MiguelLabs commited on
Commit
ff39fef
·
verified ·
1 Parent(s): ede0424

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +121 -0
README.md CHANGED
@@ -1,3 +1,124 @@
1
  ---
 
 
2
  license: mit
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language:
3
+ - pt
4
  license: mit
5
+ pipeline_tag: text-generation
6
+ tags:
7
+ - transformer
8
+ - pytorch
9
+ - portuguese
10
+ - clash-royale
11
+ - chatbot
12
+ - question-answering
13
+ - educational
14
  ---
15
+
16
+ # ClashAI-300k
17
+
18
+ **ClashAI-300k** is a lightweight Transformer language model with approximately **300,000 parameters**, designed to answer questions about **Clash Royale** in Portuguese.
19
+
20
+ This project was created as an educational experiment to demonstrate how a small language model can be trained from scratch using PyTorch, a custom tokenizer, and a handcrafted dataset.
21
+
22
+ ## Features
23
+
24
+ - 🧠 Approximately **300K parameters**
25
+ - 🇧🇷 Portuguese language
26
+ - 🎮 Specialized in Clash Royale
27
+ - ⚡ Lightweight and fast inference
28
+ - 🔤 Custom BPE tokenizer
29
+ - 🔥 Built entirely with PyTorch
30
+ - 📚 Trained from scratch (no pretrained weights)
31
+
32
+ ## Intended Use
33
+
34
+ ClashAI-300k is intended for:
35
+
36
+ - Answering Clash Royale questions
37
+ - Educational purposes
38
+ - Learning how Transformer models work
39
+ - Small AI experiments
40
+ - Running on low-end hardware
41
+
42
+ It is **not** intended to replace large language models or provide general-purpose knowledge.
43
+
44
+ ## Model Details
45
+
46
+ | Property | Value |
47
+ |----------|-------|
48
+ | Model | Transformer Decoder |
49
+ | Parameters | ~300,000 |
50
+ | Framework | PyTorch |
51
+ | Tokenizer | Custom Byte Pair Encoding (BPE) |
52
+ | Language | Portuguese |
53
+ | Domain | Clash Royale |
54
+
55
+ ## Training
56
+
57
+ The model was trained from scratch using a manually curated dataset containing hundreds of Clash Royale question-answer pairs.
58
+
59
+ Training includes:
60
+
61
+ - Custom tokenizer training
62
+ - Sequence-to-sequence language modeling
63
+ - Cross-entropy loss optimization
64
+ - Autoregressive text generation
65
+
66
+ ## Example
67
+
68
+ **Input**
69
+
70
+ ```
71
+ Como conseguir ouro?
72
+ ```
73
+
74
+ **Output**
75
+
76
+ ```
77
+ Você pode conseguir ouro vencendo batalhas, abrindo baús, participando de eventos e doando cartas ao clã.
78
+ ```
79
+
80
+ ---
81
+
82
+ **Input**
83
+
84
+ ```
85
+ Como derrotar um Corredor?
86
+ ```
87
+
88
+ **Output**
89
+
90
+ ```
91
+ Use construções defensivas para desviar o Corredor e tropas de alto dano para eliminá-lo rapidamente.
92
+ ```
93
+
94
+ ## Limitations
95
+
96
+ ClashAI-300k is intentionally small.
97
+
98
+ Some limitations include:
99
+
100
+ - Limited world knowledge
101
+ - Specialized only in Clash Royale
102
+ - May generate incorrect or repetitive responses
103
+ - Performs best on questions similar to its training data
104
+
105
+ ## Future Improvements
106
+
107
+ Planned improvements include:
108
+
109
+ - Larger training dataset
110
+ - Better sampling strategies
111
+ - Larger context window
112
+ - Improved architecture
113
+ - More gameplay knowledge
114
+ - Better generalization
115
+
116
+ ## Disclaimer
117
+
118
+ This is an independent educational project.
119
+
120
+ Clash Royale is a trademark of Supercell. This project is not affiliated with, endorsed by, or sponsored by Supercell.
121
+
122
+ ## License
123
+
124
+ MIT License