45 lines
1.3 KiB
Markdown
45 lines
1.3 KiB
Markdown
---
|
|
language:
|
|
- my
|
|
license: apache-2.0
|
|
base_model: URajinda/Qwen2.5-MM-1.5B-Base
|
|
tags:
|
|
- burmese
|
|
- myanmar
|
|
- shweyon
|
|
- qwen
|
|
- tokenization
|
|
library_name: transformers
|
|
---
|
|
|
|
# 🚀 ShweYon-Qwen2.5-Burmese-1.5B-v1.2(Enhanced Burmese LLM)
|
|
|
|
**ShweYon-v1.2-Base** is a specialized language model based on the Qwen2.5-1.5B architecture, meticulously optimized for the Myanmar (Burmese) language. This version features a significant **Vocabulary Expansion** designed to solve common tokenization inefficiencies in Burmese NLP.
|
|
|
|
|
|
### 📊 Technical Specifications
|
|
|
|
| Feature | Specification |
|
|
| :--- | :--- |
|
|
| **Base Architecture** | Qwen2.5-1.5B |
|
|
| **New Vocab Size** | 152,858 |
|
|
| **Added Tokens** | 1,418 (Cumulative) |
|
|
| **Model Size Increase** | ~4.73 MB (+0.40% total weight) |
|
|
| **Training Type** | Continual Pre-training (CPT) |
|
|
| **Language** | Myanmar (Burmese) |
|
|
|
|
|
|
### 💻 Usage
|
|
|
|
```python
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
|
|
model_name = "URajinda/ShweYon-Qwen2.5-Burmese-1.5B-v1.2"
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
|
model = AutoModelForCausalLM.from_pretrained(model_name)
|
|
|
|
text = "မြန်မာနိုင်ငံသည်"
|
|
inputs = tokenizer(text, return_tensors="pt")
|
|
outputs = model.generate(**inputs, max_new_tokens=50)
|
|
print(tokenizer.decode(outputs[0], skip_special_tokens=True)) |