cstr/Spaetzle-v8-7b

TEXT GENERATIONConcurrent Unit Cost:1Model Size:7BQuant:FP8Context Size:4kTool Calling:SupportedPublished:Mar 10, 2024License:cc-by-sa-4.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

cstr/Spaetzle-v8-7b is a 7B parameter merged language model designed for adequate performance in both German and English. It is a blend of several models, including NeuDist-Ro-7B, Brezn3, and Flora_DPO_7B, based on Wiedervereinigung-7b-dpo-laser, focusing on instruction following and reasoning. This model aims to provide a balanced performance for bilingual tasks, even with some noted weaknesses in German grammar compared to specialized alternatives.

Loading preview...

Spaetzle-v8-7b: A Bilingual Merged Model

cstr/Spaetzle-v8-7b is a 7-billion parameter language model created by cstr through a merge of several existing models using LazyMergekit. Built upon mayflowergmbh/Wiedervereinigung-7b-dpo-laser, it integrates flemmingmiguel/NeuDist-Ro-7B, johannhartmann/Brezn3, and ResplendentAI/Flora_DPO_7B to achieve a balanced performance.

Key Capabilities

  • Bilingual Proficiency: Designed for adequate performance in both German and English tasks.
  • Instruction Following & Reasoning: Prioritizes instruction adherence and logical reasoning over strict grammatical perfection in German.
  • Robustness: Aims to avoid common issues like rambling or template intermixing during generation.
  • Performance Metrics: Achieves an average of 72.27 on the Open LLM Leaderboard, with notable scores like 86.68 on HellaSwag (10-Shot) and 68.16 on GSM8k (5-Shot). It also scores 61.04 on EQ-Bench (v2_de) and 78.3 on EQ-Bench (english v2).
  • Context Length: Supports a context length of 32768 tokens.

Good for

  • Bilingual Applications: Ideal for use cases requiring functional performance in both German and English.
  • Instruction-Driven Tasks: Suitable for applications where precise instruction following and reasoning are more critical than perfect German grammar or orthography.
  • Exploration of Merged Models: Provides an example of a model created via weight merging, offering insights into this development approach.