Update README.md
Browse files
README.md
CHANGED
|
@@ -1,65 +1,69 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: other
|
| 3 |
-
license_name: hyperclovax-seed
|
| 4 |
-
license_link: LICENSE
|
| 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 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
license_name: hyperclovax-seed
|
| 4 |
+
license_link: LICENSE
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
|
| 11 |
+
## Overview
|
| 12 |
+
|
| 13 |
+
HyperCLOVAX-SEED-Text-Instruct-0.5B is a Text-to-Text model with instruction-following capabilities that excels in understanding Korean language and culture. Compared to external competitors of similar scale, it demonstrates improved mathematical performance and a substantial enhancement in Korean language capability. The HyperCLOVAX-SEED-Text-Instruct-0.5B is currently the smallest model released by the HyperCLOVAX, representing a lightweight solution suitable for deployment in resource‑constrained environments such as edge devices. It supports a maximum context length of 4K and functions as a versatile small model applicable to a wide range of tasks. The total cost of a single training run for HyperCLOVAX-SEED-Text-Instruct-0.5B was 4.358K A100 GPU hours (approximately USD 6.537K), which is 39 times lower than the cost of training the `QWEN2.5‑0.5B‑instruct` model.
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
## Basic Information
|
| 17 |
+
|
| 18 |
+
- **Architecture**: Transformer‑based (Dense Model)
|
| 19 |
+
- **Parameters**: 0.57 B (total); 0.45 B (excluding token embeddings, tied embeddings)
|
| 20 |
+
- **Input/Output Format**: Text / Text
|
| 21 |
+
- **Maximum Context Length**: 4 K tokens
|
| 22 |
+
- **Knowledge Cutoff Date**: Trained on data up to January 2025
|
| 23 |
+
|
| 24 |
+
|
| 25 |
+
## Training and Data
|
| 26 |
+
|
| 27 |
+
The training dataset for HyperCLOVAX-SEED-Text-Instruct-0.5B consists of diverse sources, including the high‑quality data accumulated during the development of HyperCLOVAX-SEED-Text-Instruct-0.5B. Training was conducted in three main stages:
|
| 28 |
+
1. **Pretraining**: Knowledge acquisition using high‑quality data and a high‑performance pretrained model.
|
| 29 |
+
2. **Rejection Sampling Fine‑Tuning (RFT)**: Enhancement of multi‑domain knowledge and complex reasoning capabilities.
|
| 30 |
+
3. **Supervised Fine‑Tuning (SFT)**: Improvement of instruction‑following proficiency.
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
## Training Cost
|
| 34 |
+
|
| 35 |
+
HyperCLOVAX-SEED-Text-Instruct-0.5B leveraged HyperCLOVA X’s lightweight training process and high‑quality data to achieve significantly lower training costs compared to industry‑leading competitors of similar scale. Excluding the SFT stage, a single pretraining run incurred:
|
| 36 |
+
|
| 37 |
+
| Pretraining Cost Category | HyperCLOVAX-SEED-Text-Instruct-0.5B | QWEN2.5‑0.5B‑instruct |
|
| 38 |
+
|---------------------------------|-----------------------------------------------|-------------------------------------|
|
| 39 |
+
| **A100 GPU Hours** | 4.358 K | 169.257 K |
|
| 40 |
+
| **Cost (USD)** | 6.537 K | 253.886 K |
|
| 41 |
+
|
| 42 |
+
This represents approximately a 39× reduction in pretraining cost relative to `QWEN2.5‑0.5B-instruct`.
|
| 43 |
+
|
| 44 |
+
## Benchmarks
|
| 45 |
+
|
| 46 |
+
| **Model** | **KMMLU (5-shot, acc)** | **HAE-RAE (5-shot, acc)** | **CLiCK (5-shot, acc)** | **KoBEST (5-shot, acc)** |
|
| 47 |
+
| --- | --- | --- | --- | --- |
|
| 48 |
+
| HyperCLOVAX-SEED-Text-Base-0.5B | 0.4181 | 0.6370 | 0.5373 | 0.6963
|
| 49 |
+
| HyperCLOVAX-SEED-Text-Instruct-0.5B | 0.3815 | 0.5619 | 0.4446 | 0.6299 |
|
| 50 |
+
| QWEN2.5-0.5B-instruct | 0.2968 | 0.3428 | 0.3805 | 0.5025 |
|
| 51 |
+
|
| 52 |
+
## HuggingFace Usage Example
|
| 53 |
+
|
| 54 |
+
```python
|
| 55 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 56 |
+
model = AutoModelForCausalLM.from_pretrained("/path/to/HyperCLOVAX-SEED-Text-Instruct-0.5B")
|
| 57 |
+
tokenizer = AutoTokenizer.from_pretrained("/path/to/HyperCLOVAX-SEED-Text-Instruct-0.5B")
|
| 58 |
+
|
| 59 |
+
chat = [
|
| 60 |
+
{"role": "tool_list", "content": ""},
|
| 61 |
+
{"role": "system", "content": "- AI 언어모델의 이름은 \"CLOVA X\" 이며 네이버에서 만들었다.\n- 오늘은 2025년 04월 24일(목)이다."},
|
| 62 |
+
{"role": "user", "content": "슈뢰딩거 방정식과 양자역학의 관계를 최대한 자세히 알려줘."},
|
| 63 |
+
]
|
| 64 |
+
|
| 65 |
+
inputs = tokenizer.apply_chat_template(chat, add_generation_prompt=True, return_dict=True, return_tensors="pt")
|
| 66 |
+
output_ids = model.generate(**inputs, max_length=1024, stop_strings=["<|endofturn|>", "<|stop|>"], tokenizer=tokenizer)
|
| 67 |
+
print(tokenizer.batch_decode(output_ids))
|
| 68 |
+
```
|
| 69 |
+
|