Description

Falcon-7B is a large language model with 7 billion parameters. It is a causal decoder-only model developed by TII and trained on 1,500 billion tokens of RefinedWeb dataset, which was enhanced with curated corpora. The model is available under the Apache 2.0 license. It outperforms comparable open-source models and features an architecture optimized for inference. However, it is a raw, pretrained model that should be further finetuned for most use cases. The model is recommended for research on large language models and as a foundation for further specialization and finetuning for specific tasks. It should not be used in production without adequate assessment of risks and mitigation. The model carries biases commonly encountered online and is trained on English and French data only. The training details of Falcon-7B include information about the training data, training procedure, and hyperparameters used. It was trained on 384 A100 40GB GPUs using a 2D parallelism strategy combined with ZeRO. The model description mentions the architectural adaptations from the GPT-3 model, such as rotary positional embeddings, multiquery attention, and FlashAttention.

The above summary was generated using ChatGPT. Review the original model card to understand the data used to train the model, evaluation metrics, license, intended uses, limitations and bias before using the model. Some of the content has been made available below.

Training Details

Training Data

Falcon-7B was trained on 1,500B tokens of RefinedWeb , a high-quality filtered and deduplicated web dataset which we enhanced with curated corpora. Significant components from our curated copora were inspired by The Pile (Gao et al., 2020 ).

Data source	Fraction	Tokens	Sources
RefinedWeb-English	79%	1,185B	massive web crawl
Books	7%	110B
Conversations	6%	85B	Reddit, StackOverflow, HackerNews
Code	3%	45B
RefinedWeb-French	3%	45B	massive web crawl
Technical	2%	30B	arXiv, PubMed, USPTO, etc.

The data was tokenized with the Falcon-7B /40B tokenizer.

Training Procedure

Falcon-7B was trained on 384 A100 40GB GPUs, using a 2D parallelism strategy (PP=2, DP=192) combined with ZeRO.

Hyperparameter	Value	Comment
Precision	`bfloat16`
Optimizer	AdamW
Learning rate	6e-4	4B tokens warm-up, cosine decay to 1.2e-5
Weight decay	1e-1
Z-loss	1e-4
Batch size	2304	30B tokens ramp-up

Speeds, Sizes, Times

Training happened in early March 2023 and took about two weeks.

Evaluation

Paper coming soon. See the OpenLLM Leaderboard for early results.

Technical Specifications

Model Architecture and Objective

Falcon-7B is a causal decoder-only model trained on a causal language modeling task (i.e., predict the next token). The architecture is broadly adapted from the GPT-3 paper (Brown et al., 2020 ), with the following differences:

Positionnal embeddings: rotary (Su et al., 2021 );
Attention: multiquery (Shazeer et al., 2019 ) and FlashAttention (Dao et al., 2022 );
Decoder-block: parallel attention/MLP with a single layer norm.

Hyperparameter	Value	Comment
Layers	32
`d_model`	4544	Increased to compensate for multiquery
`head_dim`	64	Reduced to optimise for FlashAttention
Vocabulary	65024
Sequence length	2048

Compute Infrastructure

Hardware

Falcon-7B was trained on AWS SageMaker, on 384 A100 40GB GPUs in P4d instances.

Software

Falcon-7B was trained a custom distributed training codebase, Gigatron. It uses a 3D parallelism approach combined with ZeRO and high-performance Triton kernels (FlashAttention, etc.)

License

Falcon-7B is made available under the Apache 2.0 license.

Finetuning samples

Task	Use case	Dataset	Python sample (Notebook)	CLI with YAML
Text Classification	Emotion Detection	Emotion	emotion-detection.ipynb	emotion-detection.sh

Model Evaluation Sample

Task	Use case	Dataset	Python sample (Notebook)	CLI with YAML
Text generation	Text generation	cnn_dailymail	evaluate-model-text-generation.ipynb	evaluate-model-text-generation.yml

Inference samples

Inference type	Python sample (Notebook)	CLI with YAML
Real time	text-generation-online-endpoint.ipynb	text-generation-online-endpoint.sh
Batch	text-generation-batch-endpoint.ipynb	coming soon

Sample input (for real-time inference)

{
  "input_data": {
      "input_string":["the meaning of life is"]
  }
}

Sample output

[
  {
    "0": "the meaning of life is to find your gift. the purpose of life is to give it away."
  }
]

Model Specifications

LicenseApache-2.0

Last UpdatedAugust 2024

Providertiiuae

Languages1 Language

Quick Start