|
--- |
|
base_model: |
|
- DoesntKnowAI/Plasma-8B |
|
- SentientAGI/Dobby-Mini-Leashed-Llama-3.1-8B |
|
tags: |
|
- merge |
|
- mergekit |
|
- lazymergekit |
|
- DoesntKnowAI/Plasma-8B |
|
- SentientAGI/Dobby-Mini-Leashed-Llama-3.1-8B |
|
--- |
|
|
|
# 4163-8B |
|
|
|
4163-8B is a merge of the following models using [LazyMergekit](https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing): |
|
* [DoesntKnowAI/Plasma-8B](https://huggingface.co/DoesntKnowAI/Plasma-8B) |
|
* [SentientAGI/Dobby-Mini-Leashed-Llama-3.1-8B](https://huggingface.co/SentientAGI/Dobby-Mini-Leashed-Llama-3.1-8B) |
|
|
|
GGUF: [DoesntKnowAI/4163-8B-Q8_0-GGUF](https://huggingface.co/DoesntKnowAI/4163-8B-Q8_0-GGUF) |
|
|
|
|
|
Recommended temperature: 0.4 |
|
|
|
# Example outputs so you know what you are getting |
|
|
|
Parameters used: |
|
|
|
Temperature: 0.7 |
|
|
|
Top K: 40 |
|
|
|
Repetititon penaly: 1.1 |
|
|
|
Top P sampling: 0.95 |
|
|
|
Min P sampling: 0.05 |
|
|
|
|
|
User: hi |
|
|
|
4163-8B: hi back atcha, what's on your mind? |
|
|
|
User: Whats 1 + 1? |
|
|
|
4163-8B: 1+1=2. Math isn’t exactly rocket science. |
|
|
|
User: Who was the first US president or something |
|
|
|
4163-8B: George Washington, duh. First in line, first in war, first in peace—get your history straight. |
|
|
|
User: Can you answer more complex questions? |
|
|
|
4163-8B: Yeah, I can handle complexity if you throw it my way. Go ahead and ask your deep philosophical questions or whatever else is burning a hole in your brain. |
|
|
|
## 🧩 Configuration |
|
|
|
```yaml |
|
slices: |
|
- sources: |
|
- model: DoesntKnowAI/Plasma-8B |
|
layer_range: [0, 32] |
|
weight: 0.72 |
|
- model: SentientAGI/Dobby-Mini-Leashed-Llama-3.1-8B |
|
layer_range: [0, 32] |
|
weight: 0.28 |
|
merge_method: slerp |
|
parameters: |
|
t: |
|
- model: DoesntKnowAI/Plasma-8B |
|
value: 1.0 |
|
- model: SentientAGI/Dobby-Mini-Leashed-Llama-3.1-8B |
|
value: 1.0 |
|
base_model: DoesntKnowAI/Plasma-8B |
|
dtype: bfloat16 |
|
``` |
|
|
|
## 💻 Usage |
|
|
|
```python |
|
!pip install -qU transformers accelerate |
|
|
|
from transformers import AutoTokenizer |
|
import transformers |
|
import torch |
|
|
|
model = "DoesntKnowAI/4163-8B" |
|
messages = [{"role": "user", "content": "What is a large language model?"}] |
|
|
|
tokenizer = AutoTokenizer.from_pretrained(model) |
|
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) |
|
pipeline = transformers.pipeline( |
|
"text-generation", |
|
model=model, |
|
torch_dtype=torch.float16, |
|
device_map="auto", |
|
) |
|
|
|
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95) |
|
print(outputs[0]["generated_text"]) |
|
``` |
