[whatmodel.sh](../../README.md) / BLOOM 3B
# BLOOM 3B
BigScience · released 2022-07 · 3B params · bigscience-bloom-rail-1.0
**score 16.0** ±0.7 · #710 of 732 · from 47 benchmark results across 8 sources
Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only.
## What it's good at
● marks BLOOM 3B among every model on that task or test, worst to best.
task vs the others rank counts tests
coding █··█·█·●█··█ #3 of 6* 4.1% 1
math ▁●▃▅█▆▄█▄▂▁▁ #422 of 431* 2.3% 1
coding · #3 of 6 · 4.1% of its score
test result vs the others in its run difficulty counts why
HumanEval (maker) 6.5% ▄···▄·▄·●··▄ beats 3/5 easy · avg 79% 4.1% sharp · too hard for it
math · #422 of 431 · 2.3% of its score
Beats [Flan-T5 Large](flan-t5-large.md), [OPT-66B](opt-66b.md), [Flan-T5 Base](flan-t5-base.md) here, though they rank higher overall. Trails [GPT-Neo 125M](gpt-neo-125m.md), [OPT-125M](opt-125m.md), [GPT-2 XL (1.5B)](gpt-2-xl.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
GSM8K 1.2% ▄▄●▄···▄·▄·▄ beats 2/7 easy · avg 65% 2.3% sharp · too hard for it
other tests (23)
test result vs the others in its run difficulty counts why
ARC Challenge 30.4% ▁·▁··●▁▇▂█▆▃ beats 2/26 easy · avg 82% 18% sharp · too hard for it · several runs
ARC Easy 53.3% ▁▁·●▁▄▅▄█▅▁▃ beats 2/26 easy · avg 87% 13% blunt · at its level · several runs
WinoGrande 58.5% ▁····▁··●▂█▄ beats 3/26 easy · avg 88% 11% several runs
PIQA 70.6% ▁·▁··●▁▃█▅█▂ beats 2/26 easy · avg 87% 8.6% blunt · at its level · several runs
SciQ 89.1% ▁·▁·▁·●▅▃█▄▁ beats 4/26 very easy · avg 97% 6.8% blunt · several runs
HellaSwag 54.5% ▃···●▇▅█▇▃▁▁ beats 2/26 very easy · avg 91% 6.6% at its level · several runs
LAMBADA (OpenAI version), last-wo… 51.7% ▁··●▁▁▁▂▆█▅▃ beats 1/26 very easy · avg 87% 5.4% blunt · at its level · several runs
BoolQ 61.7% ▂▃●▅▆▂▅▆█▂▂▂ beats 3/26 easy · avg 88% 4.4% several runs
TriviaQA 17.9% ▁··▁█▃●▁▄▅▄█ beats 10/28 medium · avg 45% 3.8% too hard for it · several runs
Story Cloze Test 67.8% ▂▂▄▂·●·▆·▂█▂ beats 5/14 easy · avg 89% 2.9% blunt · at its level · several runs
COPA (maker) 62.0% ▃··█▅●▃▅·▃▃▃ beats 6/12 very easy · avg 95% 2.8% at its level · several runs
CoQA 61.5% ▄····▄▄·●▄·█ beats 3/6 very easy · avg 92% 1.9% at its level
Natural Questions (closed book) 4.8% ▁·▁·▁●▁▂·▅▂█ beats 5/28 medium · avg 48% 1.8% too hard for it
Winograd Schema Challenge (WSC273) 76.9% ▄▄·▄·●··█··█ beats 3/7 very easy · avg 93% 1.6% blunt · at its level
OpenBookQA 21.8% ●·▁▁▃·▁▃▅▇▃▄ beats 3/26 hard · avg 40% 1.4% too hard for it · several runs
RACE-high 35.2% ▄···█··●▄··█ beats 3/6 medium · avg 51% 0.9% blunt
Social IQA 40.2% ▁···▁●█▁▁▁·▁ beats 2/26 hard · avg 56% 0.8% blunt
DROP 8.9% ▄▄···●·▄·▄·▄ beats 3/6 hard · avg 34% 0.8% blunt
Open PT-LLM Leaderboard average (… 33.5% ▃●▁▁▁▁▁▁▂█▅▁ beats 38/276 medium · avg 72% 0.4% too hard for it · noisy
TruthfulQA MC2 40.6% ▅▃●▆▁▂▅···▁▁ beats 10/31 medium · avg 57% 0.3% blunt · at its level · several runs
HaluEval QA hallucination detecti… 50.4% █●▁·▁▁·▁▁▁▁▁ beats 16/27 hard · avg 66% 0.1% too hard for it · noisy
WebQuestions (closed book) 1.7% █···▄··▄▄●·▄ beats 5/6 very hard · avg 3.5% 0.0% blunt · too hard for it
LogiQA (harness) 20.7% ●··········▁ beats 10/14 very hard · avg 25% 0.0% too hard for it · noisy
source results share of the score
Stability AI 10 30% ▇▇▇▇
EleutherAI lm-evaluation-harness result… 17 30% ▇▇▇▇
Pythia: A Suite for Analyzing Large Lan… 6 17% ▇▇
GPTQ: Accurate Post-Training Quantizati… 5 14% ▇▇
BLOOM: A 176B-Parameter Open-Access Mul… 1 4% ▇
Hallucinations Leaderboard results 4 2% ▇
Crosslingual Generalization through Mul… 3 1% ▇
Open Portuguese LLM Leaderboard 1 0% ▇
prior: pull toward the middle 0% ▇
weighted average = 16.0
Licences: Stability AI ([Apache-2.0](../../licenses/stablelm-original-eval-json-Apache-2.0.txt)) · EleutherAI lm-evaluation-harness results folder ([MIT](../../licenses/lm-eval-harness-results-2023-05-MIT.txt)) · Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling ([CC BY-SA 4.0](../../licenses/paper-2304.01373v2-CC-BY-SA-4.0.txt)) · GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers ([CC BY 4.0](../../licenses/paper-2210.17323v2-gptq-CC-BY-4-0.txt)) · BLOOM: A 176B-Parameter Open-Access Multilingual Language Model ([CC BY 4.0](../../licenses/paper-2211.05100v4-CC-BY-4.0.txt)) · Hallucinations Leaderboard results ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0)) · Crosslingual Generalization through Multitask Finetuning ([CC BY 4.0](../../licenses/acl-2023-acl-long-891-CC-BY-4-0.txt)) · Open Portuguese LLM Leaderboard ([Apache-2.0](../../licenses/hf-leaderboard-spaces-Apache-2.0.txt))
## 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 BLOOM 3B (Q4_K_M, needs ~2.4 GB, from [huggingface.co/mradermacher](https://huggingface.co/mradermacher/bloom-3b-GGUF/tree/6bac74d31c0688d974348d302b63e314e6be960e))
```sh
llama cli -hf mradermacher/bloom-3b-GGUF:Q4_K_M
```
3. Or start a local server and open the address it prints (http://localhost:9931)
```sh
llama serve -hf mradermacher/bloom-3b-GGUF:Q4_K_M
```
Other versions: Q8_0 (~3.9 GB) `llama cli -hf mradermacher/bloom-3b-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 BLOOM 3B (Q4_K_M, needs ~2.4 GB, from [huggingface.co/mradermacher](https://huggingface.co/mradermacher/bloom-3b-GGUF/tree/6bac74d31c0688d974348d302b63e314e6be960e))
```sh
lms get https://huggingface.co/mradermacher/bloom-3b-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 (~3.9 GB) `lms get https://huggingface.co/mradermacher/bloom-3b-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 BLOOM 3B (Q4_K_M, needs ~2.4 GB, from [huggingface.co/mradermacher](https://huggingface.co/mradermacher/bloom-3b-GGUF/tree/6bac74d31c0688d974348d302b63e314e6be960e))
```sh
ollama run hf.co/mradermacher/bloom-3b-GGUF:Q4_K_M
```
Other versions: Q8_0 (~3.9 GB) `ollama run hf.co/mradermacher/bloom-3b-GGUF:Q8_0`
### Other ways to run it
1. Run it as published (fp16, needs ~6.01 GB, from [huggingface.co/bigscience](https://huggingface.co/bigscience/bloom-3b/tree/52bc5b43010b4844513826b8be3f78c7344c37d7))
```sh
uv run --with torch --with 'transformers<5' python -c "from transformers import pipeline; print(pipeline('text-generation', model='bigscience/bloom-3b')('Once upon a time', max_new_tokens=40)[0]['generated_text'])"
```
Model repository and documentation: [huggingface.co/bigscience](https://huggingface.co/bigscience/bloom-3b)
[interactive version](https://whatmodel.sh/model/bloom-3b) · [JSON](https://whatmodel.sh/api?format=json&all=1)