File size: 3,006 Bytes
0b2e266 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 |
---
license: apache-2.0
base_model:
- Trelis/Llama-3.2-1B-Instruct-MATH-3ep
- huihui-ai/Llama-3.2-1B-Instruct-abliterated
- passing2961/Ultron-Summarizer-1B
- unsloth/Llama-3.2-1B-Instruct
tags:
- moe
- frankenmoe
- merge
- mergekit
- lazymergekit
- Trelis/Llama-3.2-1B-Instruct-MATH-3ep
- huihui-ai/Llama-3.2-1B-Instruct-abliterated
- passing2961/Ultron-Summarizer-1B
- unsloth/Llama-3.2-1B-Instruct
---
# DaRuukLLM-Refresh-4x1B-v1
DaRuukLLM-Refresh-4x1B-v1 is a Mixure of Experts (MoE) made with the following models using [LazyMergekit](https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing):
* [Trelis/Llama-3.2-1B-Instruct-MATH-3ep](https://huggingface.co/Trelis/Llama-3.2-1B-Instruct-MATH-3ep)
* [huihui-ai/Llama-3.2-1B-Instruct-abliterated](https://huggingface.co/huihui-ai/Llama-3.2-1B-Instruct-abliterated)
* [passing2961/Ultron-Summarizer-1B](https://huggingface.co/passing2961/Ultron-Summarizer-1B)
* [unsloth/Llama-3.2-1B-Instruct](https://huggingface.co/unsloth/Llama-3.2-1B-Instruct)
## 🧩 Configuration
```yaml
base_model: unsloth/Llama-3.2-1B-Instruct # Base model for self-attention and layer normalization
gate_mode: hidden # Use hidden state representations for MoE gate parameters
dtype: bfloat16 # Output data type for the merged model
experts:
- source_model: Trelis/Llama-3.2-1B-Instruct-MATH-3ep # Expert for math-related tasks
positive_prompts:
- "Solve the following math problem:"
- "Calculate the value of:"
- "What is the result of:"
- source_model: huihui-ai/Llama-3.2-1B-Instruct-abliterated # Expert for uncensored queries
positive_prompts:
- "Explain the following controversial topic:"
- "Discuss the implications of:"
- "Provide an uncensored analysis of:"
- source_model: passing2961/Ultron-Summarizer-1B # Expert for summarization tasks
positive_prompts:
- "Summarize the following text:"
- "Provide a concise summary of:"
- "Generate a brief overview of:"
- source_model: unsloth/Llama-3.2-1B-Instruct # Base model also acts as the chat expert
positive_prompts:
- "How can I assist you today?"
- "What would you like to discuss?"
- "Let's have a conversation about:"
```
## 💻 Usage
```python
!pip install -qU transformers bitsandbytes accelerate
from transformers import AutoTokenizer
import transformers
import torch
model = "Xiaojian9992024/DaRuukLLM-Refresh-4x1B-v1"
tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
"text-generation",
model=model,
model_kwargs={"torch_dtype": torch.float16, "load_in_4bit": True},
)
messages = [{"role": "user", "content": "Explain what a Mixture of Experts is in less than 100 words."}]
prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
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"])
``` |