whatmodel.sh / Pythia 12B

Pythia 12B

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Pythia 12B is an open-weights language model from EleutherAI, released February 2023. Its published size is 11.8B parameters and a 2k-token context window. Licence: apache-2.0.

Catalogue identity notes

Final (step 143000) checkpoint of the standard (non-deduplicated) Pythia model, repo main branch. Total parameter count from the Pythia model card.

24.9±0.8#672 of 719 LLMsfrom 19 benchmark results across 4 sources

$ how it compares

The LLMs ranked either side of Pythia 12B, out of 828 with a score, then more from EleutherAI.

#modelscorewhereRAMreleased
775Falcon 7B Instruct TII25±2local~5.9GB2023-04
776Qwen2 0.5B Instruct Alibaba25±2local~1.8GB2024-06
777Pythia 12B EleutherAI25±1local~9.5GB2023-02
778RedPajama-INCITE Base 3B v1 Together25±3local~3.4GB2023-05
779ChatGLM2-6B Zhipu AI (THUDM)25±2local~15GB2023-06
more from EleutherAI
756GPT-NeoX-20B EleutherAI28±1local~14GB2022-02
780GPT-J 6B EleutherAI25±1local~14GB2021-06
787Pythia 6.9B EleutherAI23±1local~6.1GB2023-02

see all LLMs · find the ones that fit your machine

$ what it's good at

other tests13 not in a task · 100% of its score
testresultvs the others in its rundifficultycountswhy
ARC Challenge +135.1% ✎▁·▁··▁▁▇▂█▆▃ beats 8/26easy · avg 85%▇▇▇▇▇▇▇▇▇▇▇▇▇ 33%sharp · several runs
WinoGrande +164.0% ✎▁····▁··▁▂█▄ beats 8/26easy · avg 89%▇▇▇▇▇▇ 14%at its level · several runs
ARC Easy +163.7% ✎▁▁·▁▁▄▅▄█▅▁▃ beats 11/26very easy · avg 90%▇▇▇▇▇ 12%at its level · several runs
SciQ +190.2% ✎▁·▁·▁·▃▅▃█▄▁ beats 8/26very easy · avg 98%▇▇▇ 8.4%several runs
PIQA +176.3% ✎▁·▁··▁▁▃█▅█▂ beats 11/26easy · avg 88%▇▇▇ 7.0%blunt · at its level · several runs
BoolQ67.3% ✎▂▃▂▅▆▃▅▆█▂▂▂ beats 11/26easy · avg 91%▇▇▇ 6.8%at its level
LAMBADA (OpenAI version), last-word accuracy +170.6% ✎▁··▂▁▁▁▂▆█▅▃ beats 15/26very easy · avg 89%▇▇ 5.0%blunt · at its level · several runs
HellaSwag67.4% ✎▃···▁▇▅█▇▃▁▁ beats 10/26very easy · avg 92%▇▇ 4.5%at its level
5 more tests
testresultvs the others in its rundifficultycountswhy
Factual knowledge recall, Premium2k exact match (zero-shot)15.3% ✎▅····▃▅·██·█ beats 4/13medium · avg 59%▇ 2.5%typical
Factual knowledge recall, Premium2k exact match (10-shot)23.2% ✎▃▅▃··▅▃·█▃▃▅ beats 3/13easy · avg 67%▇ 2.4%at its level
OpenBookQA26.4% ✎█·▁▁▃·▁▃▅▇▃▄ beats 9/26medium · avg 53%▇ 1.9%too hard for it
Social IQA42.0% ✎▁···▁▃█▁▁▁·▁ beats 12/26medium · avg 59%▇ 1.6%blunt
Open PT-LLM Leaderboard average (9 Portuguese tasks)34.0% ✎▃▁▃▁▁▁▁▁▂█▅▁ beats 39/276medium · avg 73%▇ 0.7%noisy

$ where to run it

llama.cpp · recommended

version

  1. 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 | sh
    Windows
    irm https://llama.app/install.ps1 | iex

    other ways to install ↗

  2. Run Pythia 12B · needs ~7.6 GB · from huggingface.co/mradermacher
    chat in the terminal
    llama cli -hf mradermacher/pythia-12b-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf mradermacher/pythia-12b-GGUF:Q4_K_M

LM Studio

version

  1. Install LM Studio · Open the app once before using the lms command.

    download it (macOS / Windows / Linux) ↗

  2. Run Pythia 12B · needs ~7.6 GB · from huggingface.co/mradermacher

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/mradermacher/pythia-12b-GGUF@Q4_K_M -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run Pythia 12B · needs ~7.6 GB · from huggingface.co/mradermacher
    chat in the terminal
    ollama run hf.co/mradermacher/pythia-12b-GGUF:Q4_K_M

Other ways to run it

  1. Run it as published · needs ~23.84 GB · from huggingface.co/EleutherAI (official)
    uv run --with torch --with 'transformers<5' --with accelerate python -c "from transformers import pipeline; print(pipeline('text-generation', model='EleutherAI/pythia-12b', torch_dtype='auto', device_map='auto')('Once upon a time', max_new_tokens=40)[0]['generated_text'])"

Official weights and documentation: huggingface.co/EleutherAI ↗

$ how the score was worked out · Stability AI — original Stabl… 64%, Pythia: A Suite for Analyzing… 30%, Towards a Holistic Evaluation… 5% and 1 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

sourceresultsshare of the scorelicence
Stability AI — original StableLM zero-shot evaluation outputs10▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 64%Apache-2.0
Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runs · lab-reported6▇▇▇▇▇▇▇▇▇ 30%CC BY-SA 4.0
Towards a Holistic Evaluation of LLMs on Factual Knowledge Recall (arXiv:2404.16164v1)2▇ 5%CC-BY-4.0
Open Portuguese LLM Leaderboard (archived)1▇ 1%Apache-2.0
prior · pull toward the middle▇ 0%
weighted average= 24.9

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

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