Update README.md
Browse files
README.md
CHANGED
|
@@ -50,18 +50,52 @@ metrics:
|
|
| 50 |
- **Dataset Combined Using:** Mosher-R1(Propietary Software)
|
| 51 |
|
| 52 |
- MiniMaid-L1 Official Metric Score
|
| 53 |
-
,
|
| 61 |
To Properly Showcase the differences and strength of the Models
|
| 62 |
|
| 63 |
-
- **
|
| 64 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 65 |
|
| 66 |
- # Notice
|
| 67 |
- **For a Good Experience, Please use**
|
|
@@ -78,4 +112,4 @@ metrics:
|
|
| 78 |
- This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
| 79 |
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
| 80 |
- Fine-tuned Using:
|
| 81 |
-
- Google Colab
|
|
|
|
| 50 |
- **Dataset Combined Using:** Mosher-R1(Propietary Software)
|
| 51 |
|
| 52 |
- MiniMaid-L1 Official Metric Score
|
| 53 |
+
|
| 54 |
- Metrics Made By **ItsMeDevRoland**
|
| 55 |
Which compares:
|
| 56 |
- **Deepseek R1 3B GGUF**
|
| 57 |
- **Dolphin 3B GGUF**
|
| 58 |
- **Hermes 3b Llama GGUFF**
|
| 59 |
+
- **MiniMaid-L1 GGUFF**
|
| 60 |
Which are All Ranked with the Same Prompt, Same Temperature, Same Hardware(Google Colab),
|
| 61 |
To Properly Showcase the differences and strength of the Models
|
| 62 |
|
| 63 |
+
- **Visit Below to See details!**
|
| 64 |
+
|
| 65 |
+
---
|
| 66 |
+
|
| 67 |
+
## 🧵 MiniMaid-L1: A 1B Roleplay Assistant That Punches Above Its Weight
|
| 68 |
+
|
| 69 |
+
> She’s not perfect — but she’s fast, compact, and learning quick. And most importantly, **she didn’t suck**.
|
| 70 |
+
|
| 71 |
+
Despite her size, MiniMaid-L1 held her own against 3B models like **DeepSeek**, **Dolphin**, and **Hermes**.
|
| 72 |
+
|
| 73 |
+
💬 **Roleplay Evaluation (v0)**
|
| 74 |
+
- 🧠 Character Consistency: 0.50
|
| 75 |
+
- 🌊 Immersion: 0.13
|
| 76 |
+
- 🧮 Overall RP Score: 0.51
|
| 77 |
+
- ✏️ Length Score: 0.91
|
| 78 |
+
|
| 79 |
+
- Even with only 1.5K synthetic samples, MiniMaid showed strong prompt structure, consistency, and resilience.
|
| 80 |
+
---
|
| 81 |
+
|
| 82 |
+
📊 **Efficiency Wins**
|
| 83 |
+
- **Inference Time:** 49.1s (vs Hermes: 140.6s)
|
| 84 |
+
- **Tokens/sec:** 7.15 (vs Dolphin: 3.88)
|
| 85 |
+
- **BLEU/ROUGE-L:** Outperformed DeepSeek + Hermes
|
| 86 |
+
|
| 87 |
+
- MiniMaid proved that **you don’t need 3 billion parameters to be useful** — just smart distillation and a little love.
|
| 88 |
+
---
|
| 89 |
+
🛠️ **MiniMaid is Built For**
|
| 90 |
+
- Lightweight RP generation
|
| 91 |
+
- Low-resource hardware
|
| 92 |
+
- High customization potential
|
| 93 |
+
|
| 94 |
+
🌱 **She’s just getting started** — v1 is on the way with more character conditioning, dialogue tuning, and narrative personality control.
|
| 95 |
+
|
| 96 |
+
---
|
| 97 |
+
> “She’s scrappy, she’s stubborn, and she’s still learning. But MiniMaid-L1 proves that smart distillation and a tiny budget can go a long way — and she’s only going to get better from here.”
|
| 98 |
+
---
|
| 99 |
|
| 100 |
- # Notice
|
| 101 |
- **For a Good Experience, Please use**
|
|
|
|
| 112 |
- This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
| 113 |
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
| 114 |
- Fine-tuned Using:
|
| 115 |
+
- Google Colab
|