This repository contains the Falcon3-10B-Base. It achieves state-of-the-art results (at the time of release) on reasoning, language understanding, instruction following, code and mathematics tasks. Falcon3-10B-Base supports 4 languages (English, French, Spanish, Portuguese) and a context length of up to 32K.

⚠️ This is a raw, pretrained model, which should be further finetuned using SFT, RLHF, continued pretraining, etc. for most use cases.

Model Details

Architecture
- Transformer-based causal decoder-only architecture
- 40 decoder blocks
- Grouped Query Attention (GQA) for faster inference: 12 query heads and 4 key-value heads
- Wider head dimension: 256
- High RoPE value to support long context understanding: 1000042
- Uses SwiGLu and RMSNorm
- 32K context length
- 131K vocab size
Depth up-scaled from Falcon3-7B-Base with continual pretraining on 2 Teratokens of datasets comprising of web, code, STEM, high quality and mutlilingual data using 1024 H100 GPU chips
Supports EN, FR, ES, PT
Developed by Technology Innovation Institute
License: TII Falcon-LLM License 2.0
Model Release Date: December 2024

Getting started

Click to expand

python

import torch
from transformers import pipeline

pipe = pipeline(
    "text-generation", 
    model="tiiuae/Falcon3-10B-Base", 
    torch_dtype=torch.bfloat16, 
    device_map="auto"
)
response = pipe("Question: How many hours in one day? Answer: ")
print(response[0]['generated_text'])

Benchmarks

We report in the following table our internal pipeline benchmarks.

We use lm-evaluation harness.
We report raw scores.
We use same batch-size across all models.

Category	Benchmark	Gemma2-9B	Yi1.5-9B	Mistral-Nemo-Base-2407 (12B)	Falcon3-10B-Base
General	MMLU (5-shot)	70.8	69.6	68.8	73.1
MMLU-PRO (5-shot)	41.4	39.3	34.7	42.5
IFEval	21.3	29.1	16.1	36.4
Math	GSM8K (5-shot)	69.1	63.8	55.3	81.4
MATH Lvl-5 (4-shot)	10.5	9.2	4.9	22.9
Reasoning	Arc Challenge (25-shot)	67.5	61.7	64.4	66.8
GPQA (0-shot)	33.4	36.6	28.8	34.1
MUSR (0-shot)	45.3	43.3	39.2	44.2
BBH (3-shot)	54.3	51.3	50.2	59.7
CommonSense Understanding	PIQA (0-shot)	83.0	80.5	82.1	79.4
SciQ (0-shot)	97.1	95.2	95.2	93.5
Winogrande (0-shot)	74.2	72.7	73.2	73.6
OpenbookQA (0-shot)	47.2	45.2	47.2	45.0

Useful links

View our release blogpost.
Feel free to join our discord server if you have any questions or to interact with our researchers and developers.

Technical Report

Coming soon....

Citation

If the Falcon3 family of models were helpful to your work, feel free to give us a cite.

@misc{Falcon3,
    title = {The Falcon 3 Family of Open Models},
    url = {https://huggingface.co/blog/falcon3},
    author = {Falcon-LLM Team},
    month = {December},
    year = {2024}
}

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	27.59
IFEval (0-Shot)	36.48
BBH (3-Shot)	41.38
MATH Lvl 5 (4-Shot)	24.77
GPQA (0-shot)	12.75
MuSR (0-shot)	14.17
MMLU-PRO (5-shot)	36.00

ChipHolmes/Falcon3-10B-Base-archive

Falcon3-10B-Base