README.md

---
language:
- th
- en
license: apache-2.0
library_name: transformers
base_model:
- Qwen/Qwen2.5-14B-Instruct
- Qwen/Qwen2.5-14B
pipeline_tag: text-generation
---
<img src="./Tsunami.webp" alt="Tsunami Model" width="800" style="margin-left:'auto' margin-right:'auto' display:'block'"/>

# Tsunami-1.0-14B-Instruct
**TSUNAMI**: Transformative Semantic Understanding and Natural Augmentation Model for Intelligence.

**TSUNAMI** full name was created by ChatGPT.

---

### infomation
**Tsunami-1.0-14B-Instruct** is Thai Large Language Model that fine-tuned from **Qwen2.5-14B** in Thai dataset.

---

### Author
 - Pollakrit Lorprasertkul | game.pollakrit@gmail.com

---

### Performance Evaluation

Below are the benchmark results of **Tsunami-1.0-14B-Instruct** compared to similar models in its class:

| Model | Average | Thai Exam | M3Exam |
| --- | --- | --- | --- |
| Qwen2.5-14B-Instruct | 58.45 | 57.35 | 59.55 |
| Meta-Llama-3.1-70B-Instruct | 59.38 | 58.23 | 60.52 |
| llama-3-typhoon-v1.5x-70b-instruct | 59.34 | 58.76 | 59.92 |
| openthaigpt1.5-14b-instruct | 60.41 | 58.41 | 62.41 |
| **Tsunami-1.0-14B-Instruct** | **62.05** | **61.06** | **63.05** |

---

### Prompt Template

This model uses `ChatML` prompt template:

```
<|im_start|>system
{System}<|im_end|>
<|im_start|>user
{User}<|im_end|>
<|im_start|>assistant
{Assistant}
````

---

### How to use


```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
model_name = "Tsunami-th/Tsunami-1.0-14B-Instruct"
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "สวัสดีครับ"}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
inputs = tokenizer(text, return_tensors="pt")
inputs = inputs.to(model.device)
with torch.no_grad():
   output = model.generate(**inputs, max_new_tokens=512)
response = tokenizer.decode(output[0, len(inputs['input_ids'][0]):], skip_special_tokens=True)
```

---