File size: 1,240 Bytes
43d0420
 
 
 
 
 
 
 
292e627
43d0420
292e627
 
 
 
 
 
 
 
 
 
 
4491063
292e627
 
 
 
 
 
 
 
 
 
43d0420
292e627
 
 
 
43d0420
 
 
 
292e627
 
 
43d0420
 
292e627
43d0420
292e627
 
 
 
43d0420
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
---
datasets:
- sean0042/KorMedMCQA
language:
- ko
- en
pipeline_tag: text-generation
---
### Model Card for Model ID
base_model : [google/gemma-2b-it](https://huggingface.co/google/gemma-2b-it)

### Basic usage
```python
# pip install accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

tokenizer = AutoTokenizer.from_pretrained("MDDDDR/gemma-2b-it-v0.1")
model = AutoModelForCausalLM.from_pretrained(
    "MDDDDR/gemma-2b-it-v0.1",
    device_map="auto",
    torch_dtype=torch.bfloat32
)

input_text = "사과가 뭐야?"
input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")

outputs = model.generate(**input_ids)
print(tokenizer.decode(outputs[0]))
```

### Training dataset
dataset : [sean0042/KorMedMCQA](https://huggingface.co/datasets/sean0042/KorMedMCQA)

### lora_config and bnb_config in Training
```python
bnd_config = BitsAndBytesConfig(
  load_in_4bit = True,
  bnb_4bit_use_double_quant = True,
  bnb_4bit_quant_type = 'nf4',
  bnb_4bit_compute_dtype = torch.bfloat16
)

lora_config = LoraConfig(
  r = 32,
  lora_alpha = 32,
  lora_dropout = 0.05,
  target_modules = ['q_proj', 'k_proj', 'v_proj', 'o_proj', 'gate_proj', 'up_proj', 'down_proj']
)
```

### Hardware
A100 40GB x 1