stellia/stellialm_mini_qwen_9tasks
Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:3.1BQuant:BF16Ctx Length:32kArchitecture:Transformer Warm

The stellia/stellialm_mini_qwen_9tasks is a 3.1 billion parameter language model, fine-tuned from Qwen/Qwen2.5-3B-Instruct with a specific LoRA adapter. It is optimized for high accuracy across 9 specific tasks in both English and French, including summarization, keyword extraction, and question answering. This model is designed to specialize small LLMs for client-specific needs while maintaining general task quality, offering performance comparable to larger models like GPT-mini on its target tasks.

Loading preview...

stellia/stellialm_mini_qwen_9tasks Overview

This model is a public release of a 3.1 billion parameter language model, fine-tuned by Stellia from the Qwen/Qwen2.5-3B-Instruct base using a specialized LoRA adapter. Its core focus is to deliver high accuracy on 9 predefined tasks in both English and French.

Key Capabilities & Performance

The model demonstrates strong performance across its target tasks, with an overall score of 0.8 in Stellia's internal evaluation pipeline. Specific task scores include:

  • Summarization: 0.92
  • Query Reformulation: 0.83
  • Fill-in Generation: 0.86
  • MCQ (Multiple Choice Questions): 0.86
  • Keyword Extraction: 0.78
  • Answer Reformulation: 0.73
  • GQA (General Question Answering): 0.81
  • True/False: 0.77
  • Keyword Update: 0.66

This model aims to provide specialized small LLM solutions for client-specific requirements, maintaining high quality on general tasks while competing effectively with larger models like GPT-mini on its specialized benchmarks.

Use Cases

This model is particularly well-suited for applications requiring precise and efficient execution of the 9 specified tasks in English and French, making it ideal for scenarios where specialized performance from a compact model is critical.