Poro 2 8B Instruct is an instruction-following chatbot model created through supervised fine-tuning (SFT) and Direct Preference Optimization (DPO) of the Poro 2 8B Base model. This model is designed for conversational AI applications and instruction following in both Finnish and English. It was trained on a carefully curated mix of English and Finnish instruction data, followed by preference tuning to improve response quality.

Poro 2 was created in a collaboration between AMD Silo AI, the TurkuNLP group of the University of Turku, and High Performance Language Technologies (HPLT). Training was conducted on the LUMI supercomputer, using compute resources generously provided by CSC - IT Center for Science, Finland.

This model demonstrates how continued pretraining followed by instruction tuning can efficiently add new language capabilities to existing models while maintaining strong conversational abilities in both the original and target languages.

For more details on our training and data generation pipeline, check out our Continued Pretraining Playbook.

Poro 2 Model Family

The Poro 2 model family includes both 8B and 70B models, and there are three different versions released of the Poro 2 models: a base model, a post-training SFT-only checkpoint, and the final instruct model which is the SFT model plus a round of DPO.

Model	Based on	Base Model	SFT	Instruct
Poro 2 8B	Llama 3.1 8B	Poro 2 8B Base	Poro 2 8B SFT	Poro 2 8B Instruct
Poro 2 70B	Llama 3.1 70B	Poro 2 70B Base	Poro 2 70B SFT	Poro 2 70B Instruct

What does Poro mean? Poro is the Finnish word for Reindeer! 🦌 These animals are native to Finland and hold a significant role in Finnish culture and history.

Model Overview

Poro 2 8B Instruct is based on the Llama 3.1 8B architecture and has been fine-tuned for instruction following and conversational AI applications. The model supports both English and Finnish conversations.

Hyperparameter	Value
n_parameters	8.03B
n_layers	32
n_heads	32
n_kv_heads	8
d_model	4096
vocab_size	128256
max_sequence_length	8192
base_model	Llama-3.1-8B

Training Process

Continued Pretraining

The base Poro 2 8B model was created through continued pretraining on 165B tokens of Finnish, English, code, and math data.

Supervised Fine-Tuning (SFT)

The SFT phase used 1.4M instruction-following examples in English and Finnish, including:

English and Finnish Tulu 3 prompts with Llama-3.3-70B-Instruct responses
Multi-turn conversations generated using the Magpie method
Top-rated conversations from OASST2 and Avoin Avustaja datasets
Translation samples from EuroParl

We also release the Poro 2 instruction collection.

Direct Preference Optimization (DPO)

The final model underwent preference tuning using the HelpSteer3 dataset to improve response quality and alignment.

Post-Training Hyperparameters

SFT

Hyperparameter	Value
Epochs	2
Global batch size	64
Learning rate	5e-6
LR scheduler	linear
Warmup ratio	0.03
Max sequence length	4,096

DPO

Hyperparameter	Value
Epochs	3
Global batch size	64
Beta	0.01
Learning rate	5e-7
LR scheduler	cosine
Warmup ratio	0.1
Max length	4,096

Evaluation Results

Poro 2 8B Instruct shows substantial improvements in Finnish instruction-following capabilities compared to Llama 3.1 8B Instruct, while maintaining strong English performance. We also outperform Gemma-2-9B-it and EuroLLM-9B-Instruct in Finnish.

Finnish Instruction Following

	Poro 2 8B Instruct	Llama 3.1 8B Instruct	Gemma-2-9B-it	EuroLLM-9B-Instruct
IFEval Finnish	66.54	47.31	55.82	44.17
MTBench Finnish	6.75	4.10	6.7	4.46
AlpacaEval 2 Finnish	28.89	2.05	21.85	8.15

English Instruction Following

	Poro 2 8B Instruct	Llama 3.1 8B Instruct	Gemma-2-9B-it	EuroLLM-9B-Instruct
IFEval	79.29	79.48	72.45	61.36
MTBench	7.33	7.70	7.85	6.25
AlpacaEval 2	35.30	32.70	46.67	15.87

MTBench scores per-category

	Finnish	English
Coding	4.95	5.8
Extraction	6.70	7.35
Humanities	8.55	8.85
Math	5.80	6.7
Reasoning	4.80	5.45
Roleplay	7.65	8.10
STEM	7.55	8.05
Writing	8.05	8.40

MTBench scores per-turn

	Finnish	English
first turn	7.23	7.75
second turn	6.27	6.92

Pairwise Comparisons on MTBench

Finnish: 85% win rate vs Llama 3.1 8B Instruct
Finnish: 51% win rate vs Llama 3.3 70B Instruct
English: 49% win rate vs Llama 3.1 8B Instruct

Overall: ~24% average improvement in Finnish instruction-following benchmarks while maintaining English performance.

Usage

python

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_name = "LumiOpen/Llama-Poro-2-8B-Instruct"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

# Finnish conversation example
messages = [
    {"role": "user", "content": "Kerro minulle Suomen historiasta."}
]

inputs = tokenizer.apply_chat_template(
    messages, 
    add_generation_prompt=True,
    return_tensors="pt"
)

outputs = model.generate(
    inputs,
    max_new_tokens=500,
    temperature=0.7,
    do_sample=True,
    pad_token_id=tokenizer.eos_token_id
)

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Intended Use

Poro 2 8B Instruct is designed for:

Conversational AI applications in Finnish and English
Question answering and information retrieval
Content generation and creative writing
Educational applications
Customer service and support applications
Translation between Finnish and English

Ethical Considerations and Limitations

Poro 2 8B Instruct is an advanced conversational AI model optimized for English and Finnish instruction following. As with most AI-driven systems, this model may reflect imperfections, biases, and idiosyncrasies present in its training data.

Key limitations:

Limited proficiency in languages other than English and Finnish
May occasionally generate biased, inappropriate, or factually incorrect content
Performance may vary significantly for specialized or technical domains
Context window limited to 8,192 tokens
May struggle with very recent events (knowledge cutoff limitations)

Safety Considerations:

Users should verify important factual claims independently
The model should not be used for medical, legal, or financial advice without human oversight
Responses should be reviewed for appropriateness in sensitive contexts

License

Built with Llama

Poro 2 8B Instruct is released under the Llama 3.3 Community License. Please review the license terms before use.

Citation

bibtex

@misc{poro2_2025,
    title={Poro 2: Continued Pretraining for Language Acquisition},
    author={Elaine Zosa and Jouni Louma and Kai Hakala and Antti Virtanen and Mika Koistinen and Risto Luukkonen and Akseli Reunamo and Sampo Pyysalo and Jonathan Burdge},
    year={2025},
    howpublished={LumiOpen}
}

Acknowledgments

We thank CSC - IT Center for Science, Finland for providing access to the LUMI supercomputer. This work was supported by the High Performance Language Technologies (HPLT) project and conducted in collaboration with TurkuNLP from the University of Turku. This project has received funding from the European Union's Horizon Europe research and innovation programme under grant agreement No 101070350.