en license: apache-2.0 library_name: transformers tags:
merge
mergekit
lazymergekit
creative
roleplay
instruct
qwen
model_stock
bfloat16 base_model:
newsbang/Homer-v0.5-Qwen2.5-7B
allknowingroger/HomerSlerp1-7B
bunnycore/Qwen2.5-7B-Instruct-Fusion
bunnycore/Qandora-2.5-7B-Creative model-index:
name: Qwen2.5-7B-HomerCreative-Mix results:
- task: type: text-generation name: Text Generation dataset: name: IFEval (0-Shot) type: HuggingFaceH4/ifeval args: num_few_shot: 0 metrics:
  - type: inst_level_strict_acc and prompt_level_strict_acc value: 78.35 name: strict accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ZeroXClem/Qwen2.5-7B-HomerCreative-Mix name: Open LLM Leaderboard
- task: type: text-generation name: Text Generation dataset: name: BBH (3-Shot) type: BBH args: num_few_shot: 3 metrics:
  - type: acc_norm value: 36.77 name: normalized accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ZeroXClem/Qwen2.5-7B-HomerCreative-Mix name: Open LLM Leaderboard
- task: type: text-generation name: Text Generation dataset: name: MATH Lvl 5 (4-Shot) type: hendrycks/competition_math args: num_few_shot: 4 metrics:
  - type: exact_match value: 32.33 name: exact match source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ZeroXClem/Qwen2.5-7B-HomerCreative-Mix name: Open LLM Leaderboard
- task: type: text-generation name: Text Generation dataset: name: GPQA (0-shot) type: Idavidrein/gpqa args: num_few_shot: 0 metrics:
  - type: acc_norm value: 6.6 name: acc_norm source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ZeroXClem/Qwen2.5-7B-HomerCreative-Mix name: Open LLM Leaderboard
- task: type: text-generation name: Text Generation dataset: name: MuSR (0-shot) type: TAUR-Lab/MuSR args: num_few_shot: 0 metrics:
  - type: acc_norm value: 13.77 name: acc_norm source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ZeroXClem/Qwen2.5-7B-HomerCreative-Mix name: Open LLM Leaderboard
- task: type: text-generation name: Text Generation dataset: name: MMLU-PRO (5-shot) type: TIGER-Lab/MMLU-Pro config: main split: test args: num_few_shot: 5 metrics:
  - type: acc value: 38.3 name: accuracy source: url: https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=ZeroXClem/Qwen2.5-7B-HomerCreative-Mix name: Open LLM Leaderboard

ZeroXClem/Qwen2.5-7B-HomerCreative-Mix

ZeroXClem/Qwen2.5-7B-HomerCreative-Mix is an advanced language model meticulously crafted by merging four pre-trained models using the powerful mergekit framework. This fusion leverages the Model Stock merge method to combine the creative prowess of Qandora, the instructive capabilities of Qwen-Instruct-Fusion, the sophisticated blending of HomerSlerp1, and the foundational conversational strengths of Homer-v0.5-Qwen2.5-7B. The resulting model excels in creative text generation, contextual understanding, and dynamic conversational interactions.

🚀 Merged Models

This model merge incorporates the following:

bunnycore/Qandora-2.5-7B-Creative: Specializes in creative text generation, enhancing the model's ability to produce imaginative and diverse content.
bunnycore/Qwen2.5-7B-Instruct-Fusion: Focuses on instruction-following capabilities, improving the model's performance in understanding and executing user commands.
allknowingroger/HomerSlerp1-7B: Utilizes spherical linear interpolation (SLERP) to blend model weights smoothly, ensuring a harmonious integration of different model attributes.
newsbang/Homer-v0.5-Qwen2.5-7B: Acts as the foundational conversational model, providing robust language comprehension and generation capabilities.

🧩 Merge Configuration

The configuration below outlines how the models are merged using the Model Stock method. This approach ensures a balanced and effective integration of the unique strengths from each source model.

yaml

# Merge configuration for ZeroXClem/Qwen2.5-7B-HomerCreative-Mix using Model Stock

models:
  - model: bunnycore/Qandora-2.5-7B-Creative
  - model: bunnycore/Qwen2.5-7B-Instruct-Fusion
  - model: allknowingroger/HomerSlerp1-7B
merge_method: model_stock
base_model: newsbang/Homer-v0.5-Qwen2.5-7B
normalize: false
int8_mask: true
dtype: bfloat16

Key Parameters

Merge Method (merge_method): Utilizes the Model Stock method, as described in Model Stock, to effectively combine multiple models by leveraging their strengths.
Models (models): Specifies the list of models to be merged:
- bunnycore/Qandora-2.5-7B-Creative: Enhances creative text generation.
- bunnycore/Qwen2.5-7B-Instruct-Fusion: Improves instruction-following capabilities.
- allknowingroger/HomerSlerp1-7B: Facilitates smooth blending of model weights using SLERP.
Base Model (base_model): Defines the foundational model for the merge, which is newsbang/Homer-v0.5-Qwen2.5-7B in this case.
Normalization (normalize): Set to false to retain the original scaling of the model weights during the merge.
INT8 Mask (int8_mask): Enabled (true) to apply INT8 quantization masking, optimizing the model for efficient inference without significant loss in precision.
Data Type (dtype): Uses bfloat16 to maintain computational efficiency while ensuring high precision.

🏆 Performance Highlights

Creative Text Generation: Enhanced ability to produce imaginative and diverse content suitable for creative writing, storytelling, and content creation.
Instruction Following: Improved performance in understanding and executing user instructions, making the model more responsive and accurate in task execution.
Optimized Inference: INT8 masking and bfloat16 data type contribute to efficient computation, enabling faster response times without compromising quality.

🎯 Use Case & Applications

ZeroXClem/Qwen2.5-7B-HomerCreative-Mix is designed to excel in environments that demand both creative generation and precise instruction following. Ideal applications include:

Creative Writing Assistance: Aiding authors and content creators in generating imaginative narratives, dialogues, and descriptive text.
Interactive Storytelling and Role-Playing: Enhancing dynamic and engaging interactions in role-playing games and interactive storytelling platforms.
Educational Tools and Tutoring Systems: Providing detailed explanations, answering questions, and assisting in educational content creation with contextual understanding.
Technical Support and Customer Service: Offering accurate and contextually relevant responses in technical support scenarios, improving user satisfaction.
Content Generation for Marketing: Creating compelling and diverse marketing copy, social media posts, and promotional material with creative flair.

📝 Usage

To utilize ZeroXClem/Qwen2.5-7B-HomerCreative-Mix, follow the steps below:

Installation

First, install the necessary libraries:

pip install -qU transformers accelerate

Example Code

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

python

from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import torch

# Define the model name
model_name = "ZeroXClem/Qwen2.5-7B-HomerCreative-Mix"

# 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 = "Once upon a time in a land far, far away,"

# 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"])

Notes

Fine-Tuning: This merged model may require fine-tuning to optimize performance for specific applications or domains.
Resource Requirements: Ensure that your environment has sufficient computational resources, especially GPU-enabled hardware, to handle the model efficiently during inference.
Customization: Users can adjust parameters such as temperature, top_k, and top_p to control the creativity and diversity of the generated text.

📜 License

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

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	34.35
IFEval (0-Shot)	78.35
BBH (3-Shot)	36.77
MATH Lvl 5 (4-Shot)	32.33
GPQA (0-shot)	6.60
MuSR (0-shot)	13.77
MMLU-PRO (5-shot)	38.30

RichardErkhov/ZeroXClem_-_Qwen2.5-7B-HomerCreative-Mix-4bits

Original model description: