mamii76/Snowball-7B-Hybrid-HF

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 21, 2026Architecture:Transformer Featherless Exclusive Cold

Snowball-7B-Hybrid-HF is a 7.6 billion parameter language model created by mamii76, merged using the TIES method from Qwen/Qwen2.5-7B as its base. This hybrid model integrates capabilities from OpenThinker2-7B and Qwen2.5-Coder-7B-Instruct, offering a blend of general reasoning and specialized coding instruction. With a 32768 token context length, it is designed for tasks requiring both broad understanding and precise code generation.

Loading preview...

Model Overview

Snowball-7B-Hybrid-HF is a 7.6 billion parameter language model developed by mamii76. It was created using the TIES merge method, combining the strengths of multiple pre-trained models. The base model for this merge is Qwen/Qwen2.5-7B, providing a robust foundation.

Key Capabilities

This model is a hybrid, integrating two distinct models to enhance its performance:

  • open-thoughts/OpenThinker2-7B: Contributes to general reasoning and understanding capabilities.
  • Qwen/Qwen2.5-Coder-7B-Instruct: Specializes in coding tasks and instruction following, leveraging its instruction-tuned nature.

Merge Details

The merge process utilized a specific configuration with equal weighting (0.5) and density (0.5) for both OpenThinker2-7B and Qwen2.5-Coder-7B-Instruct. The dtype was set to bfloat16, and the tokenizer source was Qwen/Qwen2.5-Coder-7B-Instruct. This strategic merge aims to create a model proficient in both general language understanding and specialized coding applications.

Potential Use Cases

Given its hybrid nature, Snowball-7B-Hybrid-HF is suitable for applications that require:

  • General-purpose text generation and comprehension.
  • Code generation, completion, and explanation.
  • Instruction-following tasks across various domains.
  • Scenarios benefiting from a blend of reasoning and coding expertise.