LLM에서 배우는 이미지 생성 모델 ZERO부터 학습하기

LLMীࢲ ߓ਋ח ੉޷૑ ࢤࢿ ݽ؛ ZEROࠗఠ ೟णೞӝ ӣకള DBSQFEN Training
Large-Scale Diffusion Model from Scratch

ӣకള ೧ఊ गಌ ஹೊ౴ ೟ࠗ ઔস റ 0QFO"* ੑࢎ ূ૑פয
ݺਵ۽ Ӗ۽ߥ ݅ݺ ਬ੷ܳ ݽ਷ ହস 4)*'561 carpedm20

ӣకള carpedm20 ௿ۄ੉߁ 🧗 ࢟۞٘ 🥗 ؋റ ѐߊ ೠ׮Ҋ ١ೠद
೮؍ ࣁ࢚ী ؀೧ ঌҊ रযࢲ ࢜۽਍ ݅թ੉ۆ ݽ੐ ֙ р ೧ழ ೞ਋झীࢲ ૑ղݶࢲ ଼ ੍Ҋ Ӗ ॳҊ ࢎۈ ݅աҊ ֙ ੉࢚ ਬ੄޷೮؍ ৈݺҗ੄ ؀ചܳ ӝ۾

"*$PNQJMFS4UVEZ ♟ ♟

$POUFOU അসীࢲ וՙח ੉޷૑ ࢤࢿ ݽ؛੄ ೠ҅ --.
೟णীࢲ ߓ਋ӝ #%JGGVTJPO.PEFM೟णਸ ೞݴ ߓ਍ ੼

%JGGVTJPO ݽ؛

4UBCMF%JGGVTJPO

ਗೞח Ӓܿ୓ ٣పੌೠ ಴അ

ೞ૑݅ ੉۠ ࢎ۹ীب ࠛҳೞҊ

ѱ੐ ഥࢎীࢲ וՙח %JGGVTJPO ݽ؛੄ ೠ҅

4UBCMF%JGGVTJPO

Ӓܿ୓ ٣పੌ ࠗ઒ೠ ౠ࢝ হח

೙ਃೠ ੘স दр গݒೠ Ӓܿ Ҋ஖ӝ ૒੽ Ӓܻӝ

%JGGVTJPO

ৈ੹൤ ઓ੤ೞח ݆ࣻ਷ ޙઁ

৔য੄ য۰਑ ৮ࢿب ੓ח Ӓܿ੉ ੜ աয়૑ ঋח׮
ౠ੿ झఋੌ۽ ੜ ӒܻҊ र׮ э਷ நܼఠ ঴ҷ ௏झౝ ੄ ׮ܲ ੗ࣁ Ӓܻӝ ੜ Ӓܻ૑ ޅೞҊ ݈۽ ಴അೞӝ য۰਍ Ѫٜ • ঈࣁࢲܻ • ׮নೠ ੄࢚ ੤૕ • ߓ҃ ހझఠ • ࡄ੄ ߑೱ ҳب • ޛ୓ р ҙ҅ ୨ਸ ١ী ݒҊ ੓Ѣա নࣚਵ۽ ੟Ҋ ੓ח

п ࢚ടী ݏѱ -P3"ܳ ٜ݅Ҋ $POUSPM/FUܳ ࢚ട ߹۽ ଺ח ߑߨূ
ೠ҅о ੓׮ ঈࣁࢲܻ झఋੌ ׮নೠ ੄࢚ ੤૕ ߓ҃ ހझఠ ࡄ੄ ߑೱ ҳب ޛ୓ р ҙ҅

Foundation Model જ਷ द੘੼

Training Large-Scale Diffusion Model from Scratch

Foundation Model ೟ण੄ о੢ ௾ ੢߷

NVIDIA A100 Tensor Core GPU

https://www.databricks.com/blog/stable-diffusion-2

ୌ ݅ਗ Yߣ র

যڌѱ ೞݶ ബਯ੸ਵ۽ ೟णೡ ࣻ ੓ਸө?

Diffusionࠁ׮ ഝߊ൤ োҳغҊ ੓ח LLM োҳܳ ݆੉ ଵҊ

OpenAIী ੓؍ ஘ҳо ա৬ࢲ द੘ೠ ೐۽ં౟ ࢎप࢚ पಁ೮૑݅ Compute੉ ࠗ઒ೠ
౱ীѱ ݆਷ ߓ਎ ੼ਸ ҕਬೣ *ؘ੉ఠী ցޖ ૘઺ਸ ೞ૑ ঋও਺

Continual Learning

015౱੉ प೷ ࢗ૕ ೠ ѐ੄ प೷ਸ ੿ܻೠ -PHCPPL IUUQTHJUIVCDPNGBDFCPPLSFTFBSDINFUBTFRCMPCNBJOQSPKFDUT015DISPOJDMFT015#@-PHCPPLQEG

• ؘ੉ఠо ݆ই૓ ݅ఀ ݽ؛ ௼ӝب ழઉঠ ୭ࢶ੄ ࢿמਸ ࠁ੢ೡ
ࣻ ੓׮ • $PNQVUF੉ ઁೠػ ࢚ടীࢲ ୭੸੄ ݽ؛ ߂ ؘ੉ఠ ௼ӝܳ ҅࢑ೡ ࣻ ੓׮ ؊ ݆਷ ؘ੉ఠ۽ જ਷ ࢿמਸ ղח ੘਷ ݽ؛ਸ ೟णೣ

• ؘ੉ఠо ݆ই૓ ݅ఀ ݽ؛ ௼ӝب ழઉঠ ୭ࢶ੄ ࢿמਸ ࠁ੢ೡ
ࣻ ੓׮ • $PNQVUF੉ ઁೠػ ࢚ടীࢲ ୭੸੄ ݽ؛ ߂ ؘ੉ఠ ௼ӝܳ ৘ஏೡ ࣻ ੓׮

• 110۽ ࢎۈٜ੉ ؊ ࢶഐೞח ݽ؛ਸ ೟ण • ࠺Үо ࢤࢿࠁ׮
ܻࣗझо ੸ѱ ٜӝ ٸޙ • 3FXBSE NPEFMਸ ೟ण೧ঠ ೞա %10ܳ ॳݶ ࠂ੟بо ઴য ٝ MFTTGMFYJCMF

• 110۽ ࢎۈٜ੉ ؊ ࢶഐೞח ݽ؛ਸ ೟ण • ࠺Үо ࢤࢿࠁ׮
ܻࣗझо ੸ѱ ٜӝ ٸޙ

• જ਷ ௬ܻ౭੄ 4ZOUIFUJDؘ੉ఠח 8FC$SBXM ࠁ׮ ഻ঁ ੸਷ নਵ۽ જ਷
ݽ؛ਸ ೟ण दఆ ࣻ ੓׮ • EBZTPO"(16T ୌ݅ਗ ੉ೞ ChatGPT & GPT-4

8FSFNBSLUIBUUIFFYQFSJFODFHBJOFEJOUIFQSPDFTTPGDSFBUJOHUIF USBJOJOHEBUBGPSCPUIQIJBOEQIJMFBETVTUPUIFDPODMVTJPOUIBU UIFDSFBUJPOPGBSPCVTUBOEDPNQSFIFOTJWFEBUBTFUEFNBOETNPSFUIBO SBXDPNQVUBUJPOBMQPXFS *USFRVJSFTJOUSJDBUFJUFSBUJPOT TUSBUFHJDUPQJDTFMFDUJPO BOEBEFFQ VOEFSTUBOEJOHPGLOPXMFEHFHBQTUPFOTVSFRVBMJUZBOEEJWFSTJUZPGUIF EBUB8FTQFDVMBUFUIBUUIFDSFBUJPOPGTZOUIFUJDEBUBTFUTXJMMCFDPNF JO
UIFOFBSGVUVSF BOJNQPSUBOUUFDIOJDBMTLJMM BOEBDFOUSBMUPQJDPGSFTFBSDI JO"* Textbooks Are All You Need II: phi-1.5 technical report

8FSFNBSLUIBUUIFFYQFSJFODFHBJOFEJOUIFQSPDFTTPGDSFBUJOHUIF USBJOJOHEBUBGPSCPUIQIJBOEQIJMFBETVTUPUIFDPODMVTJPOUIBUUIF DSFBUJPOPGBSPCVTUBOEDPNQSFIFOTJWFEBUBTFUEFNBOETNPSFUIBOSBX DPNQVUBUJPOBMQPXFS *USFRVJSFTJOUSJDBUFJUFSBUJPOT TUSBUFHJDUPQJDTFMFDUJPO BOEBEFFQ VOEFSTUBOEJOHPGLOPXMFEHFHBQTUPFOTVSFRVBMJUZBOEEJWFSTJUZPGUIF EBUB8FTQFDVMBUFUIBUUIFDSFBUJPOPGTZOUIFUJDEBUBTFUTXJMMCFDPNF JO
UIFOFBSGVUVSF BOJNQPSUBOUUFDIOJDBMTLJMMBOEBDFOUSBMUPQJDPGSFTFBSDI JO"* Textbooks Are All You Need II: phi-1.5 technical report જ਷ ؘ੉ఠ ࣇ ݅٘ח Ѣ ѐ য۰਑ 😓 ੉Ѣ שٜب ೙ਃೡ బؘ 🤣

LLM োҳীࢲ ଵҊೡ Ѫ

LLM োҳীࢲ ଵҊೡ Ѫ -FBSOJOHSBUF TDIFEVMF ݽ؛ ߓ஖ ࢎ੉ૉ

LLM োҳীࢲ ଵҊೡ Ѫ -FBSOJOHSBUF TDIFEVMF ݽ؛ ߓ஖ ࢎ੉ૉ উ੿੸ੋ
ݽ؛ ೟णਸ ਤ೧ ࠊঠೞח NFUSJD $POUJOVBM -FBSOJOH ۄ૑ ݽ؛ ೟ण द ߊࢤೞח ߡӒ

LLM োҳীࢲ ଵҊೡ Ѫ -FBSOJOHSBUF TDIFEVMF ݽ؛ ߓ஖ ࢎ੉ૉ উ੿੸ੋ
ݽ؛ ೟णਸ ਤ೧ ࠊঠೞח NFUSJD $POUJOVBM -FBSOJOH ۄ૑ ݽ؛ ೟ण द ߊࢤೞח ߡӒ જ਷ 4ZOUIFUJD ؘ੉ఠ ݅٘ח ߨ ݽ؛ ٣੗ੋ

੉ ৻ীب t5IFGJOBMEBUBNJYUVSFTBOEXFJHIUTXFSF EFUFSNJOFEUISPVHIBCMBUJPOTPOTNBMMFSNPEFMTu t8FTUBHFUSBJOJOHUPBMUFSUIFNJYUVSFDPNQPTJUJPOEVSJOHUSBJOJOH JODSFBTJOHUIFXFJHIUPGEPNBJOSFMFWBOUEBUBUPXBSETUIFFOEPGUSBJOJOHu Gemini: A Family of
Highly Capable Multimodal Models

੉ ৻ীب Gemini: A Family of Highly Capable Multimodal Models
5SBJOFEPOMBSHFDPMMFDUJPOPG 8FCEPDVNFOUT CPPLT BOEDPEF JNBHF BVEJP BOEWJEFP 2VBMJUZGJMUFSJOHGPSBMMEBUBTFUT )FVSJTUJDT NPEFMCBTFEDMBTTJGJFST 'JOBMEBUBNJYUVSFTXFJHIUTEFUFSNJOFEUISPVHIBCMBUJPOPOTNBMMFSNPEFMT *ODSFBTFEXFJHIUPGEPNBJOSFMFWBOUEBUB UPXBSETUIFFOEPGUSBJOJOH

੉ ৻ীب ؘ੉ఠо ઁೠػ ࢚ടীࢶ • ؘ੉ఠܳ ߈ࠂ > ݽ؛
௼ӝ ¬ • ௏٘ܳ ୶о > ӝઓ ؘ੉ఠ ߈ࠂ • 1FSQMFYJUZ ೙ఠ݂ > %FEVQMJDBUJPO Scaling Data-Constrained Language Models

׮द جই৬ࢲ

Foundation Model જ਷ द੘੼

LLMীࢲ੄ द೯ ଱য়ܳ ߄ఔਵ۽ Diffusion ݽ؛ ೟णਸ ੽Ӕ

೟ण

• striking red hair • horned black headpiece • piercing
blue eyes • crimson lipstick • sleek bodysuit • black leather belts • silver buckles • gunmetal gray rifle • imposing stance • flowing red cape • knee-high footwear • dark background • powerful posture • detailed armor • black fingerless gloves • red accents • light skin tone • animated style • sharp facial features • glossy lip shine • aggressive expression • asymmetrical design • thigh strap holster • mechanical arm pieces • crimson thigh-highs • bright red scarf • scarlet cape interior • high-heeled boots • compact shoulder pads • reflective surfaces

Caption੄ ௬ܻ౭о ୭ઙ ݽ؛ ࢿמী ݒ਋ ઺ਃ

DALL·E 3 ੉޷૑ ஭࣌ ࢿמ੉ ݒ਋ ઺ਃ ੄
ૣ਷ ஭࣌ ੄ ٣పੌೠ ஭࣌ਵ۽ ೟ण 5ఫझ౟ ੋ௏؊ ࢎਊ (15۽ ࢎਊ੗੄ ૣ਷ ೐܁೐౟ܳ VQTBNQMFೣ Improving Image Generation with Better Captions

• White naval cap • Cerulean blue hair • Military-inspired
uniform • Golden epaulettes • Stern gaze • Light blue eyes

GPT4-Vision ShareGPT4V: Improving Large Multi-Modal Models with Better Captions

GPT4-Vision ShareGPT4V: Improving Large Multi-Modal Models with Better Captions You
are a powerful image captioner. Create detailed captions describing the contents of the given image. Include the object types and colors, counting the objects, object actions, precise object locations, texts, doublechecking relative positions between objects, etc. Instead of describing the imaginary content, only describing the content one can determine confidently from the image. Do not describe the contents by itemizing them in list form. Minimize aesthetic descriptions as much as possible.

GPT4-Vision ShareGPT4V: Improving Large Multi-Modal Models with Better Captions You
are a powerful image captioner. Create detailed captions describing the contents of the given image. Include the object types and colors, counting the objects, object actions, precise object locations, texts, doublechecking relative positions between objects, etc. Instead of describing the imaginary content, only describing the content one can determine confidently from the image. Do not describe the contents by itemizing them in list form. Minimize aesthetic descriptions as much as possible. • Cerulean blue hair • Military-inspired uniform • Naval officer's cap • Golden epaulettes • Stern gaze • Light blue eyes • White thigh-highs • Naval insignia pins • Anchored naval theme • Golden sword handle • High-waisted shorts • Blue-striped white jacket • Nautical skirt flap • Battleship turret platforms • Metallic heel decorations • Flowing hair motion • Immaculate white gloves • Pearly skin tone • Soft facial features • White naval cap • Silver belt buckle • Delicate lace trim • Floating hair ribbon • Detailed jacket buttons • Crisp ironed seams • Intricate gold braiding • Platform shoe design • Elegant standing pose • Sword sheath hanging • Glossy shoe finish • Contrasting shadows • Softly lit contours • Pastel color palette • Graceful hand placement • Ship deck floor • Transparent background • High visual fidelity • Stylized anime character • Minimalist backdrop • Polished artistic rendering <image, caption> dataset

Orca: Progressive Learning from Complex Explanation Traces of GPT-4 You
are a powerful image captioner. Create detailed captions describing the contents of the given image. Include the object types and colors, counting the objects, object actions, precise object locations, texts, doublechecking relative positions between objects, etc. Instead of describing the imaginary content, only describing the content one can determine confidently from the image. Do not describe the contents by itemizing them in list form. Minimize aesthetic descriptions as much as possible. ৈ۞ ѐ੄ system instruction

are a powerful image captioner. Create detailed captions describing the contents of the given image. Include the object types and colors, counting the objects, object actions, precise object locations, texts, doublechecking relative positions between objects, etc. Instead of describing the imaginary content, only describing the content one can determine confidently from the image. Do not describe the contents by itemizing them in list form. Minimize aesthetic descriptions as much as possible. ৈ۞ ѐ੄ system instruction ݽ؛੉ যڃ о੉٘ۄੋ ഋधҗ హਵ۽ ׹ਸ ೡ Ѫੋоܳ ੿ೣ

are a powerful image captioner. Create detailed captions describing the contents of the given image. Include the object types and colors, counting the objects, object actions, precise object locations, texts, doublechecking relative positions between objects, etc. Instead of describing the imaginary content, only describing the content one can determine confidently from the image. Do not describe the contents by itemizing them in list form. Minimize aesthetic descriptions as much as possible. ৈ۞ ѐ੄ system instruction ݽ؛੉ যڃ о੉٘ۄੋ ഋधҗ హਵ۽ ׹ਸ ೡ Ѫੋоܳ ੿ೣ Instruction Tuning <User query, LFM response> Explanation Tuning <System message, User query, LFM response> ׮নࢿ ױ҅߹ ୶ۿਸ ਬب

are a powerful image captioner. Create detailed captions describing the contents of the given image. Include the object types and colors, counting the objects, object actions, precise object locations, texts, doublechecking relative positions between objects, etc. Instead of describing the imaginary content, only describing the content one can determine confidently from the image. Do not describe the contents by itemizing them in list form. Minimize aesthetic descriptions as much as possible. ৈ۞ ѐ੄ system instruction ݽ؛੉ যڃ о੉٘ۄੋ ഋधҗ హਵ۽ ׹ਸ ೡ Ѫੋоܳ ੿ೣ

are a powerful image captioner. Create detailed captions describing the contents of the given image. Include the object types and colors, counting the objects, object actions, precise object locations, texts, doublechecking relative positions between objects, etc. Instead of describing the imaginary content, only describing the content one can determine confidently from the image. Do not describe the contents by itemizing them in list form. Minimize aesthetic descriptions as much as possible. ਗೞח ஭࣌ਸ ਬبೡ ࣻ ੓ח User instruction

are a powerful image captioner. Create detailed captions describing the contents of the given image. Include the object types and colors, counting the objects, object actions, precise object locations, texts, doublechecking relative positions between objects, etc. Instead of describing the imaginary content, only describing the content one can determine confidently from the image. Do not describe the contents by itemizing them in list form. Minimize aesthetic descriptions as much as possible. ਗೞח ஭࣌ਸ ਬبೡ ࣻ ੓ח User instruction • Cerulean blue hair • Military-inspired uniform • Naval officer's cap • Golden epaulettes • Stern gaze • Light blue eyes • White thigh-highs • Naval insignia pins • Anchored naval theme • Golden sword handle • High-waisted shorts • Blue-striped white jacket • Nautical skirt flap • Battleship turret platforms • Metallic heel decorations • Flowing hair motion • Immaculate white gloves • Pearly skin tone • Soft facial features • White naval cap

Orca: Progressive Learning from Complex Explanation Traces of GPT-4

GPT4-Vision Orca: Progressive Learning from Complex Explanation Traces of GPT-4
Model Name Prompt Cost (USD) Completion Cost (USD) Max Prompt Tokens gpt-4-vision-preview $0.00001000 $0.00003000 128000 gemini-pro-vision $0.00000020 $0.00000050 30720 Gemini-Vision : 1 : 2 1 : 4 1 : 6 =

t5IFGJOBMEBUBNJYUVSFTXFSF EFUFSNJOFEUISPVHIBCMBUJPOT POTNBMMFSNPEFMTu GPT4-Vision Gemini-Vision : 1 : 2 1
: 4 1 : 6 =

ArtBook-50M

ArtBook-50M Dataset Dataset Volume Caption Misc ArtBook-50M-2D 30M - ArtBook-50M-3D
25M - ArtBook-50M-AI 5M - This is not JourneyDB ArtBook-GPT 4M GPT-4V 12 System Instruction ArtBook-Gem 24M Gemini 12 System Instruction ArtBook-50M 50M Raw

%BUBQSFQSPDFTTJOH 0$3۽ Ӗ੗ ੓ח ࢎ૓ ઁѢ $BQUJPO UPLFOӡ੉о
੉ೞੋ ੉޷૑ ઁѢ *NBHFTJ[F BTQFDUSBUJP ೙ఠ݂ ౠ੿ ஶࣆ੉ ನೣػ ੉޷૑ ઁѢ %VQMJDBUJPO SFNPWBMVTJOH.JMWVT "FTUIFUJDTDPSF

Image Database Query ?

೟णਸ ೞӝ ੹ী ށۋ؍ Ѫ

(16 ҳೞӝ औ૑ ঋ਺

ҳ೧ب ޙઁоj

)BSEXBSF-PUUFSZ JOUIF&SBPG--.T t/PUBMMIBSEXBSFJTDSFBUFEFRVBM5IFWBSJBODFPGDMVTUFSRVBMJUZBDSPTT IBSEXBSFQSPWJEFSTJTTPIJHIUIBUJUJTMJUFSBMMZBMPUUFSZQFSUBJOJOHUPIPX NVDIQBJOPOFXPVMEIBWFUPHPUISPVHIUPUSBJOHPPENPEFMT*OTIPSU B IBSEXBSFMPUUFSZJOUIFFSBPG--.Tu IUUQTXXXZJUBZOFUCMPHUSBJOJOHHSFBUMMNTFOUJSFMZGSPNHSPVOE[FSPJOUIFXJMEFSOFTT

(16о ઁ؀۽ جইоח ࢲߡੋ૑ ഛੋ ೙ࣻ

$ nvidia-smi topo -m

$ nvbandwidth -t device_to_device_memcpy_read_ce https://github.com/NVIDIA/nvbandwidth

ӝࠄ੸ੋ )೟ण ࢿמ ѐࢶ • $6%"7FSTJPO • .JYFE1SFDJTJPO • 1Z5PSDI
UPSDIDPNQJMF • 'VMMZ4IBSEFE%BUB1BSBMMFM • ೟ण ؘ੉ఠী ݏח $VTUPN5PLFOJ[FS • 1Z5PSDI 1SPGJMFS /TJHIU$PNQVUF

5SBJOJOH4QFFE&SSPS from torch.profiler import profile, record_function, ProfilerActivity with profile( activities=[ProfilerActivity.CPU,
ProfilerActivity.CUDA], record_shapes=True ) as prof: # 문제가 있던 코드 optimizer.step() prof.export_chrome_trace("trace.json")

Correct CUDA Kernel execution for Gradient Update

"OBMZ[JOHBOE*NQSPWJOH UIF5SBJOJOH%ZOBNJDTPG%JGGVTJPO.PEFMT

оࢸ प೷ ഥҊ

7BMJEBUFUIFNPEFMQFSGPSNBODF • 7BMJEBUJPO-PTT X&." • '*% • ೟ण ؘ੉ఠ੄ EJTUSJCVUJPOҗ
ਬࢎೞա MFBLBHF ݄ӝ • $-*14DPSF • 6TJOHGPT-4 (or LLaVA)BTKVEHF

7BMJEBUFUIFNPEFMQFSGPSNBODF • 7BMJEBUJPO-PTT • '*% • ೟ण ؘ੉ఠ੄ EJTUSJCVUJPOҗ ਬࢎೞա
MFBLBHF ݄ӝ • $-*14DPSF • 6TJOHGPT-4 (or LLaVA) BTKVEHF • LLMীࢲ ݆੉ ॳ੉ח ߑध PIXART-Σ:Weak-to-Strong Training of Diﬀusion Transformer for 4K Text-to-Image GeneraMon

GPT4-Vision As an AI visual assistant, you are analyzing two
specific images. Given a specific caption, you need to judge which image aligns with the caption more closely. Please pay attention to the key information, including object identities, properties, spatial relationships, object numbers and image style, etc. The caption for the two images is: {caption} Please respond me strictly in the following format: <the first image is better> or <the second image is better> or <The two images are tied.>. The reason is <give your reason here>.

GPT4-Vision As an AI visual assistant, you are analyzing two
specific images. Given a specific caption, you need to judge which image aligns with the caption more closely. Please pay attention to the key information, including object identities, properties, spatial relationships, object numbers and image style, etc. The caption for the two images is: {caption} Please respond me strictly in the following format: <the first image is better> or <the second image is better> or <The two images are tied.>. The reason is <give your reason here>. <the first image is better>

PIXART-Σ:Weak-to-Strong Training of Diﬀusion Transformer for 4K Text-to-Image GeneraMon

"-PHPG5SBJOJOH -BSHF4DBMF%JGGVTJPO.PEFM GSPN4DSBUDI

Image Synthesis ೞח ੉ਬ

%PNBJOTQFDJGJDLOPXMFEHF %PNBJOTQFDJGJD "*

LLM vs Image Synthesis

दп੸ (MPCBM

दп੸ (MPCBM द੢ী ୽࠙ೠ ӝഥ

दп੸ (MPCBM द੢ী ୽࠙ೠ ӝഥ ҙब੉ ੸Ҋ

दп੸ (MPCBM द੢ী ୽࠙ೠ ӝഥ ҙब੉ ੸Ҋ ҙ۲ ҃೷੉ ࠗ੤

द೐౟স "*-BCT

회사 소개 : 개 요 회사개요 기업명 대표 설립일 게임
임직원 홈페이지 사업영역 소재지 시프트업 김형태 2013년 12월 2일 데스티니차일드 / 승리의 여신 : 니케 스텔라 블레이드 290여명 shiftup.co.kr 모바일게임 개발 및 서비스 / 콘솔 게임 개 발 서울시 강남구 서초동

4)*'561"*-BCT ௏झೖ ࢚੢दز Ѥ द೐౟সjӝসо஖ ઑਗী ॕܻח ׀ 창업 10년
만에 코스피 상장을 추진하는 게임사 시프트업이 ‘제2의 크래프톤’이 될 수 있을지 업계의 관심이 쏠리고 있다. 시프트업 '니케', 누적 매출 1조원 돌파… 시프트업의 서브컬처 게임 ‘승리의 여신:니케’가 누적 매출 1조 원을 넘겼다. 2022년 11월 출시된 지 약 1년 4개월 만이다.

프로젝트 : 스텔라 블레이드 미국 경제지 포브스 “플레이스테이션 쇼케이스 작품
중 매우 놀라운 신작”으로 소개 회사소개 프로젝트 게임산업 시프트업채용 소니의 적극적인 구애로 국내 게임사 최초로 PS5 독점 3N도 도전하지 못했던 트리플 A급 액션 게임 2024 플레이스테이션 최고의 기대작 1위 출시일 발표 후 예약 구매 1위

시프트업의 성장 전략 NEXT <승리의 여신 : 니케> 글로벌 점유율
확대 + 국내 최초의 AAA급 콘솔 IP <스텔라 블레이드> 개발 + 끊임 없는 신규 IP 제작 Project 글로벌 장기 흥행작과 AAA게임 IP를 보유한 게 임사 회사소개 프로젝트 게임산업 시프트업채용

௝੸ೠ ъթ য়ೖझ

द೐౟স "*-BCT

ࢤࢿ "*۽ ѱ੐ ઁ੘ ࠺ਊਸ ߓ ઴੉ח ઑ૒

5SBJOJOH Large Image Synthesis Model Large Multimodal Model Video
Synthesis Model Audio Synthesis

જ਷ ֤ޙ ॳӝо ݾ಴ ࢑೟ োҳ ӝഥ KAIST 인공지능 대학원
KAIST에서 전액 출자하여 기업에서 필요한 연구과제를 수행하기 위한 CCC회원사 전용 프로그램 포항공과대학교 산학협력단 OpenAI에서의 GPT-3, 강화 학습, BlockSparse 연구 경험

੷൞৬ ೣԋ ࢿ੢ೞҊ र׮ݶ ⊙ bit.ly/shiftup-ai

➡ https://bit.ly/ai-compiler-study

LLM에서 배우는 이미지 생성 모델 ZERO부터 학습하기

LLM에서 배우는 이미지 생성 모델 ZERO부터 학습하기

More Decks by carpedm20

Other Decks in Programming

Featured

Transcript