Text Generation
Transformers
Safetensors
Thai
English
qwen2
text-generation-inference
sft
trl
4-bit precision
bitsandbytes
LoRA
Fine-Tuning with LoRA
LLM
GenAI
NT GenAI
ntgenai
lahnmah
NT Thai GPT
ntthaigpt
medical
medtech
HealthGPT
หลานม่า
NT Academy
conversational
Inference Endpoints
4-bit precision
Update README.md
Browse files
README.md
CHANGED
@@ -30,22 +30,22 @@ tags:
|
|
30 |
new_version: Aekanun/openthaigpt-MedChatModelv5.1
|
31 |
---
|
32 |
|
33 |
-
# 🇹🇭 Model Card for
|
34 |
<!-- Provide a quick summary of what the model is/does. -->
|
35 |
-
This model is fine-tuned from
|
36 |
|
37 |
## 👤 **Developed and Fine-tuned by:**
|
38 |
- **Amornpan Phornchaicharoen**
|
39 |
- **Aekanun Thongtae**
|
40 |
|
41 |
## Model Description
|
42 |
-
This model was fine-tuned using Supervised Fine-Tuning (SFT) to optimize it for medical question answering in Thai. The base model is `openthaigpt1.5-7b-instruct`, and it has been enhanced with domain-specific knowledge using the
|
43 |
|
44 |
- **Model type:** Causal Language Model (AutoModelForCausalLM)
|
45 |
- **Language(s):** Thai
|
46 |
- **License:** Apache License 2.0
|
47 |
-
- **Fine-tuned from model:**
|
48 |
-
- **Dataset used for fine-tuning:**
|
49 |
|
50 |
### Model Sources
|
51 |
|
@@ -97,120 +97,92 @@ This model can be used as a foundational model for medical assistance systems, c
|
|
97 |
|
98 |
## How to Get Started with the Model
|
99 |
|
100 |
-
Here’s how
|
101 |
|
102 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
103 |
import torch
|
104 |
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
|
105 |
-
```
|
106 |
|
107 |
-
#
|
108 |
-
|
109 |
-
|
110 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
111 |
|
112 |
-
|
113 |
-
|
|
|
|
|
|
|
114 |
bnb_config = BitsAndBytesConfig(
|
115 |
load_in_4bit=True,
|
116 |
bnb_4bit_quant_type="nf4",
|
117 |
-
bnb_4bit_compute_dtype=torch.float16
|
118 |
)
|
119 |
-
```
|
120 |
|
121 |
-
|
122 |
-
|
123 |
-
|
|
|
|
|
124 |
model = AutoModelForCausalLM.from_pretrained(
|
125 |
-
|
126 |
quantization_config=bnb_config,
|
127 |
-
trust_remote_code=True,
|
128 |
-
device_map="auto"
|
129 |
-
)
|
130 |
-
|
131 |
-
tokenizer = AutoTokenizer.from_pretrained(
|
132 |
-
model_name,
|
133 |
trust_remote_code=True
|
134 |
)
|
135 |
-
tokenizer.pad_token = tokenizer.eos_token
|
136 |
-
```
|
137 |
-
|
138 |
-
# Function to generate responses
|
139 |
-
```python
|
140 |
-
def generate_response(prompt, max_new_tokens=256):
|
141 |
-
PROMPT = f'[INST] <You are a question answering assistant. Answer the question as truthful and helpful as possible คุณคือผู้ช่วยตอบคำถาม จงตอบคำถามอย่างถูกต้องและมีประโยชน์ที่สุด<>{prompt}[/INST]'
|
142 |
-
|
143 |
-
inputs = tokenizer(PROMPT, return_tensors="pt").to(model.device)
|
144 |
-
|
145 |
-
generation_config = {
|
146 |
-
"temperature": 0.6,
|
147 |
-
"top_p": 0.95,
|
148 |
-
"repetition_penalty": 1.15,
|
149 |
-
"max_new_tokens": max_new_tokens,
|
150 |
-
"pad_token_id": tokenizer.eos_token_id
|
151 |
-
}
|
152 |
-
|
153 |
-
with torch.no_grad():
|
154 |
-
generation_output = model.generate(
|
155 |
-
**inputs,
|
156 |
-
**generation_config
|
157 |
-
)
|
158 |
-
|
159 |
-
response = tokenizer.decode(generation_output[0], skip_special_tokens=True)
|
160 |
-
response = response.split("[/INST]")[-1].strip()
|
161 |
-
return response
|
162 |
-
```
|
163 |
-
|
164 |
-
# Example usage
|
165 |
-
```python
|
166 |
-
if __name__ == "__main__":
|
167 |
-
questions = [
|
168 |
-
"โปรดอธิบายลักษณะช่องปากที่เป็นมะเร็งในระยะเริ่มต้น",
|
169 |
-
"อาการของโรคไข้เลือดออกมีอะไรบ้าง"
|
170 |
-
]
|
171 |
-
|
172 |
-
print("Testing question answering:")
|
173 |
-
for question in questions:
|
174 |
-
print("\nQuestion: ", question)
|
175 |
-
answer = generate_response(question)
|
176 |
-
print("Answer: ", answer)
|
177 |
-
print("-" * 50)
|
178 |
-
```
|
179 |
-
|
180 |
-
## 9. Output
|
181 |
|
182 |
-
|
183 |
-
คำถาม: โปรดอธิบายลักษณะช่องปากที่เป็นมะเร็งในระยะเริ่มต้น
|
184 |
-
/usr/local/lib/python3.10/dist-packages/transformers/generation/configuration_utils.py:590: UserWarning: `do_sample` is set to `False`. However, `temperature` is set to `0.6` -- this flag is only used in sample-based generation modes. You should set `do_sample=True` or unset `temperature`.
|
185 |
-
warnings.warn(
|
186 |
-
/usr/local/lib/python3.10/dist-packages/transformers/generation/configuration_utils.py:595: UserWarning: `do_sample` is set to `False`. However, `top_p` is set to `0.95` -- this flag is only used in sample-based generation modes. You should set `do_sample=True` or unset `top_p`.
|
187 |
-
warnings.warn(
|
188 |
-
คำตอบ: [ANS] ในระยะเริ่มต้นของมะเร็งช่องปาก อาจไม่มีอาการแสดงที่ชัดเจนหรือมีเพียงเล็กน้อย เช่น การเปลี่ยนแปลงทางกายภาพเล็กๆ บนเนื้อเยื่อภายในช่องปาก ซึ่งอาจรวมถึง:
|
189 |
|
190 |
-
|
191 |
|
192 |
-
|
|
|
|
|
|
|
193 |
|
194 |
-
|
195 |
|
196 |
-
|
197 |
-
--------------------------------------------------
|
198 |
|
199 |
-
|
200 |
-
|
201 |
|
202 |
-
|
203 |
|
204 |
-
|
205 |
|
206 |
-
|
|
|
|
|
207 |
|
208 |
-
|
209 |
-
|
210 |
-
5. เยื่อบุใต้ผิวหนังอักเสบ (Hemorrhagic rash) : ซึ่งเริ่มจากจุดแดงๆ บนใบหน้��� ลำคอ และแขน แล้วขยายไปทั่วร่างกาย โดยเฉพาะบริเวณขาและแขน
|
211 |
|
212 |
-
|
213 |
-
|
214 |
|
215 |
### Authors
|
216 |
* Amornpan Phornchaicharoen (amornpan@gmail.com)
|
|
|
30 |
new_version: Aekanun/openthaigpt-MedChatModelv5.1
|
31 |
---
|
32 |
|
33 |
+
# 🇹🇭 Model Card for openthaigpt1.5-7b-medical-tuned
|
34 |
<!-- Provide a quick summary of what the model is/does. -->
|
35 |
+
This model is fine-tuned from openthaigpt1.5-7b-instruct using Supervised Fine-Tuning (SFT) on the Thaweewat/thai-med-pack dataset. The model is designed for medical question-answering tasks in Thai, specializing in providing accurate and contextual answers based on medical information.
|
36 |
|
37 |
## 👤 **Developed and Fine-tuned by:**
|
38 |
- **Amornpan Phornchaicharoen**
|
39 |
- **Aekanun Thongtae**
|
40 |
|
41 |
## Model Description
|
42 |
+
This model was fine-tuned using Supervised Fine-Tuning (SFT) to optimize it for medical question answering in Thai. The base model is `openthaigpt1.5-7b-instruct`, and it has been enhanced with domain-specific knowledge using the Thaweewat/thai-med-pack dataset.
|
43 |
|
44 |
- **Model type:** Causal Language Model (AutoModelForCausalLM)
|
45 |
- **Language(s):** Thai
|
46 |
- **License:** Apache License 2.0
|
47 |
+
- **Fine-tuned from model:** openthaigpt1.5-7b-instruct
|
48 |
+
- **Dataset used for fine-tuning:** Thaweewat/thai-med-pack
|
49 |
|
50 |
### Model Sources
|
51 |
|
|
|
97 |
|
98 |
## How to Get Started with the Model
|
99 |
|
100 |
+
Here’s how to load and use the model for generating medical responses in Thai:
|
101 |
|
102 |
+
## 1. Install the Required Packages
|
103 |
+
|
104 |
+
First, ensure you have installed the required libraries by running:
|
105 |
+
|
106 |
+
python
|
107 |
+
pip install torch transformers bitsandbytes
|
108 |
+
|
109 |
+
## 2. Load the Model and Tokenizer
|
110 |
+
|
111 |
+
You can load the model and tokenizer directly from Hugging Face using the following code:
|
112 |
+
|
113 |
+
python
|
114 |
import torch
|
115 |
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
|
|
|
116 |
|
117 |
+
# Define the model path
|
118 |
+
model_path = 'amornpan/openthaigpt-MedChatModelv11'
|
119 |
+
|
120 |
+
# Load the tokenizer and model
|
121 |
+
tokenizer = AutoTokenizer.from_pretrained(model_path)
|
122 |
+
tokenizer.pad_token = tokenizer.eos_token
|
123 |
+
|
124 |
+
## 3. Prepare Your Input (Custom Prompt)
|
125 |
+
|
126 |
+
Create a custom medical prompt that you want the model to respond to:
|
127 |
+
|
128 |
+
python
|
129 |
+
custom_prompt = "โปรดอธิบายลักษณะช่องปากที่เป็นมะเร็งในระยะเริ่มต้น"
|
130 |
+
PROMPT = f'[INST] <You are a question answering assistant. Answer the question as truthfully and helpfully as possible. คุณคือผู้ช่วยตอบคำถาม จงตอบคำถามอย่างถูกต้องและมีประโยชน์ที่สุด<>{custom_prompt}[/INST]'
|
131 |
+
|
132 |
+
# Tokenize the input prompt
|
133 |
+
inputs = tokenizer(PROMPT, return_tensors="pt", padding=True, truncation=True)
|
134 |
+
|
135 |
|
136 |
+
## 4. Configure the Model for Efficient Loading (4-bit Quantization)
|
137 |
+
|
138 |
+
The model uses 4-bit precision for efficient inference. Here’s how to set up the configuration:
|
139 |
+
|
140 |
+
python
|
141 |
bnb_config = BitsAndBytesConfig(
|
142 |
load_in_4bit=True,
|
143 |
bnb_4bit_quant_type="nf4",
|
144 |
+
bnb_4bit_compute_dtype=torch.float16
|
145 |
)
|
|
|
146 |
|
147 |
+
## 5. Load the Model with Quantization Support
|
148 |
+
|
149 |
+
Now, load the model with the 4-bit quantization settings:
|
150 |
+
|
151 |
+
python
|
152 |
model = AutoModelForCausalLM.from_pretrained(
|
153 |
+
model_path,
|
154 |
quantization_config=bnb_config,
|
|
|
|
|
|
|
|
|
|
|
|
|
155 |
trust_remote_code=True
|
156 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
157 |
|
158 |
+
## 6. Move the Model and Inputs to the GPU (prefer GPU)
|
|
|
|
|
|
|
|
|
|
|
|
|
159 |
|
160 |
+
For faster inference, move the model and input tensors to a GPU, if available:
|
161 |
|
162 |
+
python
|
163 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
164 |
+
model.to(device)
|
165 |
+
inputs = {k: v.to(device) for k, v in inputs.items()}
|
166 |
|
167 |
+
## 7. Generate a Response from the Model
|
168 |
|
169 |
+
Now, generate the medical response by running the model:
|
|
|
170 |
|
171 |
+
python
|
172 |
+
outputs = model.generate(**inputs, max_new_tokens=200, do_sample=True)
|
173 |
|
174 |
+
## 8. Decode the Generated Text
|
175 |
|
176 |
+
Finally, decode and print the response from the model:
|
177 |
|
178 |
+
python
|
179 |
+
generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
180 |
+
print(generated_text)
|
181 |
|
182 |
+
## 9. Output
|
|
|
|
|
183 |
|
184 |
+
python
|
185 |
+
[INST] <You are a question answering assistant. Answer the question as truthfully and helpfully as possible. คุณคือผู้ช่วยตอบคำถาม จงตอบคำถามอย่างถูกต้องและมีประโยชน์ที่สุด<>โปรดอธิบายลักษณะช่องปากที่เป็นมะเร็งในระยะเริ่มต้น[/INST] มะเร็งช่องปากเป็นมะเร็งเพียงชนิดเดียวที่ได้รับผลกระทบจากนิโคติน มันคือผู้ชายกลุ่มอายุ 60 – 75 คน คุณจะแสดงอาการและเกิดขึ้นอย่างรวดเร็วหากเกิดมะเร็งช่องปาก คุณจะสังเกตเห็นปื้นแพร่กระจายของเนื้องอก ส่วนใหญ่ในช่องปาก เนื้องอกแสดงว่าเป็นเจ้าแห่ที่กำลังทำลายเยียวยา ค้นหาทั้งภายในและภายนอกลิ้นที่อยู่ติดกางเกงป่อง มะเร็งกระเพาะปัสสาวะหรือมะเร็งกล้ามเนื้อกระเพาะ
|
186 |
|
187 |
### Authors
|
188 |
* Amornpan Phornchaicharoen (amornpan@gmail.com)
|