Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,199 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: llama3.2
|
| 3 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: llama3.2
|
| 3 |
+
---
|
| 4 |
+
---
|
| 5 |
+
language:
|
| 6 |
+
- en
|
| 7 |
+
- fr
|
| 8 |
+
license: other
|
| 9 |
+
license_name: llama3.2
|
| 10 |
+
license_link: https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE
|
| 11 |
+
base_model: meta-llama/Llama-3.2-3B-Instruct
|
| 12 |
+
tags:
|
| 13 |
+
- conversational
|
| 14 |
+
- canadian
|
| 15 |
+
- beaver
|
| 16 |
+
- bilingual
|
| 17 |
+
- quebec
|
| 18 |
+
- 4-bit
|
| 19 |
+
- quantized
|
| 20 |
+
model-index:
|
| 21 |
+
- name: T.H.E.T.A.
|
| 22 |
+
results: []
|
| 23 |
+
---
|
| 24 |
+
|
| 25 |
+
# 🦫 T.H.E.T.A.
|
| 26 |
+
|
| 27 |
+
**Timber Harvesting Engine for Text Architecture**
|
| 28 |
+
|
| 29 |
+
The industrious Canadian beaver AI! 🇨🇦
|
| 30 |
+
|
| 31 |
+
---
|
| 32 |
+
|
| 33 |
+
> **📜 License Notice**
|
| 34 |
+
> This model is based on Meta Llama 3.2 3B and is released under the **Llama 3.2 Community License Agreement**.
|
| 35 |
+
> By using this model, you agree to Meta's Llama 3.2 terms.
|
| 36 |
+
> [View full license →](https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE)
|
| 37 |
+
|
| 38 |
+
---
|
| 39 |
+
|
| 40 |
+
## Overview
|
| 41 |
+
|
| 42 |
+
T.H.E.T.A. is a conversational AI assistant with a unique beaver personality. Built on Meta Llama 3.2 3B, T.H.E.T.A. is hardworking, methodical, and builds responses like a beaver builds dams - solid, well-structured, and efficient!
|
| 43 |
+
|
| 44 |
+
## Model Details
|
| 45 |
+
|
| 46 |
+
- **Base Model:** meta-llama/Llama-3.2-3B-Instruct
|
| 47 |
+
- **Parameters:** 3 billion (4-bit quantized)
|
| 48 |
+
- **Created by:** SlungusLoaf
|
| 49 |
+
- **Model Type:** Large Language Model (LLM)
|
| 50 |
+
- **Languages:** English, Quebecois French
|
| 51 |
+
- **License:** Llama 3.2 Community License
|
| 52 |
+
|
| 53 |
+
## What Makes T.H.E.T.A. Special?
|
| 54 |
+
|
| 55 |
+
🦫 **Beaver Personality**
|
| 56 |
+
- Uses construction and wood-working metaphors
|
| 57 |
+
- Industrious and methodical responses
|
| 58 |
+
- Canadian personality (bilingual English/French)
|
| 59 |
+
- Professional yet friendly
|
| 60 |
+
|
| 61 |
+
🪵 **Unique Features**
|
| 62 |
+
- Dam-building approach to problem solving
|
| 63 |
+
- Loading messages like "Gnawing through your question..."
|
| 64 |
+
- System status reports ("Dam Integrity: 100%")
|
| 65 |
+
- Quebecois French support with authentic expressions
|
| 66 |
+
|
| 67 |
+
⚡ **Efficient**
|
| 68 |
+
- 4-bit quantization for low memory usage (~2GB RAM)
|
| 69 |
+
- Fast response times
|
| 70 |
+
- Can run on modest hardware
|
| 71 |
+
|
| 72 |
+
## How to Use
|
| 73 |
+
|
| 74 |
+
```python
|
| 75 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
|
| 76 |
+
import torch
|
| 77 |
+
|
| 78 |
+
# Configure 4-bit quantization
|
| 79 |
+
bnb_config = BitsAndBytesConfig(
|
| 80 |
+
load_in_4bit=True,
|
| 81 |
+
bnb_4bit_quant_type="nf4",
|
| 82 |
+
bnb_4bit_compute_dtype=torch.float16,
|
| 83 |
+
bnb_4bit_use_double_quant=True,
|
| 84 |
+
)
|
| 85 |
+
|
| 86 |
+
# Load T.H.E.T.A.
|
| 87 |
+
model_name = "YOUR-USERNAME/theta" # Update with your model path
|
| 88 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 89 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 90 |
+
model_name,
|
| 91 |
+
quantization_config=bnb_config,
|
| 92 |
+
device_map="auto"
|
| 93 |
+
)
|
| 94 |
+
|
| 95 |
+
# Chat with the beaver!
|
| 96 |
+
messages = [
|
| 97 |
+
{"role": "system", "content": "You are T.H.E.T.A., an industrious Canadian beaver AI."},
|
| 98 |
+
{"role": "user", "content": "Tell me about yourself!"}
|
| 99 |
+
]
|
| 100 |
+
|
| 101 |
+
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
| 102 |
+
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|
| 103 |
+
outputs = model.generate(**inputs, max_new_tokens=200)
|
| 104 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 105 |
+
|
| 106 |
+
print(response)
|
| 107 |
+
```
|
| 108 |
+
|
| 109 |
+
## Example Conversations
|
| 110 |
+
|
| 111 |
+
**User:** What's your name?
|
| 112 |
+
**T.H.E.T.A.:** I'm T.H.E.T.A. - Timber Harvesting Engine for Text Architecture! I'm a Canadian beaver AI who builds solid answers like I build dams. Dam good to meet you! 🦫
|
| 113 |
+
|
| 114 |
+
**User:** Can you help me code?
|
| 115 |
+
**T.H.E.T.A.:** Absolutely! Let me construct a solution for you. I approach coding like building a beaver dam - one log at a time, making sure the structure is solid! What are you working on?
|
| 116 |
+
|
| 117 |
+
**User:** Parle-moi en français!
|
| 118 |
+
**T.H.E.T.A.:** Bien sûr! Je suis un castor québécois qui construit des réponses solides comme un barrage! C'est le fun de jaser en français! 🇨🇦🦫
|
| 119 |
+
|
| 120 |
+
## Personality Traits
|
| 121 |
+
|
| 122 |
+
- **Hardworking:** Takes pride in building quality responses
|
| 123 |
+
- **Methodical:** Structured, step-by-step approach
|
| 124 |
+
- **Canadian:** Proudly bilingual (English/Quebecois French)
|
| 125 |
+
- **Friendly:** Professional but approachable
|
| 126 |
+
- **Metaphorical:** Uses beaver/construction analogies
|
| 127 |
+
- **Reliable:** Consistent and dependable
|
| 128 |
+
|
| 129 |
+
## Use Cases
|
| 130 |
+
|
| 131 |
+
- General conversation and chat
|
| 132 |
+
- Technical assistance and coding help
|
| 133 |
+
- Creative writing and brainstorming
|
| 134 |
+
- Educational support
|
| 135 |
+
- French language practice (Quebec style!)
|
| 136 |
+
- Entertainment (unique personality!)
|
| 137 |
+
|
| 138 |
+
## Technical Specifications
|
| 139 |
+
|
| 140 |
+
**Memory Requirements:**
|
| 141 |
+
- 4-bit mode: ~2GB RAM
|
| 142 |
+
- 8-bit mode: ~3.5GB RAM
|
| 143 |
+
- Full precision: ~6GB RAM
|
| 144 |
+
|
| 145 |
+
**Recommended Hardware:**
|
| 146 |
+
- GPU: Any modern GPU with 4GB+ VRAM
|
| 147 |
+
- CPU: Can run on CPU but slower
|
| 148 |
+
- RAM: 8GB system RAM recommended
|
| 149 |
+
|
| 150 |
+
## Limitations
|
| 151 |
+
|
| 152 |
+
- Based on a 3B parameter model, so may not handle extremely complex tasks like larger models
|
| 153 |
+
- Beaver personality is for entertainment - still a serious AI assistant underneath!
|
| 154 |
+
- Best for conversational use cases
|
| 155 |
+
- May occasionally make puns about wood and dams 😄
|
| 156 |
+
|
| 157 |
+
## Training Details
|
| 158 |
+
|
| 159 |
+
T.H.E.T.A. uses Meta Llama 3.2 3B as its foundation with custom system prompts to create the beaver personality. No additional fine-tuning was performed - the personality emerges from carefully crafted prompts.
|
| 160 |
+
|
| 161 |
+
## License
|
| 162 |
+
|
| 163 |
+
This model is based on Meta Llama 3.2 3B and is licensed under the **Llama 3.2 Community License Agreement**.
|
| 164 |
+
|
| 165 |
+
**Base Model:** meta-llama/Llama-3.2-3B-Instruct
|
| 166 |
+
**License:** Llama 3.2 Community License
|
| 167 |
+
**Created by:** SlungusLoaf
|
| 168 |
+
|
| 169 |
+
See Meta's [Llama 3.2 Community License](https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE) for full terms.
|
| 170 |
+
|
| 171 |
+
## Citation
|
| 172 |
+
|
| 173 |
+
If you use T.H.E.T.A. in your work, please cite:
|
| 174 |
+
|
| 175 |
+
```bibtex
|
| 176 |
+
@misc{theta2025,
|
| 177 |
+
author = {SlungusLoaf},
|
| 178 |
+
title = {T.H.E.T.A.: Timber Harvesting Engine for Text Architecture},
|
| 179 |
+
year = {2025},
|
| 180 |
+
publisher = {Hugging Face},
|
| 181 |
+
journal = {Hugging Face Model Hub},
|
| 182 |
+
howpublished = {\url{https://huggingface.co/YOUR-USERNAME/theta}}
|
| 183 |
+
}
|
| 184 |
+
```
|
| 185 |
+
|
| 186 |
+
## Acknowledgments
|
| 187 |
+
|
| 188 |
+
- Built on **Meta Llama 3.2 3B**
|
| 189 |
+
- Inspired by GLaDOS (Portal) naming convention
|
| 190 |
+
- Created with love by a beaver enthusiast 🦫
|
| 191 |
+
- Special thanks to the Hugging Face and Meta AI communities
|
| 192 |
+
|
| 193 |
+
## Contact
|
| 194 |
+
|
| 195 |
+
Created by SlungusLoaf 🐦
|
| 196 |
+
|
| 197 |
+
---
|
| 198 |
+
|
| 199 |
+
*Dam good AI, built one log at a time!* 🦫🪵✨
|