proxectonos/Carballo-Science

Carballo-Science

Table of Contents

<details> <summary>Click to expand</summary>

• —Carballo-Legal
• —Table of Contents
• —Model description
• —Intended uses and limitations
• —How to use
• —Training
• —Tools
• —Training data
• —Training hyperparameters
• —Framework
• —Evaluation
• —Additional information
• —Funding
• —Cite this model

</details>

Model description

Carballo-Science is a specialized 7B-parameter instruction-tuned model designed for scientific text understanding and generation in Galician (GL) and Spanish (ES).

It is based on the foundation model BSC-LT/salamandra-7b-instruct and has been further trained on high-quality scientific corpora extracted from diverse sources.

Intended uses and limitations

Intended uses

• —Scientific-oriented text generation (summaries, rephrasing, explanations).
• —Chat-style scientific assistance (non-professional).

Limitations

• —May produce incomplete or incorrect scientific statements.
• —Not suitable for high-stakes or science decision-making.
• —Works best for GL and ES; other languages are not reinforced in this checkpoint.

How to use

from datetime import datetime
from transformers import AutoTokenizer, AutoModelForCausalLM
import transformers
import torch

model_id = "proxectonos/Carballo-Science"

text = "Qué sabes sobre o Proxecto Nós?"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype=torch.bfloat16
  )

message = [ { "role": "user", "content": text } ]
date_string = datetime.today().strftime('%Y-%m-%d')

prompt = tokenizer.apply_chat_template(
    message,
    tokenize=False,
    add_generation_prompt=True,
    date_string=date_string
)

inputs = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
outputs = model.generate(input_ids=inputs.to(model.device), max_new_tokens=200)
generated_tokens = outputs[0][len(inputs[0]):]
response = self.tokenizer.decode(generated_tokens, skip_special_tokens=False).strip()
response = response.split("<|reserved_token_1|>")[0].strip()
print(response)

Training

Training data

The model was trained on a mixture of general instructions and domain-specific legal texts.

Training hyperparameters

• —epochs: 0.5
• —dtype: bf16
• —block size: 2048
• —total batch size: 128
• —learning rate: 2e-6
• —scheduler: Linear
• —optimizations:
• —gradient checkpointing: True
• —flash attention: True
• —liger kernels: True
• —DeepSpeed stage: 2

Framework

Training was performed at the Galician Supercomputing Center (CESGA) on 2 nodes with 2× NVIDIA A100 40GB each, totaling 4 GPUs, across 2 days.

Evaluation

Formal evaluation is in progress. Early observations show improved handling of legal terminology, structured documents, and administrative phrasing in GL and ES.

Additional information

Funding

This work is funded by the Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU within the framework of the project Desarrollo de Modelos ALIA

Cite this model

Please cite the model as follows:

@misc{carballo_legal_2025,
    title     = {Carballo-Science: A Science Domain Instruction-Tuned Model for Galician and Spanish},
    author    = {Proxecto Nós Team},
    year      = {2025},
    publisher = {HuggingFace},
    howpublished = {\url{https://huggingface.co/proxectonos/Carballo-Science}},
}