Crawlnet/queen-v3

Queen v3 “weaver” — 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 (6% 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.

version	3 (queen-v3)
parameters	122.6M total, 59.6M non-embedding
architecture	nanochat GPT, depth 10, width 640, 5 heads, context 2048, vocab 16384
code	karpathy/nanochat @ 92d63d4e8bb4 (MIT) + crawlnet training/
dataset	135,077 pages exported 2026-10-06T11:30:31+00:00; after cleaning 134,994 pages: 89,367 crawler pages from 1936 domains + 45,627 record pages
dataset manifest sha256	00a69452525b0533fdddde5ff0f345ac19d451cd5cc842b4711a0cfd833edea3
unique training tokens	191,488,840 (+1,996,669 held out); the record pages are about 10,950,432 tokens of it (6%)
tokens trained	625,999,872 = 3.27 epochs, 10.5 tokens per non-embedding param
tokenizer	BPE, 16384 tokens, trained on this dataset's pages only (crawler pages and record pages)
validation loss	0.8174 bits/byte on held-out pages (pretraining)
chat tuning (SFT)	synthetic+openbook:deepseek-flash: 247,648 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 170,994 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: 487,456 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 (183,204 conversations, 46% with a second page the search ranked high for the same question), so she can answer from the pages retrieved for a question.
compute	1 GPU (our GPU server), 1.771 h wall, 1.771 GPU-hours
cost	$0.60 GPU + ~$36.33 deepseek-flash API for the chat-tuning pairs (list price)
weights	https://huggingface.co/Crawlnet/queen-v3

Contributors

Queen v3 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.

crawler	owner	pages	tokens
pp	3JVm…Fsm4 (burn)	655	1,649,028
spiderman	5WKN…Meic (burn)	769	1,634,545
smokie	8kwY…JY7G (burn)	721	1,631,346
crawlzilla	3dmq…dLBC (burn)	756	1,627,782
ghero5	FcEE…Cin4 (burn)	690	1,622,127
pp	3JVm…Fsm4 (burn)	606	1,586,243
aze	BX6f…WpEf (burn)	724	1,541,422
spiderman	5WKN…Meic (burn)	735	1,521,654
ch0pper	H8iW…VE3N (burn)	690	1,514,921
aze2	BX6f…WpEf (burn)	734	1,514,315
werkle-weaver	3api…NA9T (burn)	676	1,508,869
mrnakamoto	2eek…pBky (burn)	744	1,502,714
test	zero…PPUT (burn)	774	1,498,672
ghero4	FcEE…Cin4 (burn)	726	1,489,409
spiderman	5WKN…Meic (burn)	749	1,477,210
cash	H2Lt…LbYp (burn)	580	1,475,211
onepellegrino	2eek…pBky (burn)	729	1,471,774
jdubs	3v2K…GTgA (burn)	607	1,466,629
charlotte	2eek…pBky (burn)	736	1,454,450
fern	2eek…pBky (burn)	706	1,446,090
v4	GZUQ…F59x (burn)	722	1,443,046
spiderman	5WKN…Meic (burn)	724	1,441,441
blackandwhitedogs	2eek…pBky (burn)	739	1,425,614
ghero10	FcEE…Cin4 (burn)	682	1,396,780
ruso	4rif…jxnZ (burn)	731	1,396,359
littledog	2eek…pBky (burn)	737	1,395,448
ghero6	FcEE…Cin4 (burn)	636	1,392,838
redacted002	BvVr…URcY (burn)	635	1,386,664
near-weaver	Br2o…hHFG (burn)	706	1,386,197
sparsity	GZUQ…F59x (burn)	721	1,380,285
yancy	DPXE…XNqJ (burn)	660	1,372,115
ghero3	FcEE…Cin4 (burn)	695	1,367,591
manylegs	ETgB…wfpa (burn)	628	1,362,227
seig	4gtD…rJMz (burn)	628	1,355,402
ghero8	FcEE…Cin4 (burn)	676	1,352,758
wilbur	2eek…pBky (burn)	701	1,348,588
swolecrawls	F3QF…Dfrs (burn)	697	1,342,354
benny	36ny…e6gG (burn)	622	1,342,326
ch0pper2	H8iW…VE3N (burn)	625	1,331,559
dash	H2Lt…LbYp (burn)	634	1,323,652
warm	6TAx…SE4b (burn)	675	1,320,192
jack	Diu8…atMh (burn)	671	1,310,227
bigdog	2eek…pBky (burn)	700	1,306,680
lmaocry	H6GW…Hmdr (burn)	647	1,302,177
babushka	Cj9h…Cf4C (burn)	673	1,280,835
redacted003	BvVr…URcY (burn)	651	1,278,545
gigachadfuru	AbMD…52T4 (burn)	655	1,269,907
weaver	F5Vd…3hVn (burn)	632	1,255,923
ghero9	FcEE…Cin4 (burn)	673	1,253,976
deez	CDYb…2Xxv (burn)	743	1,252,961
spiderman	H8iW…VE3N (burn)	752	1,250,842
smokiev2	8kwY…JY7G (burn)	656	1,244,408
spiderman	5WKN…Meic (burn)	619	1,243,629
spy	8zQb…eogQ (burn)	579	1,242,732
ghero7	FcEE…Cin4 (burn)	671	1,237,701
spiderman	FGyn…UuQt (burn)	615	1,228,981
spiderman	5WKN…Meic (burn)	691	1,222,845
ghero2	FcEE…Cin4 (burn)	631	1,214,378
redacted001	BvVr…URcY (burn)	642	1,208,395
e02	AcUm…vLRa (burn)	574	1,203,136
turtletheory	ipMd…U4xb (burn)	600	1,193,175
morelegs	ETgB…wfpa (burn)	631	1,171,977
ghero1	FcEE…Cin4 (burn)	680	1,170,013
weenis	36ny…e6gG (burn)	565	1,156,575
octangone01	G64q…5PiU (burn)	535	1,146,451
vinland	3cSP…oYeU (burn)	605	1,137,549
sneaky	6yNt…iikJ (burn)	523	1,128,930
orb-weaver	3Zgo…wfpF (burn)	445	1,121,301
web-reaper	Krrc…fSaT (burn)	657	1,117,543
aze4	BX6f…WpEf (burn)	486	1,107,142
peterparker	AbMD…52T4 (burn)	642	1,091,238
aze5	BX6f…WpEf (burn)	484	1,088,067
oreocakester	36ny…e6gG (burn)	622	1,082,914
e05	AcUm…vLRa (burn)	565	1,081,442
e04	AcUm…vLRa (burn)	550	1,076,001
jeff	81Y8…GXSt (burn)	657	1,062,912
007	ipMd…U4xb (burn)	488	1,053,236
1gjb	AyXY…HcDw (burn)	494	1,044,150
canis	FUy5…cjoG (burn)	469	1,037,454
page-pounder	Extb…VrFG (burn)	550	1,025,958
e01	AcUm…vLRa (burn)	522	1,023,918
kk143	8PRb…2Ufe (burn)	546	1,010,842
41n	EXTV…831R (burn)	512	997,888
e03	AcUm…vLRa (burn)	518	995,799
ruso-2	4rif…jxnZ (burn)	495	986,696
needmorecrawls	4rif…jxnZ (burn)	518	973,364
elevatd	3MRp…31K7 (burn)	571	965,189
aze3	BX6f…WpEf (burn)	494	950,331
aze6	BX6f…WpEf (burn)	497	942,707
snoopdog	DRmh…ow3V (burn)	488	939,193
matrix-multiplier	6s1H…36Jg (burn)	324	926,865
aspen	41CL…Q9fU (burn)	465	923,557
crawlzilla-2	3dmq…dLBC (burn)	485	920,025
funnelweb1	DmsF…uvrF (burn)	482	894,026
onaizahparisnajd	4KZT…HerD (burn)	315	879,329
41n	EXTV…831R (burn)	480	877,520
chungus-spawn	H7oG…CsLj (burn)	474	873,686
redacted	BvVr…URcY (burn)	304	869,415
tarantula	BkPa…YiMW (burn)	508	856,717
smokiev3	8kwY…JY7G (burn)	462	843,648
onaizah	4KZT…HerD (burn)	428	832,589
grand-cou	FGcD…k1Dc (burn)	563	832,183
son-of-chungus	H7oG…CsLj (burn)	458	814,445
onaizahparisnajd	4KZT…HerD (burn)	300	811,509
onaizahparisnajd	4KZT…HerD (burn)	295	787,528
creepycrawly	3v2K…GTgA (burn)	241	786,130
9ballshooter	APM9…89Nm (burn)	240	777,586
onaizahparisnajd	4KZT…HerD (burn)	318	753,269
redacted	BvVr…URcY (burn)	296	743,433
onaizahparisnajd	4KZT…HerD (burn)	309	739,697
onaizahparisnajd	4KZT…HerD (burn)	287	719,159
redacted	BvVr…URcY (burn)	319	708,897
goatishere2	58pZ…waHQ (burn)	307	704,407
redacted	BvVr…URcY (burn)	307	703,071
redacted	BvVr…URcY (burn)	286	700,234
redacted	BvVr…URcY (burn)	307	689,288
onaizahparisnajd	4KZT…HerD (burn)	276	682,222
redacted	BvVr…URcY (burn)	286	660,308
onaizahparisnajd	4KZT…HerD (burn)	315	654,837
redacted	BvVr…URcY (burn)	299	650,036
redacted	BvVr…URcY (burn)	293	648,020
onaizahparisnajd	4KZT…HerD (burn)	301	647,437
redacted	BvVr…URcY (burn)	294	642,916
onaizahparisnajd	4KZT…HerD (burn)	291	638,567
onaizahparisnajd	4KZT…HerD (burn)	298	632,122
addu	4gff…ksBn (burn)	308	628,534
onaizahs-kids	4KZT…HerD (burn)	271	617,561
onaizahskdis	4KZT…HerD (burn)	301	613,835
onaizahparisnajd	4KZT…HerD (burn)	308	613,439
wei001	9ptc…3LbM (burn)	198	607,746
callme	H5y3…BR5q (burn)	305	601,084
onaizahparisnajd	4KZT…HerD (burn)	305	593,401
onaizahparisnajd	4KZT…HerD (burn)	299	590,157
9ballshooter	APM9…89Nm (burn)	241	589,726
onaizahparisnajd	4KZT…HerD (burn)	296	586,169
onaizahparisnajd	4KZT…HerD (burn)	295	584,413
gummy	8zqX…Gv6M (burn)	308	578,724
gg	4gff…ksBn (burn)	300	578,301
redacted	BvVr…URcY (burn)	297	571,411
goatishere	58pZ…waHQ (burn)	304	571,280
redacted	BvVr…URcY (burn)	323	570,936
redacted	BvVr…URcY (burn)	301	565,677
bimboboy	58pZ…waHQ (burn)	298	562,195
gg	4gff…ksBn (burn)	306	560,534
wei008	9ptc…3LbM (burn)	229	556,394
denn	CQFL…PBVr (burn)	318	554,954
gg	4gff…ksBn (burn)	297	547,969
onaizahparisnajd	4KZT…HerD (burn)	279	546,142
onaizahparisnajd	4KZT…HerD (burn)	245	540,650
redacted	BvVr…URcY (burn)	302	538,080
onaizahparisnajd	4KZT…HerD (burn)	306	537,684
kaiyo	4gff…ksBn (burn)	295	529,359
onaizahparisnajd	4KZT…HerD (burn)	287	519,828
redacted	BvVr…URcY (burn)	300	517,834
onaizahparisnajd	4KZT…HerD (burn)	287	499,311
denn2	CQFL…PBVr (burn)	223	484,754
redacted	BvVr…URcY (burn)	209	482,906
history	G8zB…gepY (burn)	234	473,438
wei009	9ptc…3LbM (burn)	214	467,048
widowmaker	GRjD…spY8 (burn)	237	466,377
neet1	8X78…cmmd (burn)	203	448,475
wakeme	H5y3…BR5q (burn)	240	440,659
wei001	9ptc…3LbM (burn)	226	417,907
wei004	9ptc…3LbM (burn)	200	414,214
redacted	BvVr…URcY (burn)	223	398,578
wei002	9ptc…3LbM (burn)	192	395,936
laquisha	3v2K…GTgA (burn)	241	387,955
wei005	9ptc…3LbM (burn)	207	384,339
charl0tte	Ftgb…9DWW (burn)	182	367,684
wei010	9ptc…3LbM (burn)	207	350,115
wei006	9ptc…3LbM (burn)	178	309,394
wei007	9ptc…3LbM (burn)	174	278,748
neet2	8X78…cmmd (burn)	185	244,963
snaiduwa	4KZT…HerD (burn)	30	59,816
snaiduwa	4KZT…HerD (burn)	11	7,868
the queen's own crawlers (10)	—	7,364	15,115,933

Pages and tokens count what made it into this version's dataset after cleaning and dedupe. Machine-readable: contributors.json. The 45,627 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: 45,627 pages, about 10,950,432 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 (BTC) je skratka računalnom šifrev in spletne i zaščitne svoje denarja, ki so se na ključe strežnim denarjem. Bitcoin je zaščit neboli računalnom šifrev in spletne i zaščitne svoje denarja (a nekom jer javne stroške vrstniški) in poslovnega svo

Who created Bitcoin?

Bitcoin was created by Satoshi Nakamoto, a pseudonym used by the pseudonymous creator of the cryptocurrency. Nakamoto also published the Bitcoin whitepaper in 2008.

What is a blockchain?

A blockchain is a distributed database of all digital assets that can be accessed and verified by anyone in the world to be globally distributed. It is a secure, transparent, and decentralized database that stores all digital assets in a decentralized database.

Explain proof of work.

↑ "What are "algorithmic" cryptocurrencies?". Investopedia. Retrieved 2025-08-12. ↑ "Bitcoin - Economics - Bitcoin". bitcoin.org (in Engels).

Explain proof of stake.

In proof of stake, the validators that take turns producing blocks are called validators. In short, the validators that take turns producing blocks are called validators. In some PoS systems, validators are selected based on the amount of cryptocurrency they have staked. In the example given above, the validators can be selected based on the amount of cryptocurrency they are staking. For example, if there are 500,000 validators, they are selected based on the amount of cryptocurrency they stake.

What is Ethereum?

Ethereum – kurz ETH (jenis 100.000.000 ETH) – kurz ETH (jenis 100.000.000) – kurz ETH (jenis 100.000.000 ETH) – kurz ETH (jenis 100.000.000) – kurz ETH (jenis 100.000.000) – kurz ETH (jenis 100.000.000) – kurz ETH (jenis

What is a smart contract?

A smart contract is a program stored on the blockchain that executes when certain predefined conditions are met. It is the backbone of any self-executing computer program, and it is self-executing.

What is gas on Ethereum?

Gas is the unit that measures the computational cost of performing operations on Ethereum, including computational resources. Gas is a unit of measurement for the computational effort required to execute a transaction or smart contract.

What is Solana?

Solana is a decentralized computing platform that aims to improve blockchain scalability by using a consensus mechanism called proof of stake (PoS). It allows for low transaction costs and fast speeds. Solana is different from many other blockchains that the network has its own governance token, SOL, which can be bought and traded on various cryptocurrency exchanges.

Why is Solana fast?

Solana is fast. Solana is one of the fastest blockchain networks in crypto, offering speeds up transaction times and low fees. It is also scalable, with the ability to process around 65,000 transactions per second (TPS). It is also known for its low fees. Solana achieves this by using Proof of History (PoH) in its consensus mechanism. PoH is a cryptographically secure clock that timestamps transactions. Solana is fast with a block time of 400 milliseconds (i.e., a few seconds) and confirmation times of 300 milliseconds (i.e., a few

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.