File size: 11,624 Bytes
6f52c3b 6f05a20 6f52c3b a63d33a 6f52c3b c2abf9a d08ff6c c2abf9a d08ff6c c2abf9a d08ff6c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 |
---
language:
- ko
license: apache-2.0
pipeline_tag: text-generation
---
# Kor-Gemma-2B
> Update @ 2024.05.10: First release of gemma-ko
<!-- Provide a quick summary of what the model is/does. -->
This model card corresponds to the 2B-it version of the **Gemma-Ko** model.
**Resources and Technical Documentation**:
* [Original Gemma-2b-it](https://huggingface.co/google/gemma-2b-it)
**Citation**
```bibtex
@misc {gemma-summary-v01 ,
author = { {frcp,nebchi,pepperonipizza} },
title = { gemma-summary-v01 },
year = 2024,
url = { https://huggingface.co/cpm-ai/gemma-ko-v01 },
publisher = { Hugging Face }
}
```
**Model Developers**: frcp, nebchi, pepperonipizza
## Model Information
I trained a language model using a dataset of 363,000 Korean text samples.
### Description
It has been trained with a large amount of Korean tokens compared to other LLMs, enabling it to generate high-quality Korean text.
Additionally, it shows improved performance with less data compared to other LLM models.
#### Running the model on a single / multi GPU
```python
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("cpm-ai/gemma-ko-v01")
model = AutoModelForCausalLM.from_pretrained("cpm-ai/gemma-ko-v01", device_map="auto")
prompt = """μμ½ ν λ¬Έμ₯ :
[μλ
νμΈμ μλ°©μ‘ ν λ‘ μΉ΄νμ
λλ€.
μ€λ μ±νμ μ μΌ μλ°©μ‘μΌλ‘ μ§νν΄ λλ¦¬κ³ μλλ°μ.
νΉμ§μΌλ‘ μ ν¬κ° λΆμκΈ°λ λ§μ΄ λ°κΏλ΄€κ³ λ μ€λμ μ¬λμ κ³μ μ΄λλ§νΌ λλμ λν΄μ μ΄μΌκΈ° ν΄λ³ΌκΉ ν©λλ€.
νμ μν μμ λλμ λ μ€μ²νκ³ κ³μλ λ€ λΆ λͺ¨μκ³ μ΄μΌκΈ° λλ 보λλ‘ νκ² μ΅λλ€ κ·ΈλΌ λ€ λΆ μκ°ν΄ λλ¦¬κ² μ΅λλ€ λ€μ΄μ€μμ£ .
λ€ μ€λ μλ°©μ‘ ν λ‘ μΉ΄νμμλ λλμ μλ―Έμ λν΄μ μ΄μΌκΈ° λλ λ³ΌκΉ νλλ°μ.
μ΄ μ μΌμ μ΄μ λ΄μΌμ΄λ©΄ ν¬λ¦¬μ€λ§μ€κ³ μ§κΈ μν μ μμΌ λΆ μ§λκ³ μκ±°λ μ μ°νν΄λ‘μ€ ν μλ²μ§κ° μλΉν λ°λΉ μ§ κ·Έλ° μκ°μ
λλ€.
μ΄λ΄ λ κ°μ‘±κ³Ό λλ μΉμ§λ€κ³Ό ν¨κ» 보λ΄μ
μΌ λ μ΄ κ·ν μκ° λ΄ μ£Όμ
μ μ€μ λ€ λΆ λ¨Όμ μκ°ν΄ λ리λλ‘ νκ² μ΅λλ€.
λ¨Όμ &party-name1&μ μμμ
λλ€.
μλ
νμΈμ.
κ·Έλ¦¬κ³ μμ€λνκ΅ μ¬νμ¬μ
νκ³Όμ κ΅μμ
λλ€.
μλ
νμΈμ.
κ·Έλ¦¬κ³ μλ¦λ€μ΄ μ¬λ¨μ μμμ΄μ¬ μ
λλ€.
μλ
νμΈμ.
κ·Έλ¦¬κ³ μ¬λμ μ₯κΈ°κΈ°μ¦μ΄λλ³ΈλΆμ κ΅μ₯λμ΄μλλ€.
μλ
νμΈμ.
μ΄λ κ² λμ μ£Όμ
μ λ€μ νλ² κ°μ¬λ리ꡬμ.
κ·Έλ¦¬κ³ μ€λ νΉλ³ν μ ν¬ ν λ‘ μΉ΄νμλ μ©μ°κ΅¬ μμλ΄μ¬ μΌν°μμ λ΄μ¬νλμ λ νμλ λΆλ€μ΄ λμμ£Όμ
¨μ΅λλ€.
μ€λ λμμ£Όμ λΆλ€ λ€μ νλ² νμνκ³ μ§μ¬μΌλ‘ κ°μ¬λ립λλ€.
λ μ΄λ° μκΈ°λ₯Ό νμ£ μ°λ¦¬μ¬νμλ μμ§λ 곡λ체 μμμ΄ λΆμ‘±νλ€ λλμ μμμ΄ λΆμ‘±νλ€ κΈ°λΆ λ¬Ένκ° μ μ°©λΌ μμ§ μλ€ κ·Έλ° μκΈ°λ€μ λ§μ΄ νλλ°μ.
μ΄λ»κ² νλ©΄ κ·Έλ¬ν λ°λ»ν μ°λ¦¬λ€μ λ§μμ λ ν€μ°κ³ λ λλ μ μκ³ λ κ·Έλ° κ²μ μ΄λ ν μ λμ μ₯μΉλ‘ μ 보μν΄ λκ° μ μμκΉ
κ·Έλ° λ¬Έμ λ€μ νλνλ μ΄μΌκΈ° λλ 보λλ‘ νκ² μ΅λλ€ λλμ΄ λλ체 μ νμνμ§ κ·Έλ¦¬κ³ μλ‘ μ μΈ μκΈ° κ² μ£ .
κ·Έκ²λΆν° νλ² μκΈ°λ₯Ό νλ² ν΄ λ³ΌκΉ ν©λλ€ λ¨Όμ λ³νΈμ¬λκ»μ μκΈ°ν΄ μ£Όμκ² μ΅λκΉ.
μκΈ° ν볡νκΈ° μν΄μμ£ .
{laughing} μμ€μ μΌλ‘ λ€λ¦½λλ€.
μ¬μ€ κΈ°λΆλΌλ μ§ λλμ΄λΌλ κ² μκΈ° μ£Όλ¨Έλμμ λμ΄ λκ°λκΉ μκΈ°νν
μν΄κ° λ κ² κ°μλ° μ€μ λ‘ λλ λ³Έ μ¬λλ§ μλλ€.
{laughing} μ΄κ² μΌλ§λ μκΈ°κ° μ€μ€λ‘ ν볡ν΄μ§λμ§ κ·Έλμ μμ λ λλκΈ°λΆμ€λ
μ΄λΌλ λ§λ μꡬμ.
λ μ ν¬λ€μ΄ μ΄λ κ² μμμ λ μ΄λ κ² λͺ¨κΈμ κ΄ν μ±
μ μ½μ΄λ³΄λ©΄
κΈ°λΆ ν΄ λ³Έ μ¬λνν
κ°μ λ λ¬λΌκ³ ν΄λΌ μ΄κ² λͺ¨κΈνλ μ¬λμ΄ μ²« λ²μ§Έ μμΉμΌλ‘ μκΈ°ν΄μ.
κ·Έ μκΈ°λ λ¬΄μ¨ μκΈ°λλ©΄ ν΄λ³Έ μ¬λμ΄ μ¦κ±°μ°λκΉ λν κ°λ₯μ±μ΄ λ§λ€λ κ±°μ§μ μλ§ μ΄κ±΄ ν΄λ³΄μ
μΌ μ΄κ±° μ κ° μ무리 λ§μλλ €λ μμ©μꡬμ.
μ€μ λλ 보μ
μΌ κ·Έ κΈ°μ¨ μ¦κ±°μμ μμκ² λ©λλ€.
κ²°κ΅μλ μκΈ°νν
λμμ¨λ€ λΌκ³ νλ κ²μ΄ μμ μ¬λλ€μκ² λ§μ΄ ν½λ°°νλλ° μ₯κΈ°κΈ°μ¦ κ°μ κ²½μ°μλ λ΄κ° κΈ°μ¦μ νλ©΄
μ κ·Έκ²μ΄ κ²°κ΅μ λνν
λμμ¨λ€λ κ·Έ μ΄μ κ° λλ νλ©΄ λ΄κ° μΈμ λ μ§ νμκ° λμμ λ
μ¬ν μ λ°μ μΌλ‘ κ·Έλ κ² κΈ°μ¦νλ κ·Έ λ¬Ένκ° νμ°λλ©΄ λ΄κ° νμκ° λμ λ κ·Έκ²μ΄ κ²°κ΅ λνν
ννμ΄ λμμ¨λ€ λΌκ³ ν΄μ
μ€νμΈ κ°μ κ²½μ°μλ λ°±λ§ λͺ
λΉ μΌμμ¬ λͺ
μΌλ‘ μ μΈκ³μ μΌλ‘ κ°μ₯ λ§μ΄ κΈ°μ¦μ νκ³ μλλ°
κ·Έλ° μμμ΄ κ²°κ΅μ λ΄κ² λμμ€λ κ²μ΄λ€λΌκ³ νλ μμμ΄ ν½λ°°νκΈ° λλ¬Έμ κ·Έλ κ² λλ€κ³ ν΄μ.
λ°±λ§ λͺ
λΉ μΌμμ¬ λͺ
μ΄λΌλ κ²μ μ€μ κΈ°μ¦νλ
μ μ«μκ°
μμΉκ² μ£ κ·Έκ² μ΄λ£¨μ΄μ§λ €λ©΄ κΈ°μ¦ μμ½μ κ΅μ₯ν λ λ§μ μ¬λλ€μ΄ νκ² λ€μ.
]"""
formatted_prompt = f"Instruction: {prompt}\n output:"
outputs = pipe_finetuned(
formatted_prompt,
temperature=0.1,
top_k=50,
top_p=0.95,
repetition_penalty=1.2,
add_special_tokens=True,
streamer = streamer
)
print(outputs[0]["generated_text"][len(formatted_prompt):])
```
### results
```python
μ λͺ©: λλμ μλ―Έμ νμμ±μ λν ν λ‘
1. λλμ μλ―Έμ μ€μμ±
- λλμ νΉμ λ μ§μ, νΉμ μ¬λλ€κ³Ό ν¨κ» νλ μκ°μ μλ―Ένλ€.
- νΉλ³ν, ν¬λ¦¬μ€λ§μ€μ μ°νν΄λ‘μ€λ₯Ό ν¬ν¨ν μΌλΆ λ μ§μλ κ°μ‘±κ³Ό μΉμ§λ€κ³Ό ν¨κ» λλμ ν μ μλ€.
- λλμ κ°μ‘±κ³Ό μΉμ§λ€κ³Ό ν¨κ» 보λ΄λ μκ°μ΄λΌλ μ μμ μ€μνλ€.
2. λλμ νμμ±
- 곡λ체 μμμ΄ λΆμ‘±νλ©°, κΈ°λΆ λ¬Ένκ° μ μ°©λμ§ μμλ€.
- λλμ μμμ΄ λΆμ‘±νμ¬, κΈ°λΆ λ¬Ένκ° μ 보μλμ§ μμλ€.
3. λλμ μλ‘ μ μκΈ°
- λλμ κΈ°λΆλ‘λΆν° μ»λ κ²μ΄ μλλΌ, κΈ°λΆλ₯Ό ν΅ν΄ μ»λ κ²μ΄ μλλΌ, κΈ°λΆλ₯Ό ν΅ν΄ μ»λ κ²μ΄ μλλΌ, κΈ°λΆλ₯Ό ν΅ν΄ μ»λ κ²μ΄λΌλ μμμ΄ νμνλ€.
- λλμ μμμ ν€μ°κ³ λλ μ μλλ‘ μ λμ μ₯μΉκ° νμνλ€.
4. λλμ μμμ λν λ
Όμ
- λ³νΈμ¬λ λλμ΄ μκΈ° ν볡μ μν κ²μ΄λΌλ μ견μ μ μνλ€.
- λλμ΄ κΈ°λΆλ‘λΆν° μ»λ κ²μ΄λΌλ μ견λ μ μλμλ€.
- λλμ΄ κ²°κ΅ νμκ° λλ κ²μ΄λΌλ μ견λ μ μλμλ€.
```
### Inputs and outputs
* **Input:** Text string, such as a question, a prompt, or a document to be summarized.
* **Output:** Generated Korea text in response to the input, such an answer to a question, or a summary of a minutes.
### Software
Training was done using QLoRA
## Usage and Limitations
These models have certain limitations that users should be aware of.
### Intended Usage
Open Large Language Models (LLMs) have a wide range of applications across
various industries and domains. The following list of potential uses is not
comprehensive. The purpose of this list is to provide contextual information
about the possible use-cases that the model creators considered as part of model
training and development.
* Content Creation and Communication
* Text Generation: These models can be used to generate creative text formats
such as poems, scripts, code, marketing copy, and email drafts.
* Research and Education
* Natural Language Processing (NLP) Research: These models can serve as a
foundation for researchers to experiment with NLP techniques, develop
algorithms, and contribute to the advancement of the field.
* Language Learning Tools: Support interactive language learning experiences,
aiding in grammar correction or providing writing practice.
* Knowledge Exploration: Assist researchers in exploring large bodies of text
by generating summaries or answering questions about specific topics.
### Limitations
* Training Data
* The quality and diversity of the training data significantly influence the
model's capabilities. Biases or gaps in the training data can lead to
limitations in the model's responses.
* The scope of the training dataset determines the subject areas the model can
handle effectively.
* Context and Task Complexity
* LLMs are better at tasks that can be framed with clear prompts and
instructions. Open-ended or highly complex tasks might be challenging.
* A model's performance can be influenced by the amount of context provided
(longer context generally leads to better outputs, up to a certain point).
* Language Ambiguity and Nuance
* Natural language is inherently complex. LLMs might struggle to grasp subtle
nuances, sarcasm, or figurative language.
* Factual Accuracy
* LLMs generate responses based on information they learned from their
training datasets, but they are not knowledge bases. They may generate
incorrect or outdated factual statements.
* Common Sense
* LLMs rely on statistical patterns in language. They might lack the ability
to apply common sense reasoning in certain situations.
### Ethical Considerations and Risks
The development of large language models (LLMs) raises several ethical concerns.
In creating an open model, we have carefully considered the following:
* Bias and Fairness
* LLMs trained on large-scale, real-world text data can reflect socio-cultural
biases embedded in the training material. These models underwent careful
scrutiny, input data pre-processing described and posterior evaluations
reported in this card.
* Misinformation and Misuse
* LLMs can be misused to generate text that is false, misleading, or harmful.
* Guidelines are provided for responsible use with the model, see the
[Responsible Generative AI Toolkit](http://ai.google.dev/gemma/responsible).
* Transparency and Accountability:
* This model card summarizes details on the models' architecture,
capabilities, limitations, and evaluation processes.
* A responsibly developed open model offers the opportunity to share
innovation by making LLM technology accessible to developers and researchers
across the AI ecosystem.
Risks identified and mitigations:
* Perpetuation of biases: It's encouraged to perform continuous monitoring
(using evaluation metrics, human review) and the exploration of de-biasing
techniques during model training, fine-tuning, and other use cases.
* Generation of harmful content: Mechanisms and guidelines for content safety
are essential. Developers are encouraged to exercise caution and implement
appropriate content safety safeguards based on their specific product policies
and application use cases.
* Misuse for malicious purposes: Technical limitations and developer and
end-user education can help mitigate against malicious applications of LLMs.
Educational resources and reporting mechanisms for users to flag misuse are
provided. Prohibited uses of Gemma models are outlined in the
[Gemma Prohibited Use Policy](https://ai.google.dev/gemma/prohibited_use_policy).
* Privacy violations: Models were trained on data filtered for removal of PII
(Personally Identifiable Information). Developers are encouraged to adhere to
privacy regulations with privacy-preserving techniques. |