FAQ (Frequently Asked Questions) on HyperCLOVAX-SEED-Text-Instruct-0.5B

#1
by hyperclovax - opened

(Last updated: 24 April, 2025)

Q. What are the requirements of HyperCLOVAX-SEED-Text-Instruct-0.5B?

  • Parameters: 0.5 billion
  • Context Length: 4K tokens
  • CPU: Modern processor with at least 8 cores
  • RAM: Minimum of 16 GB recommended.
  • GPU: NVIDIA GPU (at least 12 GB) for 16-bit mode.
  • Storage (Disk Space): Approximately 10GB for the model and associated data.
  • Estimated GPU Memory Requirements:
    (Higher Precision Modes) 32-bit Mode: ~3 GB / 16-bit Mode: ~1.5 GB
    (Lower Precision Modes) 8-bit Mode: ~0.75 GB / 4-bit Mode: ~0.325 GB
  • Operating System: Linux or Windows (Linux preferred for better performance)
  • Programming Language: Python 3.7 or higher
  • Frameworks: PyTorch (preferred) or TensorFlow
  • Libraries: Hugging Face Transformers, NumPy, Pandas.

Q. What is the HyperCLOVA X SEED model and what is its composition?
HyperCLOVA X SEED is an open-source AI model offering a set of compact yet powerful solutions engineered for adaptability. It comprises three distinct models with 3B, 1.5B, and 0.5B parameters, each tailored for specific applications and capable of being fine-tuned for customized use across diverse business sectors.

(KOR) HyperCLOVA X SEED λͺ¨λΈμ΄λž€ 무엇이며, μ–΄λ–»κ²Œ κ΅¬μ„±λ˜μ–΄ μžˆλ‚˜μš”?
HyperCLOVA X SEED μ˜€ν”ˆμ†ŒμŠ€ λͺ¨λΈμ€ κΈ°μ—…μ΄λ‚˜ κ°œλ°œμžκ°€ νŠΉμ • λΉ„μ¦ˆλ‹ˆμŠ€μ˜ μš”κ΅¬μ‚¬ν•­μ— 맞게 νŠœλ‹ν•˜μ—¬ λ‹€μ–‘ν•œ μ‚°μ—… 뢄야에 λ§žμΆ€ν˜•μœΌλ‘œ μ μš©ν•  수 μžˆμŠ΅λ‹ˆλ‹€. 3B, 1.5B, 0.5B νŒŒλΌλ―Έν„°λ₯Ό κ°€μ§„ μ„Έ κ°€μ§€ λͺ¨λΈλ‘œ μΆœμ‹œλ˜μ–΄ 각각 νŠΉν™”λœ μ˜μ—­μ—μ„œ μ‹€μ§ˆμ μΈ μ„±λŠ₯을 λ°œνœ˜ν•©λ‹ˆλ‹€.


Q. Which model best fits my needs?
Each of the three HyperCLOVA X SEED models is designed for distinct use cases. See the descriptions below to determine the most suitable option for your specific requirements.

[HyperCLOVA X SEED 3B]

  • Key Features: Multimodal (image & text understanding), excels in Korean linguistic and cultural contexts.
  • Recommended Use: Image and Korean cultural understanding (e.g., tour guidance, visual analysis, image Q&A).

[HyperCLOVA X SEED 1.5B]

  • Key Features: Instruction-following language model, versatile, good at Korean translation and complex instructions (e.g., JSON, casual).
  • Recommended Use: Translation, educational/business assistance, complex instruction-based applications, specialized chatbots.

[HyperCLOVA X SEED 0.5B]

  • Key Features: Compact language model, fluent Korean multi-turn conversations, efficient for limited resources, good for everyday inquiries.
  • Recommended Use: Mobile/smart home/wearable conversational interfaces, lightweight AI, customer support chatbots.

(KOR) μ €μ—κ²Œ κ°€μž₯ μ ν•©ν•œ λͺ¨λΈμ€ λ¬΄μ—‡μΈκ°€μš”?
HyperCLOVA X SEED μ„Έ κ°€μ§€ λͺ¨λΈμ€ 각각의 κ³ μœ ν•œ μ‚¬μš© 사둀λ₯Ό μœ„ν•΄ μ„€κ³„λ˜μ—ˆμŠ΅λ‹ˆλ‹€. μ•„λž˜μ˜ μ„€λͺ…을 μ°Έκ³ ν•˜μ—¬ 본인의 νŠΉμ • μš”κ΅¬ 사항에 κ°€μž₯ μ•Œλ§žμ€ λͺ¨λΈμ„ 선택해 λ³΄μ„Έμš”.

[HyperCLOVA X SEED 3B]

  • μ£Όμš” νŠΉμ§•: 이미지/μ˜μƒ 이해, ν•œκ΅­μ–΄ μ–Έμ–΄ 및 문화적 λ§₯락 이해에 탁월
  • ꢌμž₯ μ‚¬μš© 사둀: 이미지/μ˜μƒ 뢄석, μ‹œκ° 정보 기반 μ§ˆμ˜μ‘λ‹΅, κ΄€κ΄‘ κ°€μ΄λ“œ λ“± λ©€ν‹°λͺ¨λ‹¬ μ„œλΉ„μŠ€

[HyperCLOVA X SEED 1.5B]

  • μ£Όμš” νŠΉμ§•: μš°μˆ˜ν•œ μ§€μ‹œ 이행 λŠ₯λ ₯, ν•œκ΅­μ–΄ λ²ˆμ—­ 및 λ³΅μž‘ν•œ μ§€μ‹œ(예: JSON, 일상 λŒ€ν™”μ²΄)에 λŠ₯μˆ™
  • ꢌμž₯ μ‚¬μš© 사둀: λ‹€κ΅­μ–΄ λ²ˆμ—­ 보쑰, νŠΉμ • 포맷으둜 λ‹΅λ³€ 좜λ ₯ (JSON λ“±), μ–΄μ‘° μ œμ–΄ λ“± κ³ λ„ν™”λœ 챗봇/μ• ν”Œλ¦¬μΌ€μ΄μ…˜

[HyperCLOVA X SEED 0.5B]

  • μ£Όμš” νŠΉμ§•: μ†Œν˜• μ–Έμ–΄ λͺ¨λΈμ΄μ§€λ§Œ ν•œκ΅­μ–΄ λ©€ν‹°ν„΄ λŒ€ν™” λŠ₯λ ₯ 우수
  • ꢌμž₯ μ‚¬μš© 사둀: μ œν•œλœ λ¦¬μ†ŒμŠ€ ν™˜κ²½μ—μ„œμ˜ λŒ€ν™”ν˜• μΈν„°νŽ˜μ΄μŠ€ (λͺ¨λ°”일 μ–΄ν”Œλ¦¬μΌ€μ΄μ…˜, μŠ€λ§ˆνŠΈν™ˆ κΈ°κΈ°, μ›¨μ–΄λŸ¬λΈ” λ””λ°”μ΄μŠ€ λ“±)

Q. How to use it?
To start using these models, visit our Hugging Face page at https://huggingface.co/naver-hyperclovax. You can find more specific details about each model within its respective model card there.

(KOR) HyperCLOVA X SEEDλŠ” μ–΄λ–»κ²Œ μ‚¬μš©ν•˜λ‚˜μš”?
HyperCLOVA X SEED μ˜€ν”ˆμ†ŒμŠ€ λͺ¨λΈμ€ ν—ˆκΉ…νŽ˜μ΄μŠ€ νŽ˜μ΄μ§€ https://huggingface.co/naver-hyperclovax μ—μ„œ 직접 λ‹€μš΄λ‘œλ“œ λ°›μ•„ μ‚¬μš©ν•˜μ‹€ 수 μžˆμŠ΅λ‹ˆλ‹€. 각 λͺ¨λΈμ— λŒ€ν•œ μžμ„Έν•œ μ •λ³΄λŠ” ν•΄λ‹Ή λͺ¨λΈ μΉ΄λ“œλ₯Ό 톡해 확인해 λ³΄μ„Έμš”.


Q. Who can use the models, and what are the licensing conditions?
HyperCLOVA X SEED models are broadly available for use by businesses, researchers, and developers. We're committed to encouraging innovation in both research and commercial settings while ensuring responsible AI practices. Detailed information about the license conditions can be found in the LICENSE file, available for download on the following page:

(KOR) λˆ„κ°€ 이 λͺ¨λΈμ„ μ‚¬μš©ν•  수 있으며, λΌμ΄μ„ μŠ€ 쑰건은 μ–΄λ–»κ²Œ λ˜λ‚˜μš”?
HyperCLOVA X SEED λͺ¨λΈμ€ κΈ°μ—…, μ—°κ΅¬μž, 개발자 λͺ¨λ‘κ°€ ν­λ„“κ²Œ μ‚¬μš©ν•  수 μžˆμŠ΅λ‹ˆλ‹€. μ €ν¬λŠ” 상업, 연ꡬ ν™˜κ²½ λͺ¨λ‘μ—μ„œ ν˜μ‹ μ„ μ΄‰μ§„ν•˜κΈ° μœ„ν•΄ λ…Έλ ₯ν•˜λ©°, μ±…μž„κ° μžˆλŠ” AI μ‚¬μš©μ„ μœ„ν•΄ λ…Έλ ₯ν•˜κ³  μžˆμŠ΅λ‹ˆλ‹€. λΌμ΄μ„ μŠ€ 쑰건에 λŒ€ν•œ μžμ„Έν•œ 사항은 μ•„λž˜ νŽ˜μ΄μ§€μ—μ„œ LICENSEλ₯Ό λ‹€μš΄λ‘œλ“œ ν•˜μ—¬ ν™•μΈν•˜μ‹€ 수 μžˆμŠ΅λ‹ˆλ‹€.


Q. What if I'm unsure whether my use is allowed under the License Policy?
If you need further clarification, please contact NAVER Cloud Platform Customer Inquiry . We will consider requests on a case-by-case basis.

(KOR) μ €μ˜ λͺ¨λΈ μ‚¬μš©μ΄ λΌμ΄μ„ μŠ€ 정책에 λ§žλŠ”μ§€ ν™•μΈν•˜λ €λ©΄ μ–΄λ–»κ²Œ ν•΄μ•Όν•˜λ‚˜μš”?
넀이버 ν΄λΌμš°λ“œ ν”Œλž«νΌ 고객센터에 λ°©λ¬Έν•˜μ—¬ μ‚¬μš© λͺ©μ κ³Ό 적용 μ„œλΉ„μŠ€ 등을 λ³΄λ‚΄μ£Όμ‹œλ©΄, κ°œλ³„μ μœΌλ‘œ κ²€ν†  ν›„ λ‹΅μ‹  λ“œλ¦¬κ² μŠ΅λ‹ˆλ‹€. 


Q. What is the usage scope of HyperCLOVA X SEED? 
HyperCLOVA X SEED is primarily a commercial-free open-source resource. It's available for use in services with 10 million or fewer MAU, where "MAU" denotes the monthly active users accessing or utilizing your products and services.

A separate license from NAVER is required if: 
(i) The monthly active users of products or services (including commercial and non-commercial use) offered by the license provider or its affiliates surpassed 10 million in the preceding month. 
(ii) The license provider or its affiliates distribute or provide products or services that are substantially similar to or directly compete with NAVER's offerings.

For these situations, please contact NAVER Cloud Platform Customer Inquiry . We will consider requests on a case-by-case basis.

Detailed information about the license conditions can be found in the LICENSE file, available for download on the following page:

(KOR) HyperCLOVA X SEED의 μ‚¬μš© λ²”μœ„λŠ” μ–΄λ–»κ²Œ λ˜λ‚˜μš”?

HyperCLOVA X SEEDλŠ” μ›”κ°„ ν™œμ„± μ‚¬μš©μž(MAU)κ°€ 1천만 λͺ… μ΄ν•˜μΈ μ„œλΉ„μŠ€μ—μ„œ 무료둜 μ‚¬μš©ν•  수 μžˆλŠ” μ˜€ν”ˆμ†ŒμŠ€ μžμ›μž…λ‹ˆλ‹€. μ—¬κΈ°μ„œ 'MAU'λŠ” κ·€μ‚¬μ˜ μ œν’ˆμ΄λ‚˜ μ„œλΉ„μŠ€λ₯Ό μ›”κ°„ μ΄μš©ν•˜λŠ” ν™œμ„± μ‚¬μš©μžλ₯Ό λœ»ν•©λ‹ˆλ‹€.

λ‹€μŒκ³Ό 같은 κ²½μš°μ—λŠ” λ„€μ΄λ²„λ‘œλΆ€ν„° λ³„λ„μ˜ λΌμ΄μ„ μŠ€λ₯Ό λ°›μ•„μ•Ό ν•©λ‹ˆλ‹€. 
(i) λΌμ΄μ„ μŠ€ μ œκ³΅μžλ‚˜ κ·Έ 계열사가 μ œκ³΅ν•˜λŠ” μ œν’ˆ λ˜λŠ” μ„œλΉ„μŠ€(상업적 및 비상업적 μš©λ„ λͺ¨λ‘ 포함)의 μ›”κ°„ ν™œμ„± μ‚¬μš©μžκ°€ 이전 달에 1천만λͺ…을 μ΄ˆκ³Όν•œ 경우
(ii) λΌμ΄μ„ μŠ€ μ œκ³΅μžλ‚˜ κ·Έ 계열사가 λ„€μ΄λ²„μ˜ μ„œλΉ„μŠ€μ™€ μ‹€μ§ˆμ μœΌλ‘œ μœ μ‚¬ν•˜κ±°λ‚˜ μ§μ ‘μ μœΌλ‘œ κ²½μŸν•˜λŠ” μ œν’ˆ λ˜λŠ” μ„œλΉ„μŠ€λ₯Ό λ°°ν¬ν•˜κ±°λ‚˜ μ œκ³΅ν•˜λŠ” 경우

상기에 ν•΄λ‹Ήν•˜λŠ” 경우, 넀이버 ν΄λΌμš°λ“œ ν”Œλž«νΌ 고객센터에 λ°©λ¬Έν•˜μ—¬ μ‚¬μš© λͺ©μ κ³Ό 적용 μ„œλΉ„μŠ€ 등을 λ³΄λ‚΄μ£Όμ‹œλ©΄ κ²€ν†  ν›„ λ‹΅μ‹  λ“œλ¦¬κ² μŠ΅λ‹ˆλ‹€. 

λΌμ΄μ„ μŠ€ 쑰건에 λŒ€ν•œ μžμ„Έν•œ 사항은 μ•„λž˜ νŽ˜μ΄μ§€μ—μ„œ LICENSEλ₯Ό λ‹€μš΄λ‘œλ“œ ν•˜μ—¬ ν™•μΈν•˜μ‹€ 수 μžˆμŠ΅λ‹ˆλ‹€.


Q. What hardware is needed to test and deploy these models?
Hardware needs depend on the specific HyperCLOVA X SEED model, desired latency and throughput, and budget. These models are designed to run efficiently on diverse cloud infrastructure and hardware, including GPUs, CPUs, TPUs, NPUs, and AI Accelerators.

(KOR) λͺ¨λΈ ν…ŒμŠ€νŠΈμ™€ 배포에 ν•„μš”ν•œ μΈν”„λΌλŠ” λ¬΄μ—‡μΈκ°€μš”?
ν•„μš”ν•œ ν•˜λ“œμ›¨μ–΄λŠ” νŠΉμ • HyperCLOVA X SEED λͺ¨λΈ, μ›ν•˜λŠ” μ§€μ—° μ‹œκ°„(latency) 및 μ²˜λ¦¬λŸ‰(throughput), 그리고 μ˜ˆμ‚°μ— 따라 λ‹¬λΌμ§‘λ‹ˆλ‹€. GPU, CPU, TPU, NPU 및 AI 가속기 λ“± λ‹€μ–‘ν•œ ν΄λΌμš°λ“œ 인프라 및 ν•˜λ“œμ›¨μ–΄μ—μ„œ 효율적으둜 μž‘λ™ν•˜λ„λ‘ μ„€κ³„λ˜μ—ˆμŠ΅λ‹ˆλ‹€.


Q. Can I run the models on a consumer-grade laptop?
While technically feasible, a standard laptop isn't really ideal for running these models. You'd have a much smoother experience with a desktop that has a powerful multi-core CPU and a GPU with plenty of memory.

(KOR) λ…ΈνŠΈλΆ μˆ˜μ€€μ˜ PC둜 이 λͺ¨λΈλ“€μ„ μ‹€ν–‰ν•  수 μžˆλ‚˜μš”?
κΈ°μˆ μ μœΌλ‘œλŠ” κ°€λŠ₯ν•˜μ§€λ§Œ, λͺ¨λΈ ν…ŒμŠ€νŠΈμ™€ 배포에 이상적인 ν™˜κ²½μ€ μ•„λ‹™λ‹ˆλ‹€. κ°•λ ₯ν•œ λ©€ν‹°μ½”μ–΄ CPUλ‚˜ GPU, μΆ©λΆ„ν•œ λ©”λͺ¨λ¦¬λ₯Ό κ°–μΆ˜ λ°μŠ€ν¬ν†±μ„ μ‚¬μš©ν•˜λ©΄ 훨씬 더 μΎŒμ ν•œ κ²½ν—˜μ„ ν•˜μ‹€ 수 μžˆμŠ΅λ‹ˆλ‹€.


Q. Is there any promotional offer for test deployment?
We currently offer a program to support the cloud infrastructure needed for HyperCLOVA X SEED testing. Through this, 100 eligible applicants will receive 2 million KRW in credits for a two-month period to facilitate their testing of HyperCLOVA X SEED on NAVER Cloud Platform. (*This promotion is for domestic companies based in Korea.)

Please refer to the blog post below for more details:

(KOR) 인프라 지원 κ΄€λ ¨ ν”„λ‘œλͺ¨μ…˜μ΄ μžˆλ‚˜μš”?
HyperCLOVA X SEEDλ₯Ό ν™œμš©ν•œ ν…ŒμŠ€νŠΈκ°€ ν•„μš”ν•œ κΈ°μ—…κ³Ό 기관을 μœ„ν•΄, ν•„μš”ν•œ ν΄λΌμš°λ“œ 인프라λ₯Ό μ§€μ›ν•˜λŠ” ν˜œνƒμ„ λ§ˆλ ¨ν–ˆμŠ΅λ‹ˆλ‹€. μ„ μ •λœ 100개 κΈ°μ—…μ—κ²ŒλŠ” HyperCLOVA X SEED ν…ŒμŠ€νŠΈμ— μ‚¬μš©ν•  수 μžˆλŠ” 200만 원 ν¬λ ˆλ”§μ„ 2κ°œμ›”κ°„ μ œκ³΅ν•©λ‹ˆλ‹€.  (* λ³Έ ν”„λ‘œλͺ¨μ…˜μ€ ν•œκ΅­μ— κΈ°λ°˜μ„ λ‘” κ΅­λ‚΄ κΈ°μ—… λŒ€μƒμž…λ‹ˆλ‹€.)

μžμ„Έν•œ λ‚΄μš©μ€ μ•„λž˜ λΈ”λ‘œκ·Έλ₯Ό μ°Έκ³ ν•΄ μ£Όμ„Έμš”.


Q. Do HyperCLOVA X SEED models support languages other than Korean?
The HyperCLOVA X SEED model was primarily trained on Korean data, with some other languages included as well.

(KOR) HyperCLOVA X SEED λͺ¨λΈμ€ ν•œκ΅­μ–΄ μ™Έ λ‹€λ₯Έ 언어도 μ§€μ›ν•˜λ‚˜μš”?
HyperCLOVA X SEED λͺ¨λΈμ€ 주둜 ν•œκ΅­μ–΄ λ°μ΄ν„°λ‘œ ν•™μŠ΅λ˜μ—ˆκ³ , 일뢀 λ‹€λ₯Έ 언어도 같이 ν•™μŠ΅λ˜μ—ˆμŠ΅λ‹ˆλ‹€.

Your need to confirm your account before you can post a new comment.

Sign up or log in to comment