Son of Rhodia

Introduction

Son-of-Rhodia is a text generation model developed with a focus on merging pre-trained language models to enhance performance. This model utilizes the SLERP merge method, offering a consolidated output from different model architectures.

Architecture

The model is a result of merging two distinct language models:

• Infermatic/MN-12B-Inferor-v0.1
• allura-org/MN-12b-RP-Ink

These models have been integrated using a structured merging technique that leverages specific layer configurations and parameter adjustments.

Training

Son-of-Rhodia employs the SLERP merge method, which blends model layers from the two base models. This involves the use of specific YAML configurations to define the layer ranges and merging parameters. The configuration specifies different filters and values for self_attn and mlp, utilizing a bfloat16 data type for precision.

Guide: Running Locally

To run Son-of-Rhodia locally, follow these steps:

1. Clone the Model Repository:

   git clone https://huggingface.co/TheDrunkenSnail/Son-of-Rhodia
   cd Son-of-Rhodia
   

2. Install Dependencies: Ensure you have Python and the transformers library installed:

   pip install transformers
   

3. Load and Use the Model: Use the Hugging Face transformers library to load the model.

   from transformers import AutoModelForCausalLM, AutoTokenizer
   
   tokenizer = AutoTokenizer.from_pretrained("TheDrunkenSnail/Son-of-Rhodia")
   model = AutoModelForCausalLM.from_pretrained("TheDrunkenSnail/Son-of-Rhodia")
   
   input_text = "Your input text here"
   inputs = tokenizer(input_text, return_tensors="pt")
   outputs = model.generate(**inputs)
   print(tokenizer.decode(outputs[0]))
   

4. Consider Cloud GPUs: For enhanced performance, consider using cloud GPU services such as AWS, Google Cloud, or Azure, which provide scalable resources for model inference.

License

The Son-of-Rhodia model is distributed under an unspecified license labeled as "other." Please review the license terms on the Hugging Face model page for specific usage guidelines.