[whatmodel.sh](../../README.md) / Falcon 40B (base)
# Falcon 40B (base)
TII · released 2023-05 · 41.8B params · apache-2.0
**score 36.4** ±1.1 · #615 of 732 · from 16 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 Falcon 40B (base) among every model on that task or test, worst to best.
task vs the others rank counts tests
coding ▁▁▁●▅▇█▇▅▄▃▁ #443 of 469* 17% 2
math ▁▁▃▅●▆▄█▄▂▁▁ #327 of 431* 10% 1
coding · #443 of 469 · 17% of its score
Beats [Qwen3.5 2B](qwen3-5-2b.md), [Molmo 7B-D (0924)](molmo-7b-d.md), [Tulu 2 7B](tulu-2-7b.md) here, though they rank higher overall. Trails [Granite 3.0 1B-A400M Instruct](granite-3-0-1b-a400m.md), [Llama 3.2 1B Instruct](llama-3-2-1b.md), [SmolLM2 1.7B Instruct](smollm2-1-7b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
HumanEval 24.4% ▁▁▁▁▂●▆▆█▅▂▁ beats 36/205 easy · avg 67% 11% sharp · at its level
MBPP 36.5% ▁▁▁●▅█▆▅▄▁▂▁ beats 13/86 easy · avg 68% 6.4% at its level
math · #327 of 431 · 10% of its score
Beats [GPT-4 (0613)](gpt-4-0613.md), [Athene-70B](athene-70b.md), [Gemma 2 27B IT](gemma-2-27b.md) here, though they rank higher overall. Trails [Stable Code 3B](stable-code-3b.md), [Qwen3 0.6B](qwen3-0-6b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
GSM8K 22.4% ▂▂▂▂▃●▅▇██▆▄ beats 51/206 easy · avg 73% 10% sharp · at its level
other tests (13)
test result vs the others in its run difficulty counts why
BIG-Bench Hard 36.2% ▁▁▁▁▂●▅▆█▃▃▁ beats 26/155 medium · avg 65% 15% sharp
Natural Questions (closed book) 32.0% ▁·▁·▁▂▁▂·▅▂● beats 27/28 medium · avg 48% 12% typical
ARC Challenge 57.7% ▄▁▁▂▁▄█●▃▂▃▂ beats 102/140 easy · avg 85% 9.1% sharp · at its level
ImageNetVC visual commonsense, te… 61.7% ▂·▄·▂█▄▄█▄▂● beats 20/21 easy · avg 74% 9.0% blunt · at its level
MMLU 53.5% ▁▁▁▁▁▂▂●██▆▄ beats 62/214 easy · avg 79% 8.3% at its level
Factual knowledge recall, Premium… 26.6% ▅····▃▅·█●·█ beats 8/13 medium · avg 52% 6.7% at its level
Factual knowledge recall, Premium… 42.3% ▃▅▃··▅▃·●▃▃▅ beats 9/13 medium · avg 61% 3.8% at its level
Open PT-LLM Leaderboard average (… 60.3% ▃▁▁▁▁▁▁▁●█▅▁ beats 91/276 medium · avg 72% 2.6% at its level · noisy
C-Eval 40.3% ▁···▁▁●▁▂▂▃█ beats 3/33 medium · avg 59% 2.5% noisy
This is not a Dataset (negation),… 49.7% ●▁····▃▁·▁·▁ beats 0/14 hard · avg 66% 2.3% sharp · too hard for it · noisy
COPA 84.4% ▅·▃·▃··●▃··▃ beats 5/8 easy · avg 92% 0.6% at its level · noisy
TruthfulQA MC2 41.6% ▅▃●▆▁▂▅···▁▁ beats 12/31 medium · avg 57% 0.4% blunt · at its level
IndoMMLU (zero-shot) 32.1% ▅▃▅··▃···●▃█ beats 6/10 hard · avg 37% 0.3% blunt · noisy
source results share of the score
MiniCPM: Unveiling the Potential of Sma… 7 62% ▇▇▇▇▇▇▇
Hallucinations Leaderboard results 2 12% ▇
Towards a Holistic Evaluation of LLMs o… 2 10% ▇
ImageNetVC: Zero- and Few-Shot Visual C… 1 9% ▇
Open Portuguese LLM Leaderboard 1 3% ▇
This is not a Dataset: A Large Negation… 1 2% ▇
From Test-Taking to Test-Making: Examin… 1 1% ▇
Large Language Models Only Pass Primary… 1 0% ▇
prior: pull toward the middle 1% ▇
weighted average = 36.4
Licences: MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Hallucinations Leaderboard results ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0)) · Towards a Holistic Evaluation of LLMs on Factual Knowledge Recall ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) · ImageNetVC: Zero- and Few-Shot Visual Commonsense Evaluation on 1000 ImageNet Categories ([CC BY 4.0](../../licenses/acl-2023-findings-emnlp-133-CC-BY-4-0.txt)) · Open Portuguese LLM Leaderboard ([Apache-2.0](../../licenses/hf-leaderboard-spaces-Apache-2.0.txt)) · This is not a Dataset: A Large Negation Benchmark to Challenge Large Language Models ([CC BY 4.0](../../licenses/acl-2023-emnlp-main-531-CC-BY-4-0.txt)) · From Test-Taking to Test-Making: Examining LLM Authoring of Commonsense Assessment Items ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) · Large Language Models Only Pass Primary School Exams in Indonesia: A Comprehensive Test on IndoMMLU ([CC BY 4.0](../../licenses/acl-2023-emnlp-main-760-CC-BY-4-0.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 Falcon 40B (base) (Q4_K_M, needs ~25.5 GB, from [huggingface.co/mradermacher](https://huggingface.co/mradermacher/falcon-40b-GGUF/tree/afba375ee076322b091dfe38a5f3c3b5f791a0c4))
```sh
llama cli -hf mradermacher/falcon-40b-GGUF:Q4_K_M
```
3. Or start a local server and open the address it prints (http://localhost:9931)
```sh
llama serve -hf mradermacher/falcon-40b-GGUF:Q4_K_M
```
Other versions: Q8_0 (~44.5 GB) `llama cli -hf mradermacher/falcon-40b-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 Falcon 40B (base) (Q4_K_M, needs ~25.5 GB, from [huggingface.co/mradermacher](https://huggingface.co/mradermacher/falcon-40b-GGUF/tree/afba375ee076322b091dfe38a5f3c3b5f791a0c4))
```sh
lms get https://huggingface.co/mradermacher/falcon-40b-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 (~44.5 GB) `lms get https://huggingface.co/mradermacher/falcon-40b-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 Falcon 40B (base) (Q4_0, needs ~24 GB, from [ollama.com](https://ollama.com/library/falcon/tags))
```sh
ollama run falcon:40b-text
```
### Other ways to run it
1. Run it as published (BF16, needs ~83.67 GB, from [huggingface.co/tiiuae](https://huggingface.co/tiiuae/falcon-40b/tree/05ab2ee8d6b593bdbab17d728de5c028a7a94d83))
```sh
uv run --with torch --with 'transformers<5' --with accelerate python -c "from transformers import pipeline; print(pipeline('text-generation', model='tiiuae/falcon-40b', torch_dtype='auto', device_map='auto')('Once upon a time', max_new_tokens=40)[0]['generated_text'])"
```
Model repository and documentation: [huggingface.co/tiiuae](https://huggingface.co/tiiuae/falcon-40b)
[interactive version](https://whatmodel.sh/model/falcon-40b) · [JSON](https://whatmodel.sh/api?format=json&all=1)