Falcon3-Mamba-7B-Base

Falcon3 family of Open Foundation Models is a set of pretrained and instruct LLMs ranging from 1B to 10B.

This repository contains the Falcon3-Mamba-7B. It achieves, compared to similar SSM-based models of the same size, state of art results (at release's time) on reasoning, language understanding, instruction following, code and mathematics tasks. Falcon3-Mamba-7B-Base supports a context length up to 32K and was mainly trained on english corpus.

Model Details

• Architecture (same as Falcon-Mamba-7b)
  • Mamba1 based causal decoder only architecture trained on a causal language modeling task (i.e., predict the next token).
  • 64 decoder blocks
  • width: 4096
  • state dimension: 16
  • 32k context length
  • 65k vocab size
• Continue Pretrained from Falcon Mamba 7B, with another 1500 Gigatokens of data comprising of web, code, STEM and high quality data.
• Postrained on 1.2 million samples of STEM, conversations, code, and safety.
• Developed by Technology Innovation Institute
• License: TII Falcon-LLM License 2.0
• Model Release Date: December 2024

Getting started

Click to expand

from transformers import AutoTokenizer, AutoModelForCausalLM


from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "tiiuae/Falcon3-Mamba-7B-Base"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "How many hours in one day?"
messages = [
    {"role": "system", "content": "You are a helpful friendly assistant Falcon3 from TII, try to follow instructions as much as possible."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=1024
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)

Benchmarks

We report in the following table our internal pipeline benchmarks. For the benchmarks marked by star, we normalize the results with HuggingFace score normalization:

Category	Benchmark	Zamba2-7B	Llama-3.1-8B	Falcon-Mamba-7B	Falcon3-Mamba-7B-Base
General	MMLU (5-shot)	64.9	66.4	59.9	64.9
	MMLU-PRO (5-shot)*	24.5	24.9	14.5	22.6
	IFEval	37.4	12.7	33.4	30.1
Math	GSM8K (5-shot)	55.8	47.9	51.3	65.9
	MATH (4-shot)	10.3	5.1	3.6	15.6
Reasoning	Arc Challenge (25-shot)	54.1	58.5	55.9	56.7
	GPQA (0-shot)*	9.4	6.2	8.1	10.6
	MUSR (0-shot)*	7.5	8.9	10.9	4.5
	BBH (3-shot)*	27.9	25.3	19.9	25.6
CommonSense Understanding	PIQA (0-shot)	79.27	81.2	80.2	79.54
	SciQ (0-shot)	94.4	94.6	96.3	92.0
	Winogrande (0-shot)	77.4	74.0	74.9	71.27

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.

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},
    author = {Falcon-LLM Team},
    month = {December},
    year = {2024}
}