This LLAMA-3 model was finetuned to enhance capabilities in text generation for nuclear-related topics. The training was accelerated using Unsloth and Huggingface's TRL library, achieving a 2x faster performance.

Finetuning Process

The model was finetuned using the Unsloth library, leveraging its efficient training capabilities. The process included the following steps:

Data Preparation: Loaded and preprocessed nuclear-related data.
Model Loading: Utilized unsloth/llama-3-8b-bnb-4bit as the base model.
LoRA Patching: Applied LoRA (Low-Rank Adaptation) for efficient training.
Training: Finetuned the model using Hugging Face's TRL library with optimized hyperparameters.

Model Details

Base Model: unsloth/llama-3-8b-bnb-4bit
Language: English (en)
License: Apache-2.0

Files and Versions

File Name	Description
.gitattributes	Initial commit
README.md	Model description and usage
adapter_config.json	Configuration for adapter
adapter_model.safetensors	Finetuned model weights
config.json	Configuration for base model
generation_config.json	Generation configuration for model
model-00001-of-00007.safetensors	Part of the base model weights
model-00002-of-00007.safetensors	Part of the base model weights
model-00003-of-00007.safetensors	Part of the base model weights
model-00004-of-00007.safetensors	Part of the base model weights
model-00005-of-00007.safetensors	Part of the base model weights
model-00006-of-00007.safetensors	Part of the base model weights
model-00007-of-00007.safetensors	Part of the base model weights
model.safetensors.index.json	Index for the model weights
special_tokens_map.json	Special tokens mapping
tokenizer.json	Tokenizer data
tokenizer_config.json	Configuration for tokenizer

Model Card Authors

MUSTAFA UMUT OZBEK

Contact

https://www.linkedin.com/in/mustafaumutozbek/ https://x.com/m_umut_ozbek

Usage

Loading the Model

You can load the model and tokenizer using the following code snippet:

python

from transformers import AutoModelForCausalLM, AutoTokenizer

# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("inetnuc/llama-3-8b-chat-nuclear")
model = AutoModelForCausalLM.from_pretrained("inetnuc/llama-3-8b-chat-nuclear")

# Example of generating text
inputs = tokenizer("what is the iaea approach for cyber security?", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))