trinhkhng/nuslerp_Merged_Qwen2-0.5B_0.1

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 6, 2026Architecture:Transformer Featherless Exclusive Cold

trinhkhng/nuslerp_Merged_Qwen2-0.5B_0.1 is a 0.5 billion parameter language model merged from two Qwen2-0.5B variants using the NuSLERP method. This model integrates a debiased Qwen2-0.5B with a standard Qwen2-0.5B, with a 0.1 weight given to the debiased component. It is designed for general language tasks, leveraging its merged architecture to potentially offer nuanced performance characteristics.

Loading preview...

Model Overview

trinhkhng/nuslerp_Merged_Qwen2-0.5B_0.1 is a 0.5 billion parameter language model created by merging two distinct Qwen2-0.5B models. This merge was performed using the NuSLERP method, a technique for combining pre-trained language models to potentially enhance or modify their capabilities.

Merge Details

This model is a composite of:

  • A debiased version of Qwen2-0.5B (/kaggle/working/debias_Qwen2-0.5B)
  • The base Qwen2-0.5B model (/kaggle/working/Qwen2-0.5B)

The NuSLERP merge configuration applied a weight of 0.9 to the base Qwen2-0.5B model and a weight of 0.1 to the debiased Qwen2-0.5B model. This specific weighting suggests an intent to retain most of the base model's characteristics while incorporating a minor influence from the debiased variant. The merge process utilized mergekit and configured nuslerp_flatten to true and nuslerp_row_wise to false.

Potential Use Cases

Given its merged nature and the inclusion of a debiased component, this model could be suitable for:

  • General natural language understanding and generation tasks where a compact model size is beneficial.
  • Applications requiring a balance between performance and a reduced bias profile, influenced by the debiased component.
  • Experimentation with merged model architectures to observe the effects of NuSLERP on Qwen2-0.5B variants.