[whatmodel.sh](../../README.md) / Gemma 7B IT # Gemma 7B IT Google · released 2024-02 · 8.54B params · gemma **score 34.0** ±0.6 · #636 of 732 · from 17 benchmark results across 11 sources Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only. ## What it's good at ● marks Gemma 7B IT among every model on that task or test, worst to best. 
task            vs the others         rank  counts  tests
coding          ▁▁▁●▅▇█▇▅▄▃▁   #435 of 469     72%      6
chat & writing  ▁▁▁●▆▄▆█▆▇▇▃   #372 of 402     14%      4
coding · #435 of 469 · 72% of its score Beats [DeepSeek R1 Distill Qwen 7B](deepseek-r1-distill-qwen-7b.md), [Qwen3.5 2B](qwen3-5-2b.md), [DeepSeek R1 Distill Llama 8B](deepseek-r1-distill-llama-8b.md) here, though they rank higher overall. Trails [Llama 3.2 1B Instruct](llama-3-2-1b.md), [SmolLM2 1.7B Instruct](smollm2-1-7b.md), [Mistral 7B Instruct v0.1](mistral-7b-instruct-v0-1.md) here, though they rank lower overall. 
test                                result  vs the others in its run  difficulty        counts  why
HumanEval                            28.7%  ▁·▁▁●▄▆▅█▃▁▁ beats 12/99  easy · avg 76%       40%  sharp · at its level · several runs
HumanEval+                           25.0%  ▁·▁▁▁●▄▄▄█▃▁ beats 13/99  easy · avg 70%       18%  sharp · at its level
HumanEvalSynthesize (HumanEvalPac…   19.8%  ▁·●▁▂▆▅█▄▄▁▂ beats 2/35   medium · avg 62%    5.4%  at its level
MBPP+                                36.8%  ▁···▁▁●▅▇█▆▂ beats 5/75   easy · avg 66%      4.4%  at its level
HumanEvalExplain (HumanEvalPack, …   21.5%  ▄▇█▃●▇▄▁▇▃·▁ beats 17/35  medium · avg 46%    2.6%  typical
HumanEvalFix (HumanEvalPack, 6-la…   13.7%  ▁·▁▁▄█●▄█▁▅▅ beats 12/35  medium · avg 51%    2.1%  sharp
chat & writing · #372 of 402 · 14% of its score Beats [Code Llama 70B Instruct](codellama-70b-instruct.md), [Gemma 4 E2B](gemma-4-e2b.md), [Phi-2 (2.7B)](phi-2.md) here, though they rank higher overall. Trails [Gemma 3 270M IT](gemma-3-270m.md), [SmolLM2 360M Instruct](smollm2-360m.md), [Granite 4.0 350M](granite-4-0-350m.md) here, though they rank lower overall. 
test                                result  vs the others in its run   difficulty        counts  why
AlpacaEval 2 length-controlled       10.4%  ▁▂▂▃●▆█▆▄▄▁▁ beats 23/122  medium · avg 35%    5.1%  sharp · too hard for it
LMArena (text)                      1137.8  ▁▁▁●▃▃▃▆▆▇█▃ beats 30/385  avg 1260            3.9%  blunt
Polyrating rating (bias-aware re-…  1029.0  ▂▁▂▁●▇█▆▇▄▂▂ beats 19/95   avg 1187            3.6%  typical
Arena-Hard (v0.1, GPT-4-1106 judg…    7.5%  ▂●▁▅█▅▄▃▃▅▃▁ beats 3/60    medium · avg 47%    1.7%  sharp · too hard for it
other tests (6) 
test                                         result  vs the others in its run   difficulty           counts  why
EvoEval (overall)                             17.7%  ▂●▃▄▄█▁▄▅▃▁▃ beats 3/42    medium · avg 52%       7.0%  typical
WildBench v2 WB-Score (task macro)  6.6198 WB-Score  ▁·●▁·▃█▄▇▆▅▆ beats 1/52    avg 43.019 WB-Score    4.2%  sharp
Open PT-LLM Leaderboard average (…            48.0%  ▃▁▁▁▁▁▁●▂█▅▁ beats 64/276  medium · avg 72%       0.7%  at its level · noisy
EuroEval Life in the UK (English …            64.0%  ▁▁▁▁▁▁▁●▄█▄▃ beats 34/146  easy · avg 82%         0.6%  at its level · noisy
Vectara hallucination leaderboard…            85.2%  ▁▁●▃▄▄█▆▃▂▂▁ beats 14/140  very easy · avg 94%    0.5%  blunt · noisy
EuroEval HellaSwag (English), tes…            26.5%  ▃·●▁▂▁▂▅█▇▄▁ beats 8/100   medium · avg 67%       0.4%  sharp · noisy
## How the score was worked out Each result implies a score on its own, after adjusting for how hard that test is and who ran it. The model's 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 is added up by who ran the tests. [Method](../method.md) 
source                                    results  share of the score
EvalPlus / EvalPerf published evaluatio…        3   48% ▇▇▇▇▇▇
Top Leaderboard Ranking = Top Coding Pr…        2   21% ▇▇▇
Granite Code Models: A Family of Open F…        3   10% ▇
AlpacaEval 1.0 and 2.0 leaderboards             1    5% ▇
WildBench v2 leaderboard data                   1    4% ▇
LMArena leaderboard dataset                     1    4% ▇
Polyrating: A Cost-Effective and Bias-A…        1    4% ▇
Arena-Hard-Auto v0.1 leaderboard                1    2% ▇
EuroEval leaderboards                           2    1% ▇
Open Portuguese LLM Leaderboard                 1    1% ▇
Vectara hallucination leaderboard               1    0% ▇
prior: pull toward the middle                        0% ▇
weighted average                                   = 34.0
Licences: EvalPlus / EvalPerf published evaluation runs ([Apache-2.0](../../licenses/evalplus-published-runs-Apache-2.0.txt)) · Top Leaderboard Ranking = Top Coding Proficiency ([CC BY 4.0](../../licenses/paper-2403.19114v1-CC-BY-4.0.txt)) · Granite Code Models: A Family of Open Foundation Models for Code Intelligence ([CC BY 4.0](../../licenses/paper-2405.04324v1-CC-BY-4.0.txt)) · AlpacaEval 1.0 and 2.0 leaderboards ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0)) · WildBench v2 leaderboard data ([Apache-2.0](https://github.com/allenai/WildBench/blob/d6b8dcaf377d173d031980f97c16e1a82618c03d/LICENSE)) · LMArena leaderboard dataset ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Polyrating: A Cost-Effective and Bias-Aware Rating System for LLM Evaluation ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) · Arena-Hard-Auto v0.1 leaderboard ([Apache-2.0](../../licenses/arena-hard-auto-v0-1-Apache-2.0.txt)) · EuroEval leaderboards ([MIT](https://github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/LICENSE)) · Open Portuguese LLM Leaderboard ([Apache-2.0](../../licenses/hf-leaderboard-spaces-Apache-2.0.txt)) · Vectara hallucination leaderboard ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0)) ## How to run it ### llama.cpp (recommended) 1. Install llama.cpp. Homebrew and winget installs name the commands llama-cli and llama-server instead. [Other ways to install](https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md) ```sh # macOS / Linux curl -LsSf https://llama.app/install.sh | sh # Windows irm https://llama.app/install.ps1 | iex ``` 2. Run Gemma 7B IT (Q4_K_M, needs ~5.3 GB, from [huggingface.co/MaziyarPanahi](https://huggingface.co/MaziyarPanahi/gemma-7b-it-GGUF/tree/fc1d4d26cf05555b86004a23ff462860d372d64b)) ```sh llama cli -hf MaziyarPanahi/gemma-7b-it-GGUF:Q4_K_M ``` 3. Or start a local server and open the address it prints (http://localhost:9931) ```sh llama serve -hf MaziyarPanahi/gemma-7b-it-GGUF:Q4_K_M ``` Other versions: Q8_0 (~9.1 GB) `llama cli -hf MaziyarPanahi/gemma-7b-it-GGUF:Q8_0` ### LM Studio 1. Install LM Studio. Open the app once before using the lms command. [Download it (macOS / Windows / Linux)](https://lmstudio.ai/download) 2. Run Gemma 7B IT (Q4_K_M, needs ~5.3 GB, from [huggingface.co/MaziyarPanahi](https://huggingface.co/MaziyarPanahi/gemma-7b-it-GGUF/tree/fc1d4d26cf05555b86004a23ff462860d372d64b)) ```sh lms get https://huggingface.co/MaziyarPanahi/gemma-7b-it-GGUF@Q4_K_M -y ``` 3. Then run lms ls to see its name, and lms chat followed by that name Other versions: Q8_0 (~9.1 GB) `lms get https://huggingface.co/MaziyarPanahi/gemma-7b-it-GGUF@Q8_0 -y` ### Ollama 1. Install Ollama. [Download it (macOS / Windows)](https://ollama.com/download) ```sh # Linux curl -fsSL https://ollama.com/install.sh | sh ``` 2. Run Gemma 7B IT (Q4_K_M, needs ~5.62 GB, from [ollama.com](https://ollama.com/library/gemma/tags)) ```sh ollama run gemma:7b-instruct-q4_K_M ``` Other versions: Q8_0 (~10.75 GB) `ollama run gemma:7b-instruct-q8_0` Model repository and documentation: [huggingface.co/google](https://huggingface.co/google/gemma-7b-it) [interactive version](https://whatmodel.sh/model/gemma-7b) · [JSON](https://whatmodel.sh/api?format=json&all=1)