|
|
--- |
|
|
license: apache-2.0 |
|
|
language: |
|
|
- en |
|
|
- de |
|
|
- es |
|
|
- fr |
|
|
- it |
|
|
pipeline_tag: text-generation |
|
|
library_name: transformers |
|
|
--- |
|
|
|
|
|
# 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](https://huggingface.co/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 |
|
|
```python |
|
|
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**: |
|
|
|
|
|
| **Model** | **Training Tokens (T)** | **Average** | **arc_easy** | **hellaswag** | **hellaswag_de** | **hellaswag_es** | **hellaswag_fr** | **hellaswag_it** | **openbookqa** | **piqa** | **sciq** | **winogrande** | **xcopa_it** | **xnli_de** | **xnli_en** | **xnli_es** | **xnli_fr** | |
|
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---| |
|
|
| Minerva-3B-base-v1.0 | 0.66 | 47.20 | 62.33 | 46.28 | 27.20 | 29.69 | 29.02 | 40.01 | 24.60 | 74.27 | 88.00 | 56.75 | **69.60** | 34.54 | 52.13 | 36.31 | 37.35 | |
|
|
| EuroLLM-1.7B | 4 | 52.35 | 69.07 | 45.04 | 37.97 | 40.98 | 40.05 | 39.46 | 29.80 | 72.20 | 90.60 | 61.25 | 66.00 | 47.99 | 50.24 | 45.58 | 49.00 | |
|
|
| OLMo-2-0425-1B | 4 | 49.15 | 72.73 | **50.79** | 29.79 | 31.34 | 32.60 | 29.19 | 30.00 | **75.95** | 95.30 | **64.72** | 52.60 | 40.00 | 51.77 | 37.63 | 42.89 | |
|
|
| salamandra-2b | 13 | 52.90 | 71.04 | 47.19 | 38.01 | 42.07 | 40.60 | 38.56 | 26.80 | 72.69 | 91.90 | 61.72 | 65.40 | 47.79 | 51.97 | **49.08** | 48.67 | |
|
|
| Qwen3-1.7B-Base | - | 53.32 | 73.61 | 49.29 | 37.54 | 40.73 | 39.27 | 38.45 | **30.20** | 75.90 | **95.80** | 64.01 | 64.20 | 46.47 | **54.50** | 44.06 | 45.78 | |
|
|
| **Villanova-2B-Base-2512-Preview** | 2.2 | **55.25** | **75.13** | 48.57 | **42.06** | **45.72** | **44.62** | **43.32** | 26.60 | 75.08 | 94.40 | 61.96 | 68.40 | **49.36** | 52.21 | 49.04 | **52.33** | |
|
|
|
|
|
**English only**: |
|
|
|
|
|
| **Model** | **Average** | **arc_easy** | **hellaswag** | **openbookqa** | **piqa** | **sciq** | **winogrande** | **xnli_en** | |
|
|
|---|---|---|---|---|---|---|---|---| |
|
|
| Minerva-3B-base-v1.0 | 57.76 | 62.33 | 46.28 | 24.60 | 74.27 | 88.00 | 56.75 | 52.13 | |
|
|
| EuroLLM-1.7B | 59.74 | 69.07 | 45.04 | 29.80 | 72.20 | 90.60 | 61.25 | 50.24 | |
|
|
| OLMo-2-0425-1B | 63.04 | 72.73 | **50.79** | 30.00 | **75.95** | 95.30 | **64.72** | 51.77 | |
|
|
| salamandra-2b | 60.47 | 71.04 | 47.19 | 26.80 | 72.69 | 91.90 | 61.72 | 51.97 | |
|
|
| Qwen3-1.7B-Base | **63.33** | 73.61 | 49.29 | **30.20** | 75.90 | **95.80** | 64.01 | **54.50** | |
|
|
| **Villanova-2B-Base-2512-Preview** | 61.99 | **75.13** | 48.57 | 26.60 | 75.08 | 94.40 | 61.96 | 52.21 | |
|
|
|
|
|
**Multilingual Benchmarks**: |
|
|
|
|
|
| **Model** | **Average** | **hellaswag_de** | **hellaswag_es** | **hellaswag_fr** | **hellaswag_it** | **xcopa_it** | **xnli_de** | **xnli_es** | **xnli_fr** | |
|
|
|---|---|---|---|---|---|---|---|---|---| |
|
|
| Minerva-3B-base-v1.0 | 37.96 | 27.20 | 29.69 | 29.02 | 40.01 | **69.60** | 34.54 | 36.31 | 37.35 | |
|
|
| EuroLLM-1.7B | 45.88 | 37.97 | 40.98 | 40.05 | 39.46 | 66.00 | 47.99 | 45.58 | 49.00 | |
|
|
| OLMo-2-0425-1B | 37.01 | 29.79 | 31.34 | 32.60 | 29.19 | 52.60 | 40.00 | 37.63 | 42.89 | |
|
|
| salamandra-2b | 46.27 | 38.01 | 42.07 | 40.60 | 38.56 | 65.40 | 47.79 | **49.08** | 48.67 | |
|
|
| Qwen3-1.7B-Base | 44.56 | 37.54 | 40.73 | 39.27 | 38.45 | 64.20 | 46.47 | 44.06 | 45.78 | |
|
|
| **Villanova-2B-Base-2512-Preview** | **49.36** | **42.06** | **45.72** | **44.62** | **43.32** | 68.40 | **49.36** | 49.04 | **52.33** | |
