File size: 6,371 Bytes
ec25a6e
1240f58
 
 
e3f7268
 
 
 
 
 
 
 
 
 
 
 
 
 
 
54ec1f9
e3f7268
 
 
 
 
 
 
 
 
 
 
 
54ec1f9
e3f7268
 
 
 
 
 
 
 
 
 
 
 
 
54ec1f9
e3f7268
 
 
 
 
083e15b
e3f7268
 
 
 
 
 
6ffd8e1
3ae9e9f
54ec1f9
e3f7268
 
 
 
 
 
 
 
 
 
 
 
54ec1f9
e3f7268
 
 
 
 
 
 
 
 
 
 
 
 
54ec1f9
e3f7268
ec25a6e
 
70b712a
1240f58
ec25a6e
1240f58
af0705f
 
ec25a6e
 
1240f58
 
 
ec25a6e
1240f58
ec25a6e
1240f58
ec25a6e
1240f58
 
ec25a6e
1240f58
0d8b104
1240f58
 
 
af0705f
ec25a6e
af0705f
1240f58
af0705f
1240f58
 
 
 
 
 
 
 
ec25a6e
1240f58
 
af0705f
 
 
 
 
1240f58
 
 
 
ec25a6e
1240f58
04bf1ce
 
 
27622fa
04bf1ce
af0705f
04bf1ce
 
af0705f
04bf1ce
af0705f
 
 
04bf1ce
af0705f
04bf1ce
 
27622fa
04bf1ce
af0705f
04bf1ce
 
af0705f
2999cfc
 
af0705f
ac93226
ec25a6e
af0705f
04bf1ce
b2dee45
 
 
 
 
54ec1f9
 
 
 
 
 
 
b2dee45
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
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
---
license: apache-2.0
language:
- tr
model-index:
- name: Orbita-v0.1
  results:
  - task:
      type: text-generation
      name: Text Generation
    dataset:
      name: AI2 Reasoning Challenge TR
      type: ai2_arc
      config: ARC-Challenge
      split: test
      args:
        num_few_shot: 25
    metrics:
    - type: acc
      value: 41.97
      name: accuracy
  - task:
      type: text-generation
      name: Text Generation
    dataset:
      name: HellaSwag TR
      type: hellaswag
      split: validation
      args:
        num_few_shot: 10
    metrics:
    - type: acc
      value: 48
      name: accuracy
  - task:
      type: text-generation
      name: Text Generation
    dataset:
      name: MMLU TR
      type: cais/mmlu
      config: all
      split: test
      args:
        num_few_shot: 5
    metrics:
    - type: acc
      value: 49.51
      name: accuracy
  - task:
      type: text-generation
      name: Text Generation
    dataset:
      name: TruthfulQA TR
      type: truthful_qa
      config: multiple_choice
      split: validation
      args:
        num_few_shot: 0
    metrics:
    - type: acc
      name: accuracy
      value: 50.78
  - task:
      type: text-generation
      name: Text Generation
    dataset:
      name: Winogrande TR
      type: winogrande
      config: winogrande_xl
      split: validation
      args:
        num_few_shot: 5
    metrics:
    - type: acc
      value: 56.16
      name: accuracy
  - task:
      type: text-generation
      name: Text Generation
    dataset:
      name: GSM8k TR
      type: gsm8k
      config: main
      split: test
      args:
        num_few_shot: 5
    metrics:
    - type: acc
      value: 50.41
      name: accuracy
---

<img src="https://huggingface.co/Orbina/Orbita-v0.1/resolve/main/orbita.png"
alt="Orbita LLM" width="500"/>

# Orbita-v0.1
This model is a Turkish Large Language Model with multiple abilites across different dimensions on the Turkish Language covering a variety of Tasks like Coding, Math, etc... 
This model is an extended version of a Qwen-based Large Language Model (LLM) for Turkish. It was trained on a cleaned Turkish dataset carefully annotated to carry out turkish instructions in an accurate and organized manner. This model was fully finetuned extensively on 8x H100 GPU's for 2 days using a carefully annotated Turkish dataset. 
## Model Details

- **Base Model**: Qwen 14B based LLM
- **Training Dataset**: Annotated Turkish Dataset
- **Training Method**: Full Finetuning

## Usage Examples

```python

from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # the device to load the model onto

model = AutoModelForCausalLM.from_pretrained(
    "Orbina/Orbita-v0.1",
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Orbina/Orbita-v0.1")

prompt = "Python'da ekrana 'Merhaba Dünya' nasıl yazılır?"
messages = [
    {"role": "system", "content": "Sen, Orbina AI tarafından üretilen ve verilen talimatları takip ederek en iyi cevabı üretmeye çalışan yardımcı bir yapay zekasın."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)

generated_ids = model.generate(
    model_inputs.input_ids,
    temperature=0.3,
    top_k=50,
    top_p=0.9,
    max_new_tokens=512,
    repetition_penalty=1,
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
```
## Example Generations

```
Prompt:
Türkiye'deki finansal kurumlar, yapay zeka ile birlikte gelen veri güvenliği risklerini nasıl yönetiyor, 3 örnek verir misin?

Cevap:
Türkiye'deki finansal kurumlar, yapay zeka (AI) teknolojilerinin getirdiği veri güvenliği riskleri çeşitli yöntemlerle ele almakta ve bu riskleri azaltmaya çalışmaktadır. İşte üç örnek:

1. **Veri Şifreleme:** Finansal kuruluşlar, müşteri bilgileri gibi hassas verileri şifrelemeyi kullanarak veri ihlallerine karşı koruma sağlar. AES (Advanced Encryption Standard), RSA (Rivest-Shamir-Adleman) veya ECC (Elliptik Eğri Kriptografisi) gibi güçlü şifreleme algoritmaları, verilerin okunmasını engelleyebilir.
2. **Anomali Tespit Sistemleri:** Yapay zeka destekli anomali tespit sistemleri, beklenmedik davranışları algılayabilir ve güvenlik ekiplerini uyarır. Bu sistemler, makine öğrenimi modelleri sayesinde zaman içinde daha etkili hale gelebilir ve dolandırıcılık faaliyetlerini önlemede önemli rol oynar.
3. **Sürekli İzleme ve Güvenlik Politikası Uygulama:** Türkiye'deki finansal kurumlar, sürekli izleme ve güvenlik politikalarını uygularak veri güvenliğini sürdürür. Bu süreçlerde düzenli güvenlik denetimleri yapılır ve tüm personelin güvenlik konusunda eğitildiği göz önünde bulundurulur. Ayrıca, müşterilere de güvenlik protokollerini anlatan eğitimler sunulur.

Bu örnekler, Türkiye'deki finansal kurumların yapay zeka ile ilgili veri güvenliği risklerini yönetmek için kullandıkları bazı stratejileri göstermektedir. Ancak herhangi bir güvenlik sistemi mükemmel değildir ve bu nedenle kurumlar da sürekli olarak güvenlik durumlarını değerlendirip iyileştirmeler yapmaktadırlar.
```

```
Prompt:
Python'da ekrana "Merhaba Dünya" nasıl yazılır?

Cevap:
Python'da ekrana "Merhaba Dünya" yazmak için `print` fonksiyonunu kullanabilirsiniz. İşte basit bir örnek:

``python
print("Merhaba Dünya")
``

Bu kodu çalıştırdığınızda, konsol ekranında "Merhaba Dünya" metnini göreceksiniz. `print` fonksiyonu, verilen argümanı (bu durumda "Merhaba Dünya") alır ve onu çıktı olarak verir.
```

# [Open LLM Turkish Leaderboard Evaluation Results](https://huggingface.co/spaces/malhajar/OpenLLMTurkishLeaderboard)

|             Metric              |Value|
|---------------------------------|----:|
|Avg.                             |49.47|
|AI2 Reasoning Challenge_tr       |41.97|
|HellaSwag_tr                     |48.00|
|MMLU_tr                          |49.51|
|TruthfulQA_tr                    |50.78|
|Winogrande _tr                   |56.16|
|GSM8k_tr                         |50.41|