HiTZ/Latxa-Llama-3.1-70B-Instruct

TEXT GENERATIONConcurrency Cost:4Model Size:70BQuant:FP8Ctx Length:32kPublished:Mar 28, 2025License:otherArchitecture:Transformer0.0K Cold

HiTZ/Latxa-Llama-3.1-70B-Instruct is a 70 billion parameter instruction-tuned causal language model developed by HiTZ Research Center & IXA Research group. Based on Meta's Llama-3.1-Instruct, it has been further trained on a 4.2 billion token Basque corpus using language adaptation techniques. This model significantly outperforms its parent model on Basque benchmarks and chat conversations, making it highly optimized for Basque language tasks and instruction following.

Loading preview...

Latxa-Llama-3.1-70B-Instruct Overview

Latxa-Llama-3.1-70B-Instruct is a 70 billion parameter instruction-tuned large language model developed by the HiTZ Research Center & IXA Research group. It is built upon Meta's Llama-3.1-Instruct and has undergone further training using language adaptation techniques on a substantial Basque corpus (4.3 million documents, 4.2 billion tokens).

Key Capabilities & Differentiators

  • Basque Language Specialization: Specifically designed and optimized for the Basque language, addressing the performance gap observed in low-resource languages with general-purpose LLMs.
  • Superior Basque Performance: Preliminary evaluations show it significantly outperforms the base Llama-3.1-Instruct model on standard Basque benchmarks and in chat conversations.
  • Instruction Following: Trained to follow instructions and function as a chat assistant, making it suitable for interactive applications.
  • Competitive Arena Performance: Achieved 3rd place in a public arena-based evaluation against models like GPT-4o and Claude Sonnet, outperforming other same-size competitors.

Intended Use Cases

  • Basque Language Applications: Ideal for any application requiring high-quality natural language processing or generation in Basque.
  • Instruction Following & Chatbots: Suitable for building instruction-following agents or conversational AI systems in Basque.

Limitations

  • Performance is not guaranteed for languages other than Basque.
  • Inherits potential biases and limitations from its parent Llama 3.1 models.

Finetunes
2 models