mengleap-stnap/Qwenseek-coding-Merged-GGUF

---

Qwenseek-Coding-Merged-1.5B

Qwenseek-Coding-Merged-1.5B is a merged model combining the robust code-generation capabilities of Qwen2.5-Coder-1.5B-Instruct with the deep reasoning and logic skills of DeepSeek-R1-Distill-Qwen-1.5B. It was merged using the SLERP (Spherical Linear Interpolation) method via mergekit.

This model is engineered to achieve a sweet spot between accurate code syntax generation and deep reasoning logic for bug fixing.

---

Model Capabilities & Benchmarks

Key improvements achieved through this merge:

• Logic & Bug Fixing: Significantly enhanced compared to the original Qwen2.5-Coder, thanks to DeepSeek-R1's reasoning distillation.
• Coding Syntax & Multi-language: Preserves strong code generation accuracy without performance degradation.
• Efficiency: At only 1.5B parameters, it runs extremely fast and requires minimal memory/VRAM.

---

Merge Configuration

The merge was executed using mergekit with the following configuration:

slices:
  - sources:
      - model: Qwen/Qwen2.5-Coder-1.5B-Instruct
        layer_range: [0, 28]
        weight: 0.6
      - model: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
        layer_range: [0, 28]
        weight: 0.4
merge_method: slerp
base_model: Qwen/Qwen2.5-Coder-1.5B-Instruct
parameters:
  t:
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 1]
    - filter: mlp
      value: [1, 0.5, 0.7, 0.3, 0]
    - value: 0.5
dtype: bfloat16