|
--- |
|
language: |
|
- en |
|
license: apache-2.0 |
|
tags: |
|
- text-generation-inference |
|
- transformers |
|
- unsloth |
|
- llama |
|
- trl |
|
- sft |
|
- llama3 |
|
- llama38b |
|
- 8b |
|
- dolphin |
|
- babydolphin |
|
base_model: unsloth/llama-3-8b-bnb-4bit |
|
datasets: |
|
- cognitivecomputations/dolphin |
|
--- |
|
|
|
|
|
|
|
 |
|
|
|
# BabyDolphin-8B-LLaMA3-Uncensored |
|
|
|
- **Developed by:** babycommando |
|
- **License:** apache-2.0 |
|
- **Finetuned from model :** unsloth/llama-3-8b-bnb-4bit |
|
|
|
This model, `babydolphin-8b-llama3-uncensored`, is an 8-billion parameter subset of the larger LLaMA (Large Language Model by Meta) and has been fine-tuned on the `cognitivecomputations/dolphin` dataset specifically for the FLAN1M-Alpaca-Uncensored tasks. It incorporates cutting-edge transformer architectures optimized for a balance between performance and efficiency. |
|
|
|
## Model Description |
|
|
|
`babydolphin-8b-llama3-uncensored` is designed to deliver powerful language understanding and generation capabilities while ensuring compliance with non-censorship standards for diverse application scenarios. This version is ideal for applications requiring high-quality text generation where content restrictions are minimal. |
|
|
|
### Technical Details |
|
|
|
- **Base Model**: LLaMA3 |
|
- **Parameters**: 8 billion |
|
- **Fine-tuning Dataset**: cognitivecomputations/dolphin FLAN1M-Alpaca-Uncensored |
|
|
|
### Quantization and Configuration |
|
|
|
This model is available in multiple configurations to best suit different deployment needs: |
|
|
|
- **f16**: Fastest conversion, retains 100% accuracy but is slow and memory-intensive. |
|
- **q4_k_m**: Recommended for general use, balancing between speed and efficiency. |
|
- **q3_k_m**: Good for environments where model size and speed are more critical than detailed accuracy. |
|
- **q3_k_s**: Maximizes speed and minimizes model size, suitable for very resource-constrained environments. |
|
|
|
## Intended Use |
|
|
|
This model is intended for researchers and developers needing advanced natural language processing capabilities without censorship restrictions. It is particularly well-suited for generating text in scenarios where nuanced, unrestricted content generation is crucial. |
|
|
|
## How to Use |
|
|
|
For Ollama, check their docs for [running a GGUF model on Ollama](https://github.com/ollama/ollama/blob/main/docs/import.md) |
|
|
|
Here is how to load and use the model in your projects using Hugging Face Transformers: |
|
|
|
```python |
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
|
model_name = "babycommando/babydolphin-8b-llama3-uncensored" |
|
tokenizer = AutoTokenizer.from_pretrained(model_name) |
|
model = AutoModelForCausalLM.from_pretrained(model_name) |
|
|
|
inputs = tokenizer("Hello, world!", return_tensors="pt") |
|
outputs = model.generate(inputs["input_ids"]) |
|
print(tokenizer.decode(outputs[0])) |
|
``` |
|
|
|
### Training Loss Over 60 Epochs |
|
 |
|
|
|
|
|
- |
|
|
|
This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library. |
|
|
|
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth) |
|
|
|
## Usage |
|
|
|
Copy this markdown content into your model's page on the Hugging Face Model Hub to provide users with a clear, informative description of what your model can do and how it can be used. Adjust the `model_name` variable in the Python code snippet to reflect the actual path to your model on Hugging Face for ease of use by others. |
