whatmodel.sh / Pythia 2.8B (deduped)

Pythia 2.8B (deduped)

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

Pythia 2.8B (deduped) is an open-weights language model from EleutherAI, released February 2023. Its published size is 2.91B parameters and a 2k-token context window. Licence: apache-2.0.

Catalogue identity notes

Maker repo EleutherAI/pythia-2.8b-deduped @7d977fe: licence tag apache-2.0. 2,909,427,008 parameters from the Hugging Face safetensors metadata (dense, so active = total). Released 2023-02: month the maker's Hugging Face repo was created (createdAt 2023-02-10). Pretrained base checkpoint (not an instruct/chat version).

20.2±1.1#689 of 719 LLMsfrom 15 benchmark results across 3 sources

$ how it compares

The LLMs ranked either side of Pythia 2.8B (deduped), out of 828 with a score, then more from EleutherAI.

#modelscorewhereRAMreleased
793Qwen1.5 0.5B Chat Alibaba21±3local~1.8GB2024-01
794ChatGLM-6B Zhipu AI (THUDM)20±2local~16GB2023-03
795Pythia 2.8B (deduped) EleutherAI20±1local~3.4GB2023-02
796Flan-T5 Base Google20±3local~2.5GB2022-10
797Falcon-RW 1B TII20±2local~4.3GB2023-04
more from EleutherAI
756GPT-NeoX-20B EleutherAI28±1local~14GB2022-02
777Pythia 12B EleutherAI25±1local~9.5GB2023-02
780GPT-J 6B EleutherAI25±1local~14GB2021-06

see all LLMs · find the ones that fit your machine

$ what it's good at

other tests15 not in a task · 100% of its score
testresultvs the others in its rundifficultycountswhy
ARC Challenge32.9% ✎▁·▁··▁▁▇▂█▆▃ beats 4/26easy · avg 85%▇▇▇▇▇▇▇▇▇▇▇▇ 29%sharp
WinoGrande58.2% ✎▁····▁··▃▂█▄ beats 2/26easy · avg 89%▇▇▇▇ 11%typical
SciQ88.2% ✎▁·▁·▁·▃▅▃█▄▁ beats 3/26very easy · avg 98%▇▇▇▇ 11%typical
BoolQ64.1% ✎▂▃▂▅▆▂▅▆█▂▂▂ beats 6/26easy · avg 91%▇▇▇▇ 10%at its level
ARC Easy59.1% ✎▁▁·▁▁▄▅▄█▅▁▃ beats 5/26easy · avg 88%▇▇▇▇ 9.2%at its level
HellaSwag59.4% ✎▃···▁▇▅█▇▃▁▁ beats 3/26very easy · avg 92%▇▇▇ 7.9%at its level
PIQA74.1% ✎▁·▁··▁▁▃█▅█▂ beats 6/26easy · avg 88%▇▇ 5.8%blunt · at its level
LAMBADA (OpenAI version), last-word accuracy65.1% ✎▁··▂▁▁▁▃▆█▅▃ beats 6/26very easy · avg 89%▇▇ 4.0%blunt · at its level
7 more tests
testresultvs the others in its rundifficultycountswhy
MedQA (USMLE, 4 options)27.7% ✎▁▁··▄▁▁··▄▃█ beats 3/17easy · avg 76%▇ 2.8%sharp · too hard for it
Social IQA40.9% ✎▁···▁▃█▁▁▁·▁ beats 5/26medium · avg 59%▇ 2.3%blunt
MedMCQA26.3% ✎▃▂··▄▄▂▂·▄██ beats 0/17medium · avg 68%▇ 2.3%typical
OpenBookQA23.8% ✎█·▁▁▃·▁▃▅▇▃▄ beats 0/26medium · avg 53%▇ 2.2%too hard for it
Open PT-LLM Leaderboard average (9 Portuguese tasks)24.7% ✎▃▁▁▁▁▁▁▁▂█▅▁ beats 34/276medium · avg 73%▇ 0.9%noisy
PubMedQA64.8% ✎▂·▂▆·▄▂▄▄▂█▂ beats 6/17medium · avg 77%▇ 0.7%blunt · at its level
TruthfulQA MC235.6% ✎█·█·█▄···▄·█ beats 4/9medium · avg 61%▇ 0.7%blunt · at its level

$ 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 2.8B (deduped) · needs ~1.8 GB · from huggingface.co/mradermacher
    chat in the terminal
    llama cli -hf mradermacher/pythia-2.8b-deduped-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf mradermacher/pythia-2.8b-deduped-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 2.8B (deduped) · needs ~1.8 GB · from huggingface.co/mradermacher

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/mradermacher/pythia-2.8b-deduped-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 2.8B (deduped) · needs ~1.8 GB · from huggingface.co/mradermacher
    chat in the terminal
    ollama run hf.co/mradermacher/pythia-2.8b-deduped-GGUF:Q4_K_M

Other ways to run it

  1. Run it as published · needs ~5.68 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-2.8b-deduped', 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… 93%, Open Medical-LLM Leaderboard… 6%, Open Portuguese LLM Leaderboa… 1%

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 outputs11▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 93%Apache-2.0
Open Medical-LLM Leaderboard results (HF dataset openlifescienceai/results)3▇▇ 6%Apache-2.0
Open Portuguese LLM Leaderboard (archived)1▇ 1%Apache-2.0
prior · pull toward the middle▇ 0%
weighted average= 20.2

$ 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