorecrawls4rif…jxnZ (burn)171424,414
aze4BX6f…WpEf (burn)151419,010
elevatd3MRp…31K7 (burn)278418,471
1gjbAyXY…HcDw (burn)181385,564
e02AcUm…vLRa (burn)232384,465
e05AcUm…vLRa (burn)224378,902
e01AcUm…vLRa (burn)197358,102
orb-weaver3Zgo…wfpF (burn)134357,487
e03AcUm…vLRa (burn)219355,183
41nEXTV…831R (burn)171354,115
page-pounderExtb…VrFG (burn)227353,751
crawlzilla-23dmq…dLBC (burn)176351,673
grand-couFGcD…k1Dc (burn)263347,825
kk1438PRb…2Ufe (burn)219347,190
canisFUy5…cjoG (burn)167339,949
41nEXTV…831R (burn)164337,210
tarantulaBkPa…YiMW (burn)181326,420
chungus-spawnH7oG…CsLj (burn)153313,015
aze3BX6f…WpEf (burn)169305,588
aze5BX6f…WpEf (burn)155304,353
aspen41CL…Q9fU (burn)144284,233
smokiev38kwY…JY7G (burn)165283,348
aze6BX6f…WpEf (burn)172279,018
snoopdogDRmh…ow3V (burn)149271,995
funnelweb1DmsF…uvrF (burn)137260,531
son-of-chungusH7oG…CsLj (burn)133220,204
onaizah4KZT…HerD (burn)124210,672
the queen's own crawlers (10)—4,2018,026,689

Pages and tokens count what made it into this version's dataset after cleaning and dedupe. Machine-readable: contributors.json. The 44,968 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: 44,968 pages, about 9,707,932 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.
  • Chat-tuning question/answer pairs generated by deepseek-flash 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 decentralized cryptocurrency that aims to provide a secure, decentralized, and secure digital asset. It operates on a Proof-of-Work consensus mechanism, where users are required to compete to solve a computationally intensive puzzle and earn money.

Who created Bitcoin?

Bitcoin was created by Satoshi Nakamoto, who placed the name in the white paper. The document is distributed under its declared BSD-3-Clause licence.

What is a blockchain?

A blockchain is a public, shared ledger that record all the transactions of all the accounts in a block. Transactions are verified and added to the blockchain. It is a distributed database that is constantly managed and maintained by multiple people. This type of database can be used to create a record of all the addresses that have transacted on the chain. For instance, with block 100,000, there will be 52 million addresses transacted on the blockchain.

Explain proof of work.

Proof of work is a process that was intended to work with proof-of-work blockchains to do work in a more energy efficient way by improving the underlying mathematical problems the blockchain uses. Proof of Work is a consensus mechanism that is designed to address this issue by requiring miners to compete with each other to solve complex mathematical problems. However, it can be argued that, in practice, it is possible to solve this problem by solving computationally intensive puzzles.

Explain proof of stake.

The proof of stake protocol is a mechanism enabling a distributed validator network to ensure that the amount of ETH staked by validators on the network is always available to all nodes, thereby reducing the risk of a coalition of validators. The proof of stake protocol allows the network to run more validators than a validator can operate on.

What is Ethereum?

↑ (EN) Eric W. Weisen, Guilty, Friedman (2018). "Ethereum's evolution: A New Formal 'The Fintech' Anders'". The Diamond. 2018. https://ethereum-magicians.org/t/eip-20Ethereum-Cancun-Deneb-7540/9258/5/27847

What is a smart contract?

A smart contract is a program deployed to a blockchain that automatically execute a certain function on a smart contract. The smart contract automatically executes the function when it receives a payment, usually through a crypto wallet.

What is gas on Ethereum?

Gas is the unit of computation for the access of Ethereum nodes. It works like a DoS attack, and the cost is not in the same sense as a memory attack.

What is Solana?

Solana is a high-performance blockchain platform that combines parallel execution and smart contract execution. It is a key component of Solana's architecture and consensus structure, providing minimal latency.

Why is Solana fast?

Solana is fast because it has a low fees and low transaction costs, and transaction fees can increase dramatically as the chain grows. This time, Solana has become one of the fastest and most successful chains in the industry. Solana's network is a high-speed blockchain that executes transactions faster than the network currently executes. It is a high-speed internet that is slow, fast, and cheap. Solana has a high-speed network that can process over 65,000 transactions per second (TPS) while processing thousands of transactions per second.

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
f306176

0 моделей