ZeroXClem/Qwen-2.5-Aether-SlerpFusion-7B

Qwen-2.5-Aether-SlerpFusion-7B is a sophisticated model merge that combines the strengths of multiple pre-trained language models using the powerful mergekit framework. This fusion leverages spherical linear interpolation (SLERP) to seamlessly blend architectural layers, resulting in a model that benefits from enhanced performance and versatility.

🚀 Merged Models

This model merge incorporates the following:

• Locutusque/StockQwen-2.5-7B: Serves as the foundational model, renowned for its robust language understanding and generation capabilities.
• allknowingroger/QwenSlerp8-7B: Contributes advanced task-specific fine-tuning, enhancing the model's adaptability across various applications.

🧩 Merge Configuration

The configuration below outlines how the models are merged using spherical linear interpolation (SLERP). This method ensures smooth transitions between the layers of both models, facilitating an optimal blend of their unique attributes:

# ZeroXClem/Qwen-2.5-Aether-SlerpFusion-7B Merge Configuration
slices:
  - sources:
      - model: Locutusque/StockQwen-2.5-7B
        layer_range: [0, 28]
      - model: allknowingroger/QwenSlerp8-7B
        layer_range: [0, 28]
merge_method: slerp
base_model: Locutusque/StockQwen-2.5-7B
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

🔑 Key Parameters

• Self-Attention Filtering (self_attn): Controls the blending extent across self-attention layers, allowing for a dynamic mix between the two source models.
• MLP Filtering (mlp): Adjusts the balance within the Multi-Layer Perceptrons, fine-tuning the model’s neural network layers for optimal performance.
• Global Weight (t.value): Sets a general interpolation factor for all unspecified layers, ensuring an equal contribution from both models.
• Data Type (dtype): Utilizes bfloat16 to maintain computational efficiency while preserving high precision.

🗣️ Inference

Below is an example of how to load and use the model for text generation:

from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import torch

# Define the model name
model_name = "ZeroXClem/Qwen-2.5-Aether-SlerpFusion-7B"

# Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Load the model
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

# Initialize the pipeline
text_generator = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

# Define the input prompt
prompt = "Explain the significance of artificial intelligence in modern healthcare."

# Generate the output
outputs = text_generator(
    prompt,
    max_new_tokens=150,
    do_sample=True,
    temperature=0.7,
    top_k=50,
    top_p=0.95
)

# Print the generated text
print(outputs[0]["generated_text"])

🎯 Use Case & Applications

Qwen-2.5-Aether-SlerpFusion-7B excels in scenarios that require both robust language understanding and specialized task performance. This merged model is ideal for:

• Advanced Text Generation and Comprehension: Crafting coherent, contextually accurate, and nuanced text for applications like content creation, summarization, and translation.
• Domain-Specific Tasks: Enhancing performance in specialized areas such as legal document analysis, medical information processing, and technical support.
• Interactive AI Systems: Powering conversational agents and chatbots that require both general language capabilities and task-specific expertise.

📜 License

This model is open-sourced under the Apache-2.0 License.

💡 Tags

• merge
• mergekit
• slerp
• Qwen
• Locutusque/StockQwen-2.5-7B
• allknowingroger/QwenSlerp8-7B

---

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	29.59
IFEval (0-Shot)	62.62
BBH (3-Shot)	36.01
MATH Lvl 5 (4-Shot)	24.17
GPQA (0-shot)	6.49
MuSR (0-shot)	11.29
MMLU-PRO (5-shot)	36.96