|
|
--- |
|
|
base_model: BSC-LT/salamandra-2b-instruct |
|
|
tags: |
|
|
- text-generation-inference |
|
|
- transformers |
|
|
- unsloth |
|
|
- llama |
|
|
- trl |
|
|
- grpo |
|
|
license: apache-2.0 |
|
|
language: |
|
|
- en |
|
|
datasets: |
|
|
- ericrisco/gsm8k-translated-catalan |
|
|
- ericrisco/gsm8k-translated-spanish |
|
|
- openai/gsm8k |
|
|
--- |
|
|
# Salamandra 2B Reasoning R1 Model Card |
|
|
|
|
|
Salamandra is a highly multilingual model pre-trained from scratch that comes in different sizes. This model card corresponds to the **2B instructed version**, fine-tuned using **GRPO (Group Reward Policy Optimization)** and **Unsloth**. |
|
|
|
|
|
To visit the model cards of other Salamandra versions, please refer to the Model Index. |
|
|
|
|
|
The entire Salamandra family is released under a permissive Apache 2.0 license. Along with the open weights, all training scripts and configuration files are made publicly available in this GitHub repository. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
### Description |
|
|
Salamandra-2B is a **reasoning-focused** transformer-based language model fine-tuned with **GRPO**. It has been trained on **high-quality datasets**, including: |
|
|
|
|
|
- **GSM8K (English)** |
|
|
- **GSM8K Translated (Spanish)** |
|
|
- **GSM8K Translated (Catalan)** |
|
|
|
|
|
This dataset selection allows the model to **reason through complex problems** in multiple languages. Instead of relying on traditional supervised fine-tuning, **GRPO optimizes the model through reward-based reinforcement learning**, making it more adaptive to structured reasoning tasks. |
|
|
|
|
|
## Intended Use |
|
|
|
|
|
### Direct Use |
|
|
The model is designed as a **reasoning assistant** capable of structured problem-solving across different domains. It can be used for: |
|
|
- Logical and mathematical reasoning tasks |
|
|
- Multi-step question answering |
|
|
- Instruction following in multilingual contexts |
|
|
|
|
|
### Out-of-scope Use |
|
|
The model is not intended for malicious applications or any activity that violates legal or ethical standards. |
|
|
|
|
|
## How to Use |
|
|
|
|
|
The instruction-following models use the **ChatML template** for structured dialogue formatting: |
|
|
|
|
|
```python |
|
|
from datetime import datetime |
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
|
import torch |
|
|
|
|
|
model_id = "ericrisco/salamandra-2b-r1" |
|
|
|
|
|
text = "At what temperature does water boil?" |
|
|
|
|
|
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) |
|
|
|
|
|
print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |
