suayptalha/Clarus-7B-v0.2

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kPublished:Feb 24, 2025License:mitArchitecture:Transformer0.0K Open Weights Warm

Clarus-7B-v0.2 by suayptalha is a 7.6 billion parameter language model merged using the Model Stock method, based on gz987/qwen2.5-7b-cabs-v0.3. This model integrates multiple Qwen2.5-7B-CABS versions to enhance its capabilities. It achieves an average score of 36.86 on the Open LLM Leaderboard, with notable performance in IFEval (76.79) and MATH Lvl 5 (48.56).

Loading preview...

Clarus-7B-v0.2: A Merged Language Model

Clarus-7B-v0.2 is a 7.6 billion parameter language model developed by suayptalha. It was created using the Model Stock merge method, leveraging mergekit to combine several pre-trained models. The base model for this merge was gz987/qwen2.5-7b-cabs-v0.3, and it integrates additional versions including gz987/qwen2.5-7b-cabs-v0.4, gz987/qwen2.5-7b-cabs-v0.1, and gz987/qwen2.5-7b-cabs-v0.2.

Key Capabilities & Performance

This model demonstrates a balanced performance across various benchmarks, as indicated by its evaluation on the Open LLM Leaderboard. Key results include:

  • Average Score: 36.86
  • IFEval (0-Shot): 76.79
  • MATH Lvl 5 (4-Shot): 48.56
  • BBH (3-Shot): 36.02
  • MMLU-PRO (5-shot): 37.78

These metrics suggest its potential for tasks requiring instruction following and mathematical reasoning. The model was configured with bfloat16 precision for both input and output, and includes int8_mask and normalize parameters in its merge configuration.

When to Consider Using Clarus-7B-v0.2

Given its merged architecture and benchmark scores, Clarus-7B-v0.2 could be suitable for applications where a 7B-class model with enhanced instruction following and mathematical capabilities is desired. Its performance on IFEval and MATH Lvl 5 indicates a focus on these areas, making it a candidate for tasks requiring precise instruction adherence and numerical problem-solving.