LeoLM/leo-hessianai-13b

TEXT GENERATIONConcurrency Cost:1Model Size:13BQuant:FP8Ctx Length:4kPublished:Sep 5, 2023Architecture:Transformer0.0K Cold

LeoLM/leo-hessianai-13b is a 13 billion parameter causal decoder-only transformer language model developed by LAION and HessianAI. It extends Llama-2's capabilities into German through continued pretraining on a large corpus of German-language text, making it a linguistically enhanced open language model. This model is designed for German-language applications, offering strong performance in both English and German contexts. It is particularly suited for research and commercial LLM development focusing on the German language.

Loading preview...

Overview

LeoLM/leo-hessianai-13b is a 13 billion parameter language model developed by LAION and HessianAI, building upon the Llama-2 architecture. It represents the first open and commercially available German Foundation Language Model, created through continued pretraining on extensive German-language datasets. The model supports both English and German, making it a versatile tool for bilingual applications.

Key Capabilities

  • German Language Enhancement: Significantly extends Llama-2's capabilities for German through specialized pretraining.
  • Bilingual Support: Functions effectively in both English and German.
  • Foundation Model: Serves as a robust base for further fine-tuning and application development.
  • Llama-2 Compatibility: Finetuned from meta-llama/Llama-2-13b-hf, ensuring architectural familiarity and leveraging Llama-2's strengths.
  • Commercial Use: Released under the Llama-2 community license, allowing for commercial applications.

Good For

  • German LLM Research: Accelerating open-source research and development in German natural language processing.
  • Commercial German Applications: Building and deploying LLM-powered solutions tailored for the German market.
  • Bilingual Text Generation: Tasks requiring generation or understanding in both English and German.
  • Continued Pretraining: As a strong base model for further domain-specific or task-specific pretraining.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p