Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- moe
|
| 4 |
+
- llama
|
| 5 |
+
- '3'
|
| 6 |
+
- llama 3
|
| 7 |
+
- 2x8b
|
| 8 |
+
---
|
| 9 |
+
<img src="https://i.imgur.com/c1Mv8cy.png" width="640"/>
|
| 10 |
+
|
| 11 |
+
# Llama-3-Teal-Instruct-2x8B-MoE
|
| 12 |
+
This is a experimental MoE created from [meta-llama/Meta-Llama-3-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct), [nvidia/Llama3-ChatQA-1.5-8B](https://huggingface.co/nvidia/Llama3-ChatQA-1.5-8B), [Salesforce/SFR-Iterative-DPO-LLaMA-3-8B-R](https://huggingface.co/Salesforce/SFR-Iterative-DPO-LLaMA-3-8B-R) and [Muhammad2003/Llama3-8B-OpenHermes-DPO](https://huggingface.co/Muhammad2003/Llama3-8B-OpenHermes-DPO) using Mergekit.
|
| 13 |
+
|
| 14 |
+
Mergekit yaml file:
|
| 15 |
+
```
|
| 16 |
+
base_model: Meta-Llama-3-8B-Instruct
|
| 17 |
+
experts:
|
| 18 |
+
- source_model: Meta-Llama-3-8B-Instruct
|
| 19 |
+
positive_prompts:
|
| 20 |
+
- "explain"
|
| 21 |
+
- "chat"
|
| 22 |
+
- "assistant"
|
| 23 |
+
- "think"
|
| 24 |
+
- "roleplay"
|
| 25 |
+
- "versatile"
|
| 26 |
+
- "helpful"
|
| 27 |
+
- "factual"
|
| 28 |
+
- "integrated"
|
| 29 |
+
- "adaptive"
|
| 30 |
+
- "comprehensive"
|
| 31 |
+
- "balanced"
|
| 32 |
+
negative_prompts:
|
| 33 |
+
- "specialized"
|
| 34 |
+
- "narrow"
|
| 35 |
+
- "focused"
|
| 36 |
+
- "limited"
|
| 37 |
+
- "specific"
|
| 38 |
+
- source_model: ChatQA-1.5-8B
|
| 39 |
+
positive_prompts:
|
| 40 |
+
- "python"
|
| 41 |
+
- "math"
|
| 42 |
+
- "solve"
|
| 43 |
+
- "code"
|
| 44 |
+
- "programming"
|
| 45 |
+
negative_prompts:
|
| 46 |
+
- "sorry"
|
| 47 |
+
- "cannot"
|
| 48 |
+
- "factual"
|
| 49 |
+
- "concise"
|
| 50 |
+
- "straightforward"
|
| 51 |
+
- "objective"
|
| 52 |
+
- "dry"
|
| 53 |
+
- source_model: SFR-Iterative-DPO-LLaMA-3-8B-R
|
| 54 |
+
positive_prompts:
|
| 55 |
+
- "chat"
|
| 56 |
+
- "assistant"
|
| 57 |
+
- "AI"
|
| 58 |
+
- "instructive"
|
| 59 |
+
- "clear"
|
| 60 |
+
- "directive"
|
| 61 |
+
- "helpful"
|
| 62 |
+
- "informative"
|
| 63 |
+
- source_model: Llama3-8B-OpenHermes-DPO
|
| 64 |
+
positive_prompts:
|
| 65 |
+
- "analytical"
|
| 66 |
+
- "accurate"
|
| 67 |
+
- "logical"
|
| 68 |
+
- "knowledgeable"
|
| 69 |
+
- "precise"
|
| 70 |
+
- "calculate"
|
| 71 |
+
- "compute"
|
| 72 |
+
- "solve"
|
| 73 |
+
- "work"
|
| 74 |
+
- "python"
|
| 75 |
+
- "code"
|
| 76 |
+
- "javascript"
|
| 77 |
+
- "programming"
|
| 78 |
+
- "algorithm"
|
| 79 |
+
- "tell me"
|
| 80 |
+
- "assistant"
|
| 81 |
+
negative_prompts:
|
| 82 |
+
- "creative"
|
| 83 |
+
- "abstract"
|
| 84 |
+
- "imaginative"
|
| 85 |
+
- "artistic"
|
| 86 |
+
- "emotional"
|
| 87 |
+
- "mistake"
|
| 88 |
+
- "inaccurate"
|
| 89 |
+
gate_mode: hidden
|
| 90 |
+
dtype: float16
|
| 91 |
+
```
|