VillanovaAI/Villanova-2B-Base-2512-Preview
Model Card for Villanova-2B-Base-2512-Preview
<img src="https://cdn-uploads.huggingface.co/production/uploads/6426a5c798a5be164d38ae44/aULFKOno-4UaH1Le3nkGY.webp" alt="Villanova.AI logo" width="96"/>
Villanova is a family of multilingual and multimodal Large Language Models (LLMs). VillanovaAI/Villanova-2B-Base-2512-Preview is a base text-only LLM.
[!WARNING] DISCLAIMER: This model is a preview.
Model Summary
Villanova-2B-Base-2512-Preview is a decoder-only transformer of 2B parameters.
Villanova-2B-Base-2512-Preview was pre-trained from scratch on 2.2 trillion tokens drawn from a curated, high-quality corpus, in a two-stage fashion.
It supports 5 languages: English, Italian, Spanish, French and German.
Stage 1 (0T → 2T tokens)
Broad, diverse multilingual data mixture with primary focus on the five core languages of the Villanova project.
Stage 2 (2T → 2.2T tokens)
Cosine annealing learning rate schedule over a mixture of 200B higher-quality tokens.
How to Use
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "VillanovaAI/Villanova-2B-Base-2512-Preview"
device = "cuda" # for GPU usage or "cpu" for CPU usage
# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name).to(device)
# prepare the model input
prompt = "What is gravity?"
model_inputs = tokenizer([prompt], return_tensors="pt").to(model.device)
# Generate the output
generated_ids = model.generate(**model_inputs, max_new_tokens=128, do_sample=True, temperature=0.7)
# Get and decode the output
output_ids = generated_ids[0][len(model_inputs.input_ids[0]) :]
print(tokenizer.decode(output_ids, skip_special_tokens=True))Evaluation
Overall performance of Villanova-2B-Base-2512-Preview on English and Multilingual Benchmarks.
<img src="https://cdn-uploads.huggingface.co/production/uploads/6426a5c798a5be164d38ae44/F2Epnvtr34egkz8hKEQ4T.png" alt="Model size/performance" width="672"/>
Detailed results are enlisted in the following tables.
Global evaluation:
English only:
Multilingual Benchmarks:
