whatmodel.sh / GPT-2 Large (774M)
GPT-2 Large (774M)
↗ share
+ contribute
GPT-2 Large (774M) is an open-weights language model from OpenAI, released August 2019. Its published size is 0.774B parameters and a 1k-token context window. Licence: mit.
Catalogue identity notes
OpenAI's released GPT-2 checkpoint as hosted at openai-community/gpt2-large (formerly the root-namespace 'gpt2-large' repo). Base (pretrained) model; parameter count as corrected in openai/gpt-2 README.
$ how it compares
The LLMs ranked either side of GPT-2 Large (774M), out of 828 with a score, then more from OpenAI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 823 |  Pythia 160M | 3±2 | local | ||
| 824 |  OPT-125M | 3±2 | local | ||
| 825 | GPT-2 Large (774M) | 1±2 | local | ||
| 826 | GPT-2 Medium (355M) | -1±2 | local | ||
| 827 | Pythia 70M | -7±2 | local | ||
| more from OpenAI | |||||
| 1 | GPT-6 Astra | 87±1 | api $20.00/M | ||
| 3 | GPT-6.1 Sol | 85±1 | api $4.00/M | ||
| 7 | GPT-6 Sol | 83±1 | api $4.00/M | ||
$ what it's good at
other tests14 not in a task · 98% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| OpenBookQA | 19.4% ✎ | █▁▁▁▄▃▃▇▂▁▄▂ beats 6/65 | 19% |
| ARC Challenge | 23.1% ✎ | ▆▁▁▂▂▅▇█▄▃▃▂ beats 0/97 | 18% |
| HellaSwag | 35.0% ✎ | ▁▁▃▁▂▂▃▇█▆▅▃ beats 2/85 | 13% |
| PROST physical reasoning (zero-shot, 10-concept macro average) | 26.4% ✎ | █···██·█···█ beats 2/4 | 13% |
| CommonsenseQA | 33.3% ✎ | █·····█·█··█ beats 2/3 | 10% |
| ARC Easy | 52.7% ✎ | ▁▂▃▂▃▅▇█▇▆▄▆ beats 12/65 | 9.0% |
| COPA | 69.8% ✎ | ▄·▄·█··▆▂··▂ beats 7/12 | 7.7% |
| Story Cloze Test | 66.1% ✎ | █····█···█·█ beats 2/3 | 6.1% |
6 more tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| COPEN concept identification in context (zero-shot, full prompt) | 33.8% ✎ | ▃···▃▃·▃·▃·█ beats 3/7 | 1.2% |
| EuroEval Life in the UK (English knowledge MC), test split, 5-shot, accuracy | 26.5% ✎ | ▁▁▃▁▁▁▁▂▄█▄▃ beats 3/146 | 0.9% |
| Open PT-LLM Leaderboard average (9 Portuguese tasks) | 27.4% ✎ | ▃▁▁▁▁▁▁▁▂█▅▁ beats 25/276 | 0.8% |
| EuroEval HellaSwag (English), test split, 5-shot generative MC, accuracy | 25.0% ✎ | ▃·▁▁▂▁▂▅█▇▄▁ beats 2/100 | 0.2% |
| COPEN conceptual similarity judgment (zero-shot, full prompt) | 11.3% ✎ | ▄···█···██·▄ beats 5/7 | 0.1% |
| COPEN conceptual property judgment (zero-shot, full prompt) | 47.2% ✎ | █·······▁··▁ beats 4/7 | 0.0% |
$ where to run it
llama.cpp · recommended
version
- Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.macOS / Linux
curl -LsSf https://llama.app/install.sh | shWindowsirm https://llama.app/install.ps1 | iex - Run GPT-2 Large (774M) · needs ~0.5 GB · from huggingface.co/mradermacherchat in the terminal
llama cli -hf mradermacher/gpt2-large-GGUF:Q4_K_M - Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf mradermacher/gpt2-large-GGUF:Q4_K_M
LM Studio
version
- Install LM Studio · Open the app once before using the lms command.
- Run GPT-2 Large (774M) · needs ~0.5 GB · from huggingface.co/mradermacher
- Or download it in the terminal
lms get https://huggingface.co/mradermacher/gpt2-large-GGUF@Q4_K_M -y - Then run lms ls to see its name, and lms chat followed by that name
Ollama
version
- Install Ollama
download it (macOS / Windows) ↗
Linuxcurl -fsSL https://ollama.com/install.sh | sh - Run GPT-2 Large (774M) · needs ~0.5 GB · from huggingface.co/mradermacherchat in the terminal
ollama run hf.co/mradermacher/gpt2-large-GGUF:Q4_K_M
MLX (Mac)
- Install MLX (Mac) · Apple silicon Macs only, text only.with uv
uv tool install mlx-lmorpip install mlx-lm - Run GPT-2 Large (774M) · needs ~1.5 GB, Mac only · from huggingface.co/MCES10chat in the terminal
mlx_lm.chat --model MCES10/gpt2-large-mlx-fp16 - Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model MCES10/gpt2-large-mlx-fp16
Other ways to run it
- Run it as published · needs ~3.25 GB · from huggingface.co/openai-community
uv run --with torch --with 'transformers<5' python -c "from transformers import pipeline; print(pipeline('text-generation', model='openai-community/gpt2-large')('Once upon a time', max_new_tokens=40)[0]['generated_text'])"
Weights and documentation: huggingface.co/openai-community ↗
$ how the score was worked out · Surface Form Competition (Hol… 82%, PROST (Aroca-Ouellette et al.… 13%, COPEN — EMNLP 2022 original z… 1% and 2 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 Surface Form Competition (Holtzman et al., EMNLP 2021) — authors' own GPT-2 runs | 7 | 82% |
| PROST (Aroca-Ouellette et al., Findings of ACL-IJCNLP 2021) - authors' own zero-shot runs | 1 | 13% |
| COPEN — EMNLP 2022 original zero-shot evaluations | 3 | 1% |
 EuroEval leaderboards (results collection), English generative leaderboard | 2 | 1% |
 Open Portuguese LLM Leaderboard (archived) | 1 | 1% |
| prior · pull toward the middle | 2% | |
| weighted average | = 0.7 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json