trinhkhng/linear_Merged_Qwen2-0.5B_0.5

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/linear_Merged_Qwen2-0.5B_0.5 is a 0.5 billion parameter language model created by trinhkhng, merged using the Linear method. This model combines a base Qwen2-0.5B with a debiased version of Qwen2-0.5B, aiming to integrate their respective characteristics. It is suitable for applications requiring a compact model with potentially balanced properties derived from its merged components.

Loading preview...

Model Overview

trinhkhng/linear_Merged_Qwen2-0.5B_0.5 is a compact 0.5 billion parameter language model, developed by trinhkhng. This model was constructed using the Linear merge method via mergekit, combining two distinct versions of the Qwen2-0.5B architecture.

Merge Details

The model integrates two primary components:

  • A base /kaggle/working/Qwen2-0.5B model.
  • A debiased version, /kaggle/working/debias_Qwen2-0.5B.

Each component contributed with a weight of 0.5, and the merge process included normalization. This approach suggests an intent to balance the characteristics of the original Qwen2-0.5B with those of its debiased counterpart, potentially offering a model with improved fairness or reduced biases while retaining core capabilities.

Potential Use Cases

Given its small parameter count and the merging strategy, this model could be suitable for:

  • Resource-constrained environments: Its 0.5B parameters make it efficient for deployment on devices with limited computational power.
  • Exploratory tasks: Ideal for experimenting with merged model behaviors, especially concerning bias mitigation.
  • Specific downstream applications: Where a balance between general language understanding and reduced bias is beneficial, without requiring the scale of larger models.