whatmodel.sh / Vicuna 13B v1.5

Vicuna 13B v1.5

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

Vicuna 13B v1.5 is an open-weights language model from LMSYS, released August 2023. Its published size is 13B parameters and a 4k-token context window. Licence: llama2.

Catalogue identity notes

Arena entry 'vicuna-13b' (text). LMArena's leaderboard table (arena-leaderboard space @6f2eeeb) labels 'vicuna-13b' as 'Vicuna-13B' (LMSYS, Llama 2 Community), linking https://huggingface.co/lmsys/vicuna-13b-v1.5. Maker repo lmsys/vicuna-13b-v1.5 @c8327bf: licence tag llama2. Parameters: nominal size from the card/model name (fine-tuned Llama 2 13B). Released 2023-08: FastChat README (pinned 2c68a13): '[2023/08] We released Vicuna v1.5 based on Llama 2'. The Arena key 'vicuna-13b' is unversioned and has been on the leaderboard since 2023; the dataset's licence field ('Llama 2 Community') and LMArena's own leaderboard table (link lmsys/vicuna-13b-v1.5) identify the Llama 2-based v1.5 checkpoint. Votes cast before v1.5 existed (Aug 2023) may be pooled under the same key.

36.2±1.5#586 of 719 LLMsfrom 5 benchmark results across 5 sources

$ how it compares

The LLMs ranked either side of Vicuna 13B v1.5, out of 828 with a score, then more from LMSYS.

#modelscorewhereRAMreleased
677Falcon3 3B Instruct TII37±3local~3.4GB2024-12
678Zephyr 7B alpha Hugging Face36±1local~6.1GB2023-10
679Vicuna 13B v1.5 LMSYS36±2local~9.8GB2023-08
680Granite 3.1 2B Instruct IBM36±1local~3.1GB2024-12
681EXAONE 3.5 2.4B Instruct LG AI Research36±2local~3GB2024-12
more from LMSYS
635Vicuna 33B v1.3 LMSYS39±1local~23GB2023-06
716Vicuna 7B v1.5 LMSYS33±1local~5.8GB2023-08
804FastChat-T5 3B LMSYS17±2local~8.5GB2023-04

see all LLMs · find the ones that fit your machine

$ what it's good at

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #357 of 393 · 95% of its score · 4 tests

Beats Gemma 4 E2B, Code Llama 70B Instruct, Code Llama 34B Instruct here, though they rank higher overall. Trails Gemma 3 270M IT, SmolLM2 360M Instruct, Qwen2.5 0.5B Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
AlpacaEval 2 length-controlled10.5% ✎▁▂▂▃▆▆█▆▄▄▁▁ beats 24/122medium · avg 39%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 38%typical
Polyrating rating (bias-aware re-fit of Chatbot Arena votes, mid-2024)1077.0 ✎▂▁▂▁▃▇█▆▇▄▂▂ beats 36/95avg 1201▇▇▇▇▇▇▇▇ 20%blunt
LMArena (text)1141.5 ✎▁▁▁▃▃▃▃▆▆▇█▃ beats 34/379avg 1268▇▇▇▇▇▇▇▇ 20%blunt
MT-Bench (GPT-4 single-answer grading, LMSYS runs) lab6.5700 MT-Bench score (1-10) ✎▆▂▂▆▆▄██▂▄▄▆ beats 18/28avg 8.819 MT-Bench score (1-10)▇▇▇▇▇▇▇ 17%sharp · noisy
other tests1 not in a task · 3.9% of its score
testresultvs the others in its rundifficultycountswhy
Open PT-LLM Leaderboard average (9 Portuguese tasks)63.7% ✎▃▁▁▁▁▁▁▁▃█▅▁ beats 104/276medium · avg 73%▇▇ 3.9%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 Vicuna 13B v1.5 · needs ~7.9 GB · from huggingface.co/TheBloke
    chat in the terminal
    llama cli -hf TheBloke/vicuna-13B-v1.5-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf TheBloke/vicuna-13B-v1.5-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 Vicuna 13B v1.5 · needs ~7.9 GB · from huggingface.co/TheBloke

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/TheBloke/vicuna-13B-v1.5-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 Vicuna 13B v1.5 · needs ~8.3 GB · from ollama.com
    chat in the terminal
    ollama run vicuna:13b-v1.5-q4_K_M

Official weights and documentation: huggingface.co/lmsys ↗

$ how the score was worked out · AlpacaEval 1.0 and 2.0 leader… 38%, Polyrating: A Cost-Effective… 20%, LMArena leaderboard dataset 20% 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

sourceresultsshare of the scorelicence
AlpacaEval 1.0 and 2.0 leaderboards (tatsu-lab/alpaca_eval repository)1▇▇▇▇▇▇▇▇▇▇▇ 38%Apache-2.0
Polyrating: A Cost-Effective and Bias-Aware Rating System for LLM Evaluation (arXiv:2409.00696v3)1▇▇▇▇▇▇ 20%CC-BY-4.0
LMArena leaderboard dataset1▇▇▇▇▇▇ 20%CC BY 4.0
LMSYS Chatbot Arena leaderboard Space: MT-Bench column of the published leaderboard table · lab-reported1▇▇▇▇▇ 17%Apache-2.0
Open Portuguese LLM Leaderboard (archived)1▇ 4%Apache-2.0
prior · pull toward the middle▇ 1%
weighted average= 36.2

$ 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