license: apache-2.0
pipeline_tag: text-generation
language:
- fr
- en
- it
- de
- es
tags:
- pretrained
- llama-3
- openllm-france
datasets:
- OpenLLM-France/Lucie-Training-Dataset
widget:
- text: |-
Quelle est la capitale de l'Espagne ? Madrid.
Quelle est la capitale de la France ?
example_title: Capital cities in French
group: 1-shot Question Answering
training_progress:
num_steps: 756291
num_tokens: 3131736326144
context_length: 32000
Model Card for Lucie-7B
Model Description
Lucie-7B is a pretrained 7B parameter causal language model built by LINAGORA and OpenLLM-France, available under the Apache 2.0 license.
Lucie-7B was trained on 3 trillion tokens of multilingual data, including English (33.2%), French (32.4%), German (6.9%), Spanish (6.6%), Italian (3.8%), and parallel data from those languages (2.5%), as well as several programming languages (14.7%).
Example code in python
Load the model
Load the model (quantized version on GPU if possible, for efficient inference):
import transformers
model_name = "OpenLLM-France/Lucie-7B"
tokenizer = transformers.AutoTokenizer.from_pretrained(model_name)
model = transformers.AutoModelForCausalLM.from_pretrained(model_name,
device_map="auto",
load_in_4bit=True # For efficient inference, if quantization is supported by the GPU card
)
Sentence completion
Wrap the model in a text generation pipeline, and prepare some generation parameters:
pipeline = transformers.pipeline("text-generation", model=model, tokenizer=tokenizer)
generation_kwargs = dict(
num_return_sequences=1, # Number of variants to generate.
return_full_text= False, # Do not include the prompt in the generated text.
do_sample=True,
temperature=1.0, top_p=1, top_k=None, # Sampling parameters.
max_new_tokens=200, # Maximum length for the output text (in number of tokens).
)
Try 1-shot question answering:
prompt = """\
Quelle est la capitale de l'Espagne ? Madrid\n\
Quelle est la capitale de la France ?\
"""
completions = pipeline(prompt, **generation_kwargs)
for completion in completions:
print(prompt + " […]" + completion['generated_text'])
This will print something like:
Quelle est la capitale de l'Espagne ? Madrid
Quelle est la capitale de la France ? […] Paris
Quelle est la capitale de l'Italie? Rome
Quelle est la capitale de la Grande-Bretagne? Londres
Quelle est la capitale de la Suisse? Berne
Quelle est la capitale du Portugal? Lisbonne
Quelle est la capitale de l'Algérie? Alger
...
If running on GPU (cuda device), you will need at least 6GB of VRAM to run inference using 4bit quantization (16GB of VRAM without 4bit quantization).
Load a checkpoint
Checkpoints at several training steps are available under revision tags, every 5000 steps during the first 25000 steps, and then every 25000 steps.
Intermediate checkpoints can be loaded using the revision parameter:
model = transformers.AutoModelForCausalLM.from_pretrained(model_name,
revision="step0753851",
...
)
where revision can be one of:
- "
step0005000", "step0010000", "step0015000", "step0020000": each 5000 steps for the first pre-training steps (with a context length of 4096). - "
step0025000", "step0050000", "step0075000", "step0100000", ..., "step0750000": each 25000 steps from 25k to 750k steps. - "
step0753851": last pre-training step before context extension and annealing. - "
extension_step0000250", "extension_step0000500", "extension_step0000750", "extension_step0001000", "extension_step0001220": several checkpoints during context extension (with a context length of 32000).
Training Details
Training Data
The training dataset used for the pretraining of Lucie-7B is available at OpenLLM-France/Lucie-Training-Dataset.
The initial composition of the training data is as follows:
Some of the data was upsampled to balance the training data distribution, and the final composition is as follows:
Training Procedure
Lucie-7B is a causal decoder-only model trained on a causal language modeling task (i.e., predict the next token).
It was pre-trained on 512 H100 80GB GPUs for about 550,000 GPU hours on Jean Zay supercomputer.
The training code is available at https://github.com/OpenLLM-France/Lucie-Training. It is based on this fork of Megatron-DeepSpeed.
Optimizer checkpoints are available at OpenLLM-France/Lucie-7B-optimizer-states.
Neural Network Architecture
Lucie-7B has the same neural network architecture as Llama3.1. It has exactly 6 706 958 336 free parameters, with the following hyperparameters:
| Hyperparameter | Value |
|---|---|
| Vocabulary size (# tokens) | 65 024 |
| # transformer blocks | 32 |
| # attention heads | 32 |
| # key-value heads | 8 |
| Hidden size | 4 096 |
| Feed-Forward hidden size | 12 288 |
| Activation | silu |
| RMS norm epsilon | 1e-5 |
The parameter "theta" of Rotary Positional Embedding (RoPE) varied during the training process and is indicated in the tables with training hyperparameters below.
Training Hyperparameters
The training consisted of three main phases:
- Main pre-training on 3.1T tokens, with a context length of 4096,
- Context extension on 5B tokens, with a context length of 32000,
- Annealing, with a selected subset of the training data with especially high quality.
The details of each phase are given below.
1. Main pre-training
Training hyperparameters in torch/Megatron-DeepSpeed were the following:
| Hyperparameter | Value |
|---|---|
| Total # samples | 762 144 586 (3.1T tokens) |
| Total # steps | 753 851 |
| RoPE theta | 500 000 |
| Context length | 4 096 |
| Initial Batch size | 256 |
| Final Batch size | 1 024 |
| Batch size rampup | by steps of 64 over 10M samples |
| Learning rate schedule | warmup (2M samples) + cosine annealing |
| Maximum Learning rate | 3e-4 |
| Final Learning rate | 3e-5 |
| Weight decay | 0.1 |
| Dropout | _ |
| Gradient clipping | 1 |
| Initializer range | 0.009 |
| Optimizer | AdamW (β₁=0.9, β₂=0.95, ε=1e-5) |
| Precision | bfloat16 |
| Tensor Parallelism (with 512 GPUs) | 4 |
| Pipeline Parallelism (with 512 GPUs) | 4 |
| Data Parallelism (with 512 GPUs) | 32 |
2. Context Extension
Training hyperparameters are the same as above, with the following changes:
| Hyperparameter | Value |
|---|---|
| Total # samples | 156 250 (5B tokens) |
| Total # steps | 1 220 |
| RoPE theta | 20 000 000 |
| Context length | 32 000 |
| Batch size | 128 |
| Learning rate | 2e-5 |
| Tensor Parallelism (with 128 GPUs) | 4 |
| Pipeline Parallelism (with 128 GPUs) | 4 |
| Data Parallelism (with 128 GPUs) | 8 |
3. Annealing
TODO
Training logs and learning curves
🚧 work in progress 🚧
Training logs can be found in Tensorboard format in:
metadata/training_logs/
├──1_pretraining.ziptraining logs for the first pre-training phases, in a zip file. Each file in the zip corresponds to a job of at most 20H of training (parallelized over 512 GPUs).
└──2_extension/folder containing the training log for the context extension phase, which was done in a single job of around 13H of training (parallelized over 128 GPUs).
Acknowledgements
This work was performed using HPC resources from GENCI–IDRIS (Grant 2024-GC011015444).
Lucie-7B was created by members of LINAGORA and OpenLLM-France community, including in alphabetical order: Christophe Cerisara (LORIA), Evan Dufraisse (CEA), Julie Hunter (LINAGORA), Jean-Pierre Lorré (LINAGORA), Jérôme Louradour (LINAGORA), Michel-Marie Maudet (LINAGORA), Olivier Gouvert (LINAGORA), Pierre-Carl Langlais (OpSci), Yaya Sy (LORIA).