metadata

license: apache-2.0
datasets:
  - fistro/gromenauer
language:
  - es
pipeline_tag: text-generation

Bertin-Gromenauer

Overview

Bertin-Gromenauer is a Spanish language model designed to understand and generate high-quality Spanish text. Developed using the robust Mistral architecture, this model has been trained on an extensive literary corpus, ensuring it captures a wide range of linguistic nuances, styles, and contexts found in Spanish literature.

Model Details

Model Type: Mistral
Sequence Length: 8192
Hidden Dimension: 4096
Intermediate Dimension: 14336
Number of Layers: 32
Number of Attention Heads: 32
Number of Key-Value Heads: 8
Activation Function: SiLU
Initializer Range: 0.02
Layer Norm Epsilon: 1.0e-05
Use Flash Attention: Yes
Gradient Checkpointing: Enabled (Block Size: 5)
Sliding Window Attention: 4096
Use Bias: No

Training Details

Tokenizer: mistralai/Mistral-7B-v0.1
Batch Size: 512
Learning Rate: 1e-5
Optimizer: Adam with beta1=0.9, beta2=0.95, epsilon=1e-8
Weight Decay: 0.1
Warmup Steps: 200
Learning Rate Schedule: Cosine
Number of Training Steps: 7000

Usage

To load the model in your project, you can use the following code:

from transformers import AutoModel, AutoTokenizer

# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained("bertin-project/bertin-gromenauer")

# Load the model
model = AutoModel.from_pretrained("bertin-project/bertin-gromenauer")

# Example usage
text = "Introduce aquí tu texto en español."
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)