Model Overview

LumiOpen/Llama-Poro-2-70B-Instruct is a 70.55 billion parameter instruction-following chatbot model, a collaborative effort by AMD Silo AI, the TurkuNLP group, and High Performance Language Technologies (HPLT). It is built upon the Llama 3.1 70B architecture and has been extensively fine-tuned using Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) for conversational AI applications.

Key Capabilities

• Bilingual Proficiency: Designed for high-performance conversational AI and instruction following in both Finnish and English.
• Enhanced Finnish Performance: Demonstrates substantial improvements in Finnish instruction-following capabilities, outperforming Llama 3.1 70B Instruct and Llama 3.3 70B Instruct on metrics like MTBench Finnish and AlpacaEval 2 Finnish.
• Strong English Performance: Maintains excellent English performance, on par with or exceeding Llama 3.3 70B Instruct.
• Robust Training: Underwent continued pretraining on 165 billion tokens of Finnish, English, code, and math data, followed by SFT with 1.4 million instruction-following examples and DPO using the HelpSteer3 dataset.
• Context Length: Supports a maximum sequence length of 8192 tokens.

Good For

• High-performance conversational AI applications requiring strong bilingual (Finnish/English) capabilities.
• Question answering, information retrieval, and content generation.
• Educational applications, customer service, and support systems.
• Research and enterprise applications demanding robust multilingual LLM performance.