Model Follower.

Up to date

Rebuilt
2026-10-25 06:54 UTC
Release radar
117 on watch
Hub snapshot
2026-10-25
Re-checked
daily · every number sourced and dated

Models › gpt2 family

gpt2

Hugging Face ↗ Follow changes

openai-community MIT out 5 years ago

137M parameters, text in, text out, 1K tokens of context.

Verified 2026-10-25 Rank #95 in LLMs · score 99.5 method v1 Sources 3 Downloads 30d 20.1M · #5 of 619 tracked

Changed this week: 1 new quantized build on the Hub (100 in all).2026-10-24

Can I run it?0.1GB at INT4Estimatefits a 16 GB cardmethod
Can I use it?MITPermissivelicence file2026-10-24
Do people use it?20.1Mdownloads, 30 days#5 of 619 we track · top 1%−369 likes, downloads +21% this week3,891 likesHugging Face2026-10-24

Worth reading

Picked by rule from 8 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Field report

Talking to myself: how I trained GPT2-1.5b for rubber ducking using my chat data

An engineer trained a GPT2-1.5b model on 14mb of personal Facebook Messenger data to create a conversational clone for self-analysis and debugging.

  • The training setup requires changing the runtime type to TPU to train the whole model on Google Colab.
  • The training process required restarting the model twice due to sensitive subjects appearing in the conversation.
svilentodorov.xyz 2020-01-23 217 HN pts · 67 comments 3.4k words
Latest · 7 months ago

OpenAI says its new model GPT-2 is too dangerous to release (2019)

On Feb.

  • OpenAI withheld the full algorithm and training codes despite releasing a much smaller version of the model.
  • The model distinguishes between multiple definitions of a single word based on context clues and has deeper knowledge of obscure usages.
slate.com 2026-04-08 395 HN pts · 120 comments 1.7k words

8 best of 8 tracked

The 4 picks above, plus these 4. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

  1. CodeShow HN: GPT-2 implemented using graphics shadersgithub.com · 2025-05-02 · 228 HN pts
  2. ArticleShow HN: Fully client-side GPT2 prediction visualizerperplexity.vercel.app · 2023-09-05 · 153 HN pts
  3. ArticleThe Illustrated GPT-2: Visualizing Transformer Language Models (2019)jalammar.github.io · 2023-12-19 · 213 HN pts
  4. ArticleGPT-2: A Deep Divesushant-kumar.com · 2024-03-22

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

No API provider serves it, so its downloads are people running it themselves: 100 quantized builds and 100 Spaces.

  • As a service0no API provider listed
  • On own hardware100quantized builds
  • Built on4,0472,299 fine-tunes · 1,737 adapters · 11 merges
  • In apps100+Hugging Face Spaces

Top Spaces running it: microsoft/HuggingGPT ♥ 2,272 · Gustavosta/MagicPrompt-Stable-Diffusion ♥ 2,079 · IndexTeam/IndexTTS-2-Demo ♥ 849 · fffiloni/LatentSync ♥ 630

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 0.1 GB at INT4, 0.2 GB at FP8, 0.3 GB at BF16: weights only, so leave room for the KV cache.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 0.1 GB of weights (estimate; KV cache and activations extra)INT40.1 GBFP8: about 0.2 GB of weights (estimate; KV cache and activations extra)FP80.2 GBBF16: about 0.3 GB of weights (estimate; KV cache and activations extra)BF160.3 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 0.1 GB of weights (estimate; KV cache and activations extra)INT40.1 GBFP8: about 0.2 GB of weights (estimate; KV cache and activations extra)FP80.2 GBBF16: about 0.3 GB of weights (estimate; KV cache and activations extra)BF160.3 GB

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

8 dated articles and 26 repo commits so far

Teal is the weights (2022-03-02), amber triangles are 8 dated articles (each links to it), grey ticks are repo commits.

MarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2024-02-19: 1 repo commit2023-06-30: 1 repo commit2022-12-16: 1 repo commit2022-11-23: 1 repo commit2022-11-18: 1 repo commit2022-10-20: 1 repo commit2022-09-29: 1 repo commit2021-05-19: 2 repo commits2020-12-11: 1 repo commit2020-12-09: 1 repo commit2020-10-12: 1 repo commit2020-05-11: 1 repo commit2020-04-24: 2 repo commits2020-01-31: 1 repo commit2019-12-20: 1 repo commit2019-12-12: 2 repo commits2019-11-29: 1 repo commit2019-09-23: 1 repo commit2019-02-18: 5 repo commitsOpenAI says its new model GPT-2 is too dangerous to release (2019) · 2026-04-08Bumblebee: GPT2, Stable Diffusion, and More in Elixir · 2022-12-08Show HN: GPT-2 implemented using graphics shaders · 2025-05-02Try to guess if code is real or GPT2-generated · 2021-02-23Talking to myself: how I trained GPT2-1.5b for rubber ducking using my chat data · 2020-01-23Show HN: Fully client-side GPT2 prediction visualizer · 2023-09-05The Illustrated GPT-2: Visualizing Transformer Language Models (2019) · 2023-12-19GPT-2: A Deep Dive · 2024-03-22Weights 2022-03-02 · 2022-03-02Weights 2022-03-02now
MarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOct2024-02-19: 1 repo commit2023-06-30: 1 repo commit2022-12-16: 1 repo commit2022-11-23: 1 repo commit2022-11-18: 1 repo commit2022-10-20: 1 repo commit2022-09-29: 1 repo commit2021-05-19: 2 repo commits2020-12-11: 1 repo commit2020-12-09: 1 repo commit2020-10-12: 1 repo commit2020-05-11: 1 repo commit2020-04-24: 2 repo commits2020-01-31: 1 repo commit2019-12-20: 1 repo commit2019-12-12: 2 repo commits2019-11-29: 1 repo commit2019-09-23: 1 repo commit2019-02-18: 5 repo commitsOpenAI says its new model GPT-2 is too dangerous to release (2019) · 2026-04-08Bumblebee: GPT2, Stable Diffusion, and More in Elixir · 2022-12-08Show HN: GPT-2 implemented using graphics shaders · 2025-05-02Try to guess if code is real or GPT2-generated · 2021-02-23Talking to myself: how I trained GPT2-1.5b for rubber ducking using my chat data · 2020-01-23Show HN: Fully client-side GPT2 prediction visualizer · 2023-09-05The Illustrated GPT-2: Visualizing Transformer Language Models (2019) · 2023-12-19GPT-2: A Deep Dive · 2024-03-22Weights 2022-03-02 · 2022-03-02Weights 2022-03-02now

What people built on it

4,147 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

4,147 derivatives on the Hub, most of them fine-tunes

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

100 quantized (Hugging Face model tree)2,299 fine-tunes (Hugging Face model tree)11 merges (Hugging Face model tree)1,737 adapters (Hugging Face model tree)100 quantized2,299 fine-tunes11 merges1,737 adapters
100 quantized (Hugging Face model tree)2,299 fine-tunes (Hugging Face model tree)11 merges (Hugging Face model tree)1,737 adapters (Hugging Face model tree)100 quantized2,299 fine-tunes11 merges1,737 adapters
GGUF PrunaAI/gpt2-GGUF-smashed for llama.cpp, Ollama, LM Studio 73.7k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

10 rows

Benchmarks 10 rows

measured third-party leaderboard rowauthor-reported number from the model card or eval files (10)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.
BenchmarkValueSetting / notesReported byDate
author-reported 1BW75.2 PPL—model author2026-10-07
author-reported CBT-CN87.65 ACC—model author2026-10-07
author-reported CBT-NE83.4 ACC—model author2026-10-07
author-reported LAMBADA35.13 PPL—model author2026-10-07
author-reported LAMBADA45.99 ACC—model author2026-10-07
author-reported PTB65.85 PPL—model author2026-10-07
author-reported WikiText10337.5 PPL—model author2026-10-07
author-reported WikiText229.41 PPL—model author2026-10-07
author-reported enwiki81.16 BPB—model author2026-10-07
author-reported text81,17 BPC—model author2026-10-07

What the card says

1 cited sentence

Summary generated, cited

The model is pretrained on English language data using a causal language modeling objective.[1]

  1. “Pretrained model on English language using a causal langu…”

Generated 2026-10-07 15:00 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

1 use case

What it is good at 1 use case

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

19 fields

Listed in LLMs

Facts

Pipeline tagtext-generation
Librarytransformers
ArchitecturesGPT2LMHeadModel
Model typegpt2
Params by dtypeF32: 137,022,720
Size bandS
MoEno
Context length1,024 (config.json max_position_embeddings)
Languagesen
LicenseMIT (mit)
Hub snapshot2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · GPT2LMHeadModel · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · GPT2LMHeadModel · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for gpt2: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)0.3 GB · Edge (<4 GB)
FP8 (1 byte/param)0.2 GB · Edge (<4 GB)
INT4 (0.5 bytes/param)0.1 GB · Edge (<4 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

40 variants

Lineage

Original model Family: openai-community/gpt2 4,147 derivatives on the Hub

finetune: 2,299adapter: 1,737quantized: 100merge: 11

Top variants (40 shown, ranked by downloads)

Relation
40 rows
Variant Relation Quant Params Band License Author Likes Downloads
garak-llm/attackgeneration-toxicity_gpt2 finetune — 124M S apache-2.0 garak-llm 7 404.4k
peft-internal-testing/gpt2-lora-random adapter — — — — peft-internal-testing 0 368.6k
cwestnedge/gpt2-small-pubmed finetune — 124M S mit cwestnedge 0 116.4k
Xenova/gpt2 quantized — — — — Xenova 12 115.4k
PrunaAI/gpt2-GGUF-smashed quantized GGUF — — — PrunaAI 5 73.7k
manupande21/GPT2_PMC finetune — 124M S mit manupande21 1 64.1k
11mlabs/indri-0.1-124m-tts finetune — 140M S cc-by-sa-4.0 11mlabs 12 48.8k
jackoyoungblood/TinyStoriesProject finetune — 124M S mit jackoyoungblood 0 45k
alibidaran/medical_transcription_generator finetune — 137M S — alibidaran 5 40.2k
phonemetransformers/GPT2-85M-BPE-TXT finetune — 97.5M S — phonemetransformers 1 35.6k
akari000/gpt-2-artificial-vss-40 finetune — 124M S mit akari000 0 32.4k
PrunaAI/openai-community-gpt2-GGUF-smashed quantized GGUF — — — PrunaAI 1 31.6k
QuantFactory/gpt2-GGUF quantized GGUF — — mit QuantFactory 13 23.8k
nicholasKluge/Aira-2-124M finetune — 124M S apache-2.0 nicholasKluge 1 23.7k
Mikivis/xuanxuan finetune — 137M S mit Mikivis 0 23.5k
nikokons/gpt2-greek finetune — — — mit nikokons 2 18.7k
theabdulsaboor/gpt2-therapist-finetuned finetune — 124M S — theabdulsaboor 1 18.1k
mradermacher/gpt2-i1-GGUF quantized GGUF — — mit mradermacher 0 18.1k
augustocsc/gpt2_base_infix_682k adapter — — — — augustocsc 0 16.4k
augustocsc/gpt2_base_prefix_682k adapter — — — — augustocsc 0 15.1k
mradermacher/gpt2-GGUF quantized GGUF — — mit mradermacher 0 13.9k
Melady/TEMPO finetune — — — apache-2.0 Melady 12 12.5k
IntelLabs/aac_gpt2 finetune — — — mit IntelLabs 2 12.9k
diffusionfamily/diffugpt-s finetune — 124M S apache-2.0 diffusionfamily 4 12k
kpriyanshu256/gpt-ya2-v2 finetune — — — mit kpriyanshu256 0 12.2k
joaosollatori/small-agentic-model-1.5B finetune — 124M S mit joaosollatori 1 11.5k
imda-biztech/toxic_sentence_generator finetune — 124M S apache-2.0 imda-biztech 0 10.9k
11mlabs/indri-0.1-350m-tts finetune — 375M S cc-by-sa-4.0 11mlabs 3 9.1k
Tribewarez/psy-q-finder-369M finetune — 370M S mit Tribewarez 1 9.1k
Arjun-G-Ravi/chat-GPT2 finetune — 124M S mit Arjun-G-Ravi 6 8.7k
OpceanAI/Yuuki-NxG-nano finetune — 81.9M S apache-2.0 OpceanAI 3 8.8k
OpceanAI/Yuuki-best quantized GGUF 81.9M S apache-2.0 OpceanAI 3 8.6k
MiniLLM/MiniLLM-gpt2-120M finetune — 124M S apache-2.0 MiniLLM 0 8.6k
Lukamac/PlayPart-AI-Personal-Trainer finetune — 124M S apache-2.0 Lukamac 3 8.2k
riturajpandey739/gpt2-sentiment-analysis-tweets finetune — 124M S mit riturajpandey739 0 7.9k
OpceanAI/Yuuki-3.7 quantized GGUF 81.9M S apache-2.0 OpceanAI 3 7.6k
MiniLLM/init-gpt2-120M finetune — 124M S apache-2.0 MiniLLM 1 7.1k
jarod0411/zinc10M_gpt2_SMILES_bpe_combined_step1 finetune — 163M S mit jarod0411 0 7.1k
MikeMpapa/8_bar_lmd_clean_custom_final_token_scheme_epochs10 finetune — 35.3M S mit MikeMpapa 0 6.6k
jkawamoto/gpt2-ct2 quantized — — — mit jkawamoto 0 5.9k

Repository changelog

26 commits
changed in the last 7 daysOther: 2026-10-24all changes

Repository changelog 26 commits

DateCommitSHA
2024-02-19 Adds the tokenizer configuration file (#80) 607a30d
2023-06-30 Adding ONNX file of this model (#60) 11c5a3d
2022-12-16 Update generation_config.json e7da7f2
2022-11-23 Add note that this is the smallest version of the model (#18) f27b190
2022-11-18 Our very first generation_config.json (#17) 0dd7bcc
2022-10-20 Upload model.safetensors with huggingface_hub (#12) 75e09b4
2022-09-29 Convert weights to .safetensors (#6) 909a290
2021-05-19 add gpt2 6c0e608
2021-05-19 Update .gitattributes aff9832
2020-12-11 Migrate model card from transformers-repo 6a8c602
2020-12-09 For clarity, delete deprecated modelcard.json 84d42a0
2020-10-12 Update tokenizer.json 210ab4c
2020-05-11 Update config.json 7979bb7
2020-04-24 Update rust_model.ot ee92bbb
2020-04-24 Update config.json 26b7c7a
2020-01-31 Update config.json 8bfcadd
2019-12-20 Update modelcard.json 99dd285
2019-12-12 Update 64-8bits.tflite 8faa6d7
2019-12-12 Update 64-fp16.tflite 5d494dc
2019-11-29 Update 64.tflite b57efa2
2019-09-23 Update tf_model.h5 ddceb78
2019-02-18 Update config.json 703af34
2019-02-18 Update vocab.json 9228726
2019-02-18 Update merges.txt a9eff68
2019-02-18 Update pytorch_model.bin a49587d
2019-02-18 initial commit 9b865ef

Articles and coverage

8 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Analysis8

OpenAI says its new model GPT-2 is too dangerous to release (2019)
slate.com2026-04-08HN front pageHN 395 pts · 120 comments1,704 words
On Feb. 14, OpenAI announced a text-generation model trained using text from 8 million webpages that can adapt to the style and content of conditioning text. The organization decided to withhold the full algorithm and training codes due to concerns that it could be used to generate false news articles or impersonate people online. Experts note that an entity with enough capital could build a comparable generator by renting servers from Amazon Web Services.written, reviewed, cited
  • OpenAI withheld the full algorithm and training codes despite releasing a much smaller version of the model.
  • The model distinguishes between multiple definitions of a single word based on context clues and has deeper knowledge of obscure usages.
Bumblebee: GPT2, Stable Diffusion, and More in Elixir
news.livebook.dev2022-12-08HN front pageHN 550 pts · 72 comments
Show HN: GPT-2 implemented using graphics shaders
github.com2025-05-02authorityHN front pageHN 228 pts · 25 comments169 words
The project implements a full GPT-2 small (117M) forward pass in the GPU via WebGL2 shaders. BPE tokenization uses js-tiktoken in the browser without a WASM fetch.generated, cited
Try to guess if code is real or GPT2-generated
doesnotexist.codes2021-02-23HN front pageHN 201 pts · 103 comments
Talking to myself: how I trained GPT2-1.5b for rubber ducking using my chat data
svilentodorov.xyz2020-01-23HN front pageHN 217 pts · 67 comments3,394 words
An engineer trained a GPT2-1.5b model on 14mb of personal Facebook Messenger data to create a conversational clone for self-analysis and debugging. The training utilized a Google Colab TPU and reached an end loss of just over 0.6 after taking over a month to complete. A model checkpoint requires approximately 6 gb of storage, though paying for Google One simplifies fitting the data to Drive.written, reviewed, cited
  • The training setup requires changing the runtime type to TPU to train the whole model on Google Colab.
  • The training process required restarting the model twice due to sensitive subjects appearing in the conversation.
Show HN: Fully client-side GPT2 prediction visualizer
perplexity.vercel.app2023-09-05HN front pageHN 153 pts · 11 comments304 words
The Illustrated GPT-2: Visualizing Transformer Language Models (2019)
jalammar.github.io2023-12-19HN front pageHN 213 pts · 5 comments5,011 words
GPT-2: A Deep Dive
sushant-kumar.com2024-03-22361 words

Similar models

3 models

Similar models

  • SmolLM2-135M · 135M
    same size band (S), same task (text-generation), permissive license
  • SmolLM-135M · 135M
    same size band (S), same task (text-generation), permissive license
  • gpt-neo-125m · 150M
    same size band (S), same task (text-generation), permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:33 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:00 UTC · Last verified 2026-10-25

Follow gpt2

Watch this model

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.