Qwen15-DeepSeek-Coder-Merge

This is a merge of pre-trained language models created using MergeKit, combining the foundational capabilities of Qwen 1.5 with DeepSeek Coder's programming expertise through an efficient SLERP fusion.

About Me

I'm David Soeiro-Vuong, a third-year Computer Science student working as an apprentice at TW3 Partners, a company specialized in Generative AI. Passionate about artificial intelligence and language models optimization, I focus on creating efficient model merges that balance performance and capabilities.

🔗 Connect with me on LinkedIn

Merge Details

Merge Method

This model uses SLERP (Spherical Linear Interpolation) with carefully tuned parameters to achieve optimal performance balance:

• Weighted Blend: t=0.6 provides a slightly stronger influence from the DeepSeek Coder model
• Complete Layer Merging: Full layer-range coverage ensures comprehensive knowledge transfer
• Format: bfloat16 precision for efficient memory usage

Models Merged

• Qwen/Qwen1.5-7B-Chat - Alibaba's Qwen 1.5 chat model known for its strong conversational capabilities and instruction following
• deepseek-ai/deepseek-coder-6.7b-instruct - DeepSeek's specialized coding model with excellent programming language understanding and code generation abilities

Configuration

slices:
  - sources:
      - model: Qwen/Qwen1.5-7B-Chat
        layer_range: [0, 32]
      - model: deepseek-ai/deepseek-coder-6.7b-instruct
        layer_range: [0, 32]
merge_method: slerp
base_model: Qwen/Qwen1.5-7B-Chat
parameters:
  t: 0.6
dtype: bfloat16

Model Capabilities

This merge combines:

• Qwen 1.5's strong instruction following and general knowledge capabilities
• DeepSeek Coder's specialized programming expertise and code generation abilities
• Enhanced technical understanding and explanation capabilities
• Fully open architecture with no usage restrictions

The resulting model provides enhanced performance on tasks requiring both conversational fluency and programming expertise, such as:

• Code generation across multiple programming languages
• Technical documentation and explanations
• Algorithm implementation and problem-solving
• Software development assistance with natural language understanding
• Debugging and code optimization suggestions

Limitations

• Inherits limitations from both base models
• May exhibit inconsistent behavior for certain advanced programming tasks
• No additional alignment or fine-tuning beyond the base models' training
• Model was created through parameter merging without additional training data
• Slight model size mismatch (7B vs 6.7B) may introduce some parameter interpolation artifacts

License

This model is released under the Apache 2.0 license, consistent with the underlying models' licenses.