gpt2-large

812M parameters, text in, text out, 1K tokens of context.

Verified 2026-10-26 07:27 UTC Rank #353 in LLMs · score 81.2 method v1 Sources 3 Downloads 30d 793.6k · #132 of 619 tracked

Changed this week: 1 new quantized build on the Hub (15 in all).2026-10-24

Can I run it?0.5GB at INT4Estimatefits a 16 GB cardmethod

Can I use it?MITPermissivelicence file2026-10-26

Do people use it?793.6kdownloads, 30 days#132 of 619 we track · top 21%+2 likes, downloads −12% this week371 likesHugging Face2026-10-26

Worth reading

Picked by rule from 2 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Most discussed · 145 HN pts

Salesforce releases language model bigger than GPT-2 large

Salesforce releases CTRL, a 1.6 billion-parameter conditional transformer language model designed for controllable text generation.

A 36-layer model variant exists but produces markedly worse generation quality than the standard 48-layer base model.
Inference on PyTorch requires the model path to point to a .data file rather than the standard checkpoint folder.

github.com 2019-09-12 145 HN pts · 40 comments 2.9k words

2 best of 2 tracked

The 1 pick above, plus these 1. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleWrite with Transformertransformer.huggingface.co · 2019-12-24 · 1 HN pts

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: 15 quantized builds and 100 Spaces.

As a service0no API provider listed
On own hardware15quantized builds
Built on190120 fine-tunes · 70 adapters · 0 merges
In apps100+Hugging Face Spaces

Top Spaces running it: IndexTeam/IndexTTS-2-Demo ♥ 849 · pliny-the-prompter/obliteratus ♥ 621 · shi-labs/Versatile-Diffusion ♥ 408 · exbert-project/exbert ♥ 187

Will it fit?

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

Fits a 16 GB card at INT4

About 0.5 GB at INT4, 1.0 GB at FP8, 1.9 GB at BF16: weights only, so leave room for the KV cache.

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

The story so far

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

2 dated articles and 23 repo commits so far

Teal is the weights (2022-03-02), amber triangles are 2 dated articles (each links to it), grey ticks are repo commits.

What people built on it

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

205 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.

GGUF mradermacher/gpt2-large-i1-GGUF for llama.cpp, Ollama, LM Studio 14.3k downloads MLX MCES10/gpt2-large-mlx-fp16 for Apple silicon 672 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.

Benchmark	Value	Setting / notes	Reported by	Date
author-reported 1BW	44.58 PPL	zero-shot	model author	2026-10-07
author-reported CBT-CN	93.45 ACC	zero-shot	model author	2026-10-07
author-reported CBT-NE	88 ACC	zero-shot	model author	2026-10-07
author-reported LAMBADA	10.87 PPL	zero-shot	model author	2026-10-07
author-reported LAMBADA	60.12 ACC	zero-shot	model author	2026-10-07
author-reported PTB	40.31 PPL	zero-shot	model author	2026-10-07
author-reported WikiText103	22.05 PPL	zero-shot	model author	2026-10-07
author-reported WikiText2	19.93 PPL	zero-shot	model author	2026-10-07
author-reported enwiki8	0.97 BPB	zero-shot	model author	2026-10-07
author-reported text8	1.02 BPC	zero-shot	model author	2026-10-07

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.

Run on a laptop or phonestructuraldecision-model estimate p=0.82 fits: edge

Facts

20 fields

Listed in LLMs

Facts

gpt2-large has 812M parameters. (safetensors metadata, observed 2026-10-26)
Context length is 1,024 tokens (1K). (config.json max_position_embeddings, observed 2026-10-26)
License: MIT (friction: none). Permissive. (license file, observed 2026-10-26)
Created 2022-03-02, last modified 2024-02-19. (HF API, observed 2026-10-26)

Pipeline tag	text-generation
Library	transformers
Architectures	GPT2LMHeadModel
Model type	gpt2
Params by dtype	F32: 811,778,816
Size band	S
MoE	no
Context length	1,024 (config.json max_position_embeddings)
Languages	en
Papers	arXiv:1910.09700
License	MIT (mit)
Hub snapshot	2026-10-26

Servability

vLLM · SGLang

Servability

vLLM: supported · GPT2LMHeadModel · 4b1c04e76fe9ac3ca8b7e541dd423e7d5f38dc60SGLang: supported · GPT2LMHeadModel · 60d941293e3af27a12c6eb64e2c3e2e1a17f100f

vLLM registry (4b1c04e76fe9ac3ca8b7e541dd423e7d5f38dc60) · SGLang registry (60d941293e3af27a12c6eb64e2c3e2e1a17f100f) · 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-large: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	1.9 GB · Edge (<4 GB)
FP8 (1 byte/param)	1.0 GB · Edge (<4 GB)
INT4 (0.5 bytes/param)	0.5 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-large 205 derivatives on the Hub

finetune: 120quantized: 15adapter: 70

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
nicholasKluge/Aira-2-774M	finetune	—	774M	S	apache-2.0	nicholasKluge	3	24k
Mikivis/gpt2-large-lora-sft	finetune	—	774M	S	mit	Mikivis	2	23.9k
mradermacher/gpt2-large-i1-GGUF	quantized	GGUF	—	—	mit	mradermacher	0	14.3k
mradermacher/gpt2-large-GGUF	quantized	GGUF	—	—	mit	mradermacher	1	10k
DevQuasar-10/openai-community.gpt2-large-GGUF	quantized	GGUF	—	—	—	DevQuasar-10	0	4.1k
Alcoft/gpt2-large-GGUF	quantized	GGUF	—	—	mit	Alcoft	0	3.6k
augustocsc/gpt2_large_infix_682k	adapter	—	—	—	—	augustocsc	0	3.6k
augustocsc/gpt2_large_prefix_682k	adapter	—	—	—	—	augustocsc	0	3.4k
ma921/gpt2-large-sft-imdb	finetune	—	774M	S	mit	ma921	0	3.4k
payelb/GPT2L_full	finetune	—	774M	S	mit	payelb	0	2.9k
tensorblock/gpt2-large-GGUF	quantized	GGUF	—	—	mit	tensorblock	0	2.6k
MiniLLM/MiniLLM-gpt2-760M	finetune	—	—	—	apache-2.0	MiniLLM	0	1.3k
mario-rc/emotional-gpt2-large	finetune	—	774M	S	mit	mario-rc	0	1.2k
zeromodels/gpt2_large	finetune	—	—	—	mit	zeromodels	0	1.2k
schaeff/gpt2-large_LNFree600	finetune	—	774M	S	other	schaeff	0	863
course-genai-w24/week4-gpt2-lora-sft-tldr	adapter	—	—	—	—	course-genai-w24	0	722
MCES10/gpt2-large-mlx-fp16	finetune	mlx	774M	S	mit	MCES10	0	672
cwestnedge/gpt2-large-pubmed	finetune	—	774M	S	mit	cwestnedge	0	611
BroAlanTaps/Stage1-PCC-Lite-4x	finetune	—	774M	S	apache-2.0	BroAlanTaps	0	594
MiniLLM/SFT-gpt2-760M	finetune	—	774M	S	apache-2.0	MiniLLM	0	550
TahaCakir/enhanced_turkishReviews-generativeAI	finetune	—	—	—	mit	TahaCakir	0	517
MiniLLM/SeqKD-gpt2-760M	finetune	—	—	—	apache-2.0	MiniLLM	0	437
jdumanski/gpt2-large-Q4_K_M-GGUF	quantized	GGUF	—	—	mit	jdumanski	0	404
jncraton/LaMini-GPT-774M-ct2-int8	finetune	int8	—	—	cc-by-nc-4.0	jncraton	0	383
MiniLLM/KD-gpt2-760M	finetune	—	—	—	apache-2.0	MiniLLM	0	375
MiniLLM/init-gpt2-760M	finetune	—	—	—	apache-2.0	MiniLLM	0	359
jhonparra18/gpt2-large-sft-chat-eng-20ep-gguf	quantized	gguf	—	—	apache-2.0	jhonparra18	0	324
mhdang/gelatwo-common-gen-gpt2-large	finetune	—	774M	S	mit	mhdang	0	319
MagistrTheOne/RadonSAI	finetune	—	774M	S	apache-2.0	MagistrTheOne	0	300
varun-v-rao/gpt2-large-squad-model3	finetune	—	774M	S	mit	varun-v-rao	0	293
netrias/cancer_harmonization_gpt2large	finetune	—	774M	S	apache-2.0	netrias	0	283
varun-v-rao/gpt2-large-snli-model1	finetune	—	774M	S	mit	varun-v-rao	0	276
Holarissun/RM-HH-AllMixNonPeft_harmless_gpt3_20000_gpt2-large_shuffleFalse_extractchosenTrue	finetune	—	774M	S	mit	Holarissun	0	267
Holarissun/RM-HH-AllMixNonPeft_harmless_gpt3_20000_gpt2-large_shuffleFalse_extractchosenFalse	finetune	—	774M	S	mit	Holarissun	0	261
varun-v-rao/gpt2-large-snli-model3	finetune	—	774M	S	mit	varun-v-rao	0	257
schaeff/gpt2-large_vanilla600	finetune	—	774M	S	other	schaeff	0	256
netrias/alcohol100_bacteria0_harmonization_gpt2large	finetune	—	774M	S	apache-2.0	netrias	0	255
varun-v-rao/gpt2-large-snli-model2	finetune	—	774M	S	mit	varun-v-rao	0	249
gumran/gpt2-large-sft	finetune	—	774M	S	mit	gumran	0	246
AshishKJain/gpt2-large-alpaca-sft	finetune	—	838M	S	mit	AshishKJain	2	128

Repository changelog

23 commits

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

Repository changelog 23 commits

Date	Commit	SHA
2024-02-19	Adds the tokenizer configuration file (#7)	32b71b1
2023-06-30	Adding ONNX file of this model (#6)	97935fc
2023-02-22	Adding `safetensors` variant of this model (#3)	212095d
2023-01-24	Adding generation config file(s)	b54dbc3
2022-07-22	Add model card (#1)	e5ab12c
2021-09-08	Update config.json	897ad54
2021-09-08	Update config.json	1bd2ca5
2021-05-21	upload flax model	8a79112
2021-05-21	allow flax	6edb598
2020-12-11	Migrate model card from transformers-repo	d7b8769
2020-12-09	For clarity, delete deprecated modelcard.json	3673adc
2020-10-12	Update tokenizer.json	d54890b
2020-05-11	Update config.json	3779e12
2020-05-02	Update rust_model.ot	98dd51b
2020-04-24	Update config.json	5b48370
2020-01-31	Update config.json	6934902
2019-12-20	Update modelcard.json	934da44
2019-09-23	Update tf_model.h5	3c0ebb8
2019-08-21	Update config.json	a1a6494
2019-08-21	Update merges.txt	277e1e3
2019-08-21	Update pytorch_model.bin	7c42711
2019-08-21	Update vocab.json	b4787c5
2019-08-21	initial commit	dece569

Articles and coverage

2 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.

Analysis2

Salesforce releases language model bigger than GPT-2 large

github.com2019-09-12authorityHN front pageHN 145 pts · 40 comments2,893 words

Salesforce releases CTRL, a 1.6 billion-parameter conditional transformer language model designed for controllable text generation. The model uses control codes to specify domains, entities, and relationships, allowing users to guide output toward specific topics like 'Links' or 'Wikipedia'. It is released under the BSD-3 License and requires TensorFlow 1.14 for inference.written, reviewed, cited

A 36-layer model variant exists but produces markedly worse generation quality than the standard 48-layer base model.
Inference on PyTorch requires the model path to point to a .data file rather than the standard checkpoint folder.
Users must patch a specific TensorFlow file to enable custom functionality, and skipping this step causes errors.

Write with Transformer

transformer.huggingface.co2019-12-24HN 1 pts · 0 comments75 words

Papers

1 paper

Papers 1

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

Quantifying the Carbon Emissions of Machine Learning cited in card

Alexandre Lacoste, Alexandra Luccioni, Victor Schmidt +1 · 2019-10-21 · PDF · HF paper page (61 upvotes)

Similar models

3 models

Similar models

Qwen3-0.6B · 752M
same size band (S), same task (text-generation), more providers, permissive license
Qwen3Guard-Gen-0.6B · 752M
same size band (S), same task (text-generation), more providers, permissive license
bitnet-b1.58-2B-4T · 850M
same size band (S), same task (text-generation), permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-26 · HF API: 2026-10-26 07:20 UTC · Providers: 2026-10-26 07:26 UTC · Articles: 2026-10-26 07:27 UTC · Summary: 2026-10-07 15:09 UTC · Last verified 2026-10-26 07:27 UTC

Follow gpt2-large

Watch this model

/feeds/models/openai-community--gpt2-large.xml(RSS 2.0) · every change to gpt2-large
/feeds/families/openai-community--gpt2-large.xml(RSS 2.0) · every change across the gpt2-large family

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