whatmodel.sh / Llama 2 70B (base)

Llama 2 70B (base)

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

Llama 2 70B (base) is an open-weights language model from Meta, released July 2023. Its published size is 69B parameters and a 4k-token context window. Licence: llama2.

Catalogue identity notes

Pretrained (not chat) Llama 2 checkpoint; the chat models are separate identities (llama-2-*-chat). Ollama 'text' tags are the pretrained model.

40.7±0.9#519 of 719 LLMsfrom 22 benchmark results across 6 sources

$ how it compares

The LLMs ranked either side of Llama 2 70B (base), out of 828 with a score, then more from Meta.

#modelscorewhereRAMreleased
609InternLM2 Chat 20B Shanghai AI Laboratory (InternLM)41±1local~14GB2024-01
610DeepSeek LLM 67B Chat DeepSeek41±1local~31GB2023-11
611Llama 2 70B (base) Meta41±1local~32GB2023-07
612Mixtral 8x7B v0.1 (base) Mistral AI41±1local~29GB2023-12
613North Micro Vision Instruct (2.4B) Cohere41±2local~3.8GB2026-08
more from Meta
24Muse Spark 1.3 Meta77±1api $2.00/M2026-09
40Muse Spark 1.2 Meta75±1api $2.00/M2026-08
43Muse Spark 1.1 Meta75±1api $2.00/M2026-07

see all LLMs · find the ones that fit your machine

$ what it's good at

coding▁▁▁▃▆██▇▆▄▃▁ #411 of 458 · 20% of its score · 3 tests

Beats DeepSeek R1 Distill Qwen 7B, Qwen3.5 2B, DeepSeek R1 Distill Llama 8B here, though they rank higher overall. Trails Llama 3.2 1B Instruct, SmolLM2 1.7B Instruct, Code Llama 7B (base) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
HumanEval29.3% ✎█···█·███··█ beats 4/5easy · avg 66%▇▇▇▇ 11%sharp · at its level
Tulu 2: CodexEval pass@1060.1% ✎▃▃▃·▆▃▃█▃▂▂▅ beats 11/25very easy · avg 86%▇▇ 5.0%sharp · at its level
MBPP49.8% ✎▄··▄·▄··█··▄ beats 3/5easy · avg 71%▇ 3.6%at its level
math▁▂▃▅▆▅▄█▄▁▁▁ #302 of 398 · 11% of its score · 2 tests

Beats GPT-4 (0613), Athene-70B, Llama 3 70B Instruct here, though they rank higher overall. Trails Gemma 7B (base), DeepSeek R1 Distill Qwen 1.5B, CodeGemma 7B (v1.0) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
GSM8K +155.5% ✎▆▂▅█▅▅▃▆···▂ beats 18/25easy · avg 82%▇▇▇ 7.5%sharp · at its level · several runs
MATH13.8% ✎█·█·█·██···█ beats 4/5hard · avg 35%▇ 3.4%typical
other tests15 not in a task · 69% of its score
testresultvs the others in its rundifficultycountswhy
BIG-Bench Hard66.0% ✎▁·····▂█▂▅·▁ beats 20/25easy · avg 77%▇▇▇▇ 11%sharp · at its level
MMLU-Pro36.3% ✎▃▂▅█▃·█▃▅··▂ beats 4/23medium · avg 57%▇▇▇▇ 10%at its level
MMLU +164.5% ✎▁▁▃▃▄█▄▁█··▁ beats 19/25easy · avg 82%▇▇▇▇ 10%at its level · several runs
ARC Challenge56.5% ✎▄···▄···█▄·▄ beats 4/5easy · avg 82%▇▇▇▇ 10%sharp · at its level
Natural Questions (closed book)25.4% ✎█·····▄▄▄··▄ beats 4/5medium · avg 51%▇▇ 6.2%at its level
WinoGrande80.4% ✎█··█·█·██··█ beats 5/5easy · avg 89%▇▇ 5.6%at its level
TriviaQA73.0% ✎█···██··█·██ beats 5/5very easy · avg 86%▇▇ 4.7%at its level
HellaSwag85.4% ✎█···██···███ beats 5/5very easy · avg 92%▇ 3.5%typical
7 more tests
testresultvs the others in its rundifficultycountswhy
ARC Easy79.9% ✎▄···▄···█▄·▄ beats 3/5very easy · avg 89%▇ 2.3%typical
PIQA82.6% ✎▄····▄··█▄·▄ beats 4/5easy · avg 88%▇ 1.5%blunt · at its level
Open PT-LLM Leaderboard average (9 Portuguese tasks)67.7% ✎▃▁▁▁▁▁▁▁▂█▅▁ beats 139/276medium · avg 73%▇ 1.4%at its level · noisy
Tulu 2: TyDiQA gold passage, 1-shot F162.6% ✎▂··▁▃▂▂▂█▅▁▃ beats 24/25medium · avg 61%▇ 1.2%blunt · at its level
EuroEval HellaSwag (English), test split, 5-shot generative MC, accuracy59.0% ✎▃·▁▁▂▁▂▅█▇▄▁ beats 33/100medium · avg 68%▇ 1.1%at its level · noisy
EuroEval Life in the UK (English knowledge MC), test split, 5-shot, accuracy86.0% ✎▁▁▁▁▁▁▁▂▄█▄▃ beats 101/146easy · avg 83%▇ 1.0%at its level · noisy
COPA97.6% ✎▅·▃·▃··█▃··▃ beats 8/8easy · avg 92%▇ 0.3%at its level · 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 Llama 2 70B (base) · needs ~41.4 GB · from huggingface.co/TheBloke
    chat in the terminal
    llama cli -hf TheBloke/Llama-2-70B-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf TheBloke/Llama-2-70B-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 Llama 2 70B (base) · needs ~41.4 GB · from huggingface.co/TheBloke

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/TheBloke/Llama-2-70B-GGUF@Q4_K_M -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Ollama

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run Llama 2 70B (base) · needs ~39 GB · from ollama.com
    chat in the terminal
    ollama run llama2:70b-text

Weights and documentation: huggingface.co/meta-llama ↗

$ how the score was worked out · Mixtral of Experts — base-mod… 61%, Tulu 2 — base and instruction… 24%, MMLU-Pro author-released eval… 10% and 3 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
Mixtral of Experts — base-model evaluations12▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 61%CC BY 4.0
Tulu 2 — base and instruction-tuned evaluations5▇▇▇▇▇▇▇ 24%CC BY 4.0
MMLU-Pro author-released evaluation runs1▇▇▇ 10%Apache-2.0
EuroEval leaderboards (results collection), English generative leaderboard2▇ 2%MIT
Open Portuguese LLM Leaderboard (archived)1▇ 1%Apache-2.0
From Test-Taking to Test-Making: Examining LLM Authoring of Commonsense Assessment Items (arXiv:2410.14897v1)1▇ 0%CC-BY-4.0
prior · pull toward the middle▇ 0%
weighted average= 40.7

$ 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