YFC-112358/Qwen3.8-27B-Della-Carnice-Ostrich-Salience-Glimmer-v5

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 27, 2026Architecture:Transformer0.0K Featherless Exclusive Cold

YFC-112358/Qwen3.8-27B-Della-Carnice-Ostrich-Salience-Glimmer-v5 is a 27 billion parameter language model developed by YFC-112358, built upon the Qwen/Qwen3.8-27B base model. It utilizes a single-stage 'della_linear' merge method, combining four distinct source models to enhance its capabilities. This model is designed for general language tasks, leveraging its merged architecture to potentially offer improved performance over its base. Its 32768 token context length supports processing extensive inputs for various applications.

Loading preview...

Model Overview

YFC-112358/Qwen3.8-27B-Della-Carnice-Ostrich-Salience-Glimmer-v5 is a 27 billion parameter language model derived from the Qwen/Qwen3.8-27B base. It employs a unique single-stage della_linear merge method, integrating four distinct source models: kai-os/Carnice-V3, etemiz/Ostrich-27B-Qwen3.8-260815, vectionlabs/Salience-27B-R5, and YFC-112358/Qwen3.8-27B-TM-Gemma4-Glimmer-v1. This merging strategy aims to combine the strengths of these individual models into a cohesive unit.

Key Technical Details

  • Base Model: Qwen/Qwen3.8-27B
  • Merge Method: della_linear with bfloat16 dtype for computation.
  • Parameter Blending: The merge process involves specific weight, density, and epsilon parameters for each source model, influencing their contribution to the final model. Notably, normalize: true ensures that the incremental contribution of each tensor is scaled by the sum of weights of the active sources.
  • Pruning and Rescaling: Magnitude pruning is applied within row blocks, and surviving terms are rescaled by their selection probability to maintain unbiasedness.
  • Tokenizer: The tokenizer and configuration are inherited directly from the Qwen/Qwen3.8-27B base model, ensuring consistency.
  • Context Length: Supports a context length of 32768 tokens.

Unique Characteristics

This model's primary differentiation lies in its sophisticated della_linear merging approach, which does not perform sign election. This means that contributions from minority sources are not overridden by majority sources, allowing for a more direct combination of features from all four integrated models. The merging process is designed for reproducibility, using a specific random seed and tensor-level generation.

When to Use This Model

This model is suitable for users seeking a 27B parameter model that leverages a complex, multi-source merging strategy to potentially achieve enhanced performance over its base. Its design suggests a focus on integrating diverse capabilities from its constituent models, making it a candidate for general-purpose language generation and understanding tasks where a nuanced blend of model characteristics is desired.

Finetunes
2 models