TIL
  • Main
  • TIL : ML
    • AI News
      • AI-powered coding, free of charge with Colab
    • Paper
      • Retrieval-augmented generation for knowledge-intensive nlp tasks
        • reference ์ •๋ฆฌ
      • Sequence to Sequence
    • Generative AI
      • LLM
        • ๊ธฐ๋ณธ ๊ฐœ๋…
    • Boostcamp
      • Daily Report
        • 2021
          • 09
            • week4
            • week5
          • 10
            • week1
            • week2
            • week3
            • week4
          • 11
            • week1
            • week2
            • week3
            • week4
            • week5
          • 12
            • week1
            • week2
            • week3
      • Final Project
        • ํ”„๋กœ์ ํŠธ ๊ธฐํš
          • ๐ŸŒŸ์ตœ์ข… ํ”„๋กœ์ ํŠธ ๊ธฐํš
          • ์ตœ์ข… ํ”„๋กœ์ ํŠธ Version
          • ์•„์ด๋””์–ด ์ˆ˜์ง‘
          • ์š•์„ค, ํ˜์˜ค๋ฐœ์–ธ ๊ฐ์ง€
          • ๋ผ์ด๋ธŒ ์ปค๋จธ์Šค ๋ ˆํฌํŒ… ํ”„๋กœ์ ํŠธ
        • ํ”„๋กœ์ ํŠธ ์ง„ํ–‰
          • week1
          • week2
          • week3
      • Competition
        • 1. [NLP] ๋ฌธ์žฅ ๋‚ด ๊ฐœ์ฒด๊ฐ„ ๊ด€๊ณ„ ์ถ”์ถœ
          • Day1 (9.27, ์›”)
          • Day2-3 (9.28~29, ํ™”~์ˆ˜)
          • Day4 (9.30, ๋ชฉ)
          • Day5 (10.1, ๊ธˆ)
          • Day6~7 (10.2~3, ํ† ~์ผ)
          • Day8 (10.4, ์›”)
          • Day9 (10.5, ํ™”)
          • Day10 (10.6, ์ˆ˜)
          • Day 11 (10.7 ๋ชฉ)
        • 2. [NLP] MRC ํ”„๋กœ์ ํŠธ
          • Day1 (10.25, ์›”)
          • Day2 (10.26, ํ™”)
          • Day3 (10.27, ์ˆ˜)
          • Day4-5 (10.28-29, ๋ชฉ-๊ธˆ)
          • Day6 (11.1, ์›”)
          • Day7 (11.2, ํ™”)
          • Day8 (11.3, ์ˆ˜)
          • Day9 (11.4, ๋ชฉ)
        • ๐Ÿ”จ3. [NLP] ๋ฐ์ดํ„ฐ ์ œ์ž‘
          • Day1
        • ๐Ÿ”จ4. [๊ณตํ†ต] ๋ชจ๋ธ ๊ฒฝ๋Ÿ‰ํ™”
      • [U Stage] - DL basic
        • (01๊ฐ•) ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ๋ณธ ์šฉ์–ด ์„ค๋ช… - Historical Review
        • (02๊ฐ•) ๋‰ด๋Ÿด ๋„คํŠธ์›Œํฌ - MLP (Multi-Layer Perceptron)
        • (03๊ฐ•) Optimization
        • ๐Ÿ”จ(04๊ฐ•) Convolution์€ ๋ฌด์—‡์ธ๊ฐ€?
        • ๐Ÿ”จ(05๊ฐ•) Modern CNN - 1x1 convolution์˜ ์ค‘์š”์„ฑ
        • ๐Ÿ”จ(06๊ฐ•) Computer Vision Applications
        • (07๊ฐ•) Sequential Models - RNN
        • (08๊ฐ•) Sequential Models - Transformer
        • Page 2
      • [U Stage] - PyTorch
        • (01๊ฐ•) Introduction to PyTorch
        • (02๊ฐ•) PyTorch Basics
        • (03๊ฐ•) PyTorch ํ”„๋กœ์ ํŠธ ๊ตฌ์กฐ ์ดํ•ดํ•˜๊ธฐ
        • (04๊ฐ•) AutoGrad & Optimizer
        • (05๊ฐ•) Dataset & Dataloader
        • (06๊ฐ•) ๋ชจ๋ธ ๋ถˆ๋Ÿฌ์˜ค๊ธฐ
        • (07๊ฐ•) Monitoring tools for PyTorch
        • (08๊ฐ•) Multi-GPU ํ•™์Šต
        • (09๊ฐ•) Hyperparameter Tuning
        • (10๊ฐ•) PyTorch Troubleshooting
      • [U Stage] - NLP
        • (01๊ฐ•) Introduction to NLP, Bag-of-Words
        • (02๊ฐ•) Word Embedding
        • (03๊ฐ•) Recurrent Neural Network and Language Modeling
        • (04๊ฐ•) LSTM and GRU
        • (05๊ฐ•) Sequence to Sequence with Attention
        • (06๊ฐ•) Beam Search and BLEU score
        • (07-08๊ฐ•) Transformer
        • (09๊ฐ•) Self-supervised Pre-training Models
      • [P Stage] - KLUE
        • (1๊ฐ•) ์ธ๊ณต์ง€๋Šฅ๊ณผ ์ž์—ฐ์–ด ์ฒ˜๋ฆฌ
        • (2๊ฐ•) ์ž์—ฐ์–ด์˜ ์ „์ฒ˜๋ฆฌ
        • (3๊ฐ•) BERT ์–ธ์–ด๋ชจ๋ธ ์†Œ๊ฐœ
        • (4๊ฐ•) ํ•œ๊ตญ์–ด BERT ์–ธ์–ด ๋ชจ๋ธ ํ•™์Šต
        • (5๊ฐ•) BERT ๊ธฐ๋ฐ˜ ๋‹จ์ผ ๋ฌธ์žฅ ๋ถ„๋ฅ˜ ๋ชจ๋ธ ํ•™์Šต
        • (6๊ฐ•) BERT ๊ธฐ๋ฐ˜ ๋‘ ๋ฌธ์žฅ ๊ด€๊ณ„ ๋ถ„๋ฅ˜ ๋ชจ๋ธ ํ•™์Šต
        • (7๊ฐ•) BERT ์–ธ์–ด๋ชจ๋ธ ๊ธฐ๋ฐ˜์˜ ๋ฌธ์žฅ ํ† ํฐ ๋ถ„๋ฅ˜
        • ์˜คํ”ผ์Šค์•„์›Œ (9.30, ๋ชฉ)
        • (8๊ฐ•) GPT ์–ธ์–ด ๋ชจ๋ธ
        • (9๊ฐ•) GPT ์–ธ์–ด๋ชจ๋ธ ๊ธฐ๋ฐ˜์˜ ์ž์—ฐ์–ด ์ƒ์„ฑ
        • (10๊ฐ•) ์ตœ์‹  ์ž์—ฐ์–ด์ฒ˜๋ฆฌ ์—ฐ๊ตฌ
      • [P Stage] - MRC
        • Before Study
        • (1๊ฐ•) MRC Intro & Python Basics
        • (2๊ฐ•) Extraction-based MRC
        • (3๊ฐ•) Generation-based MRC
        • (4๊ฐ•) Passage Retrieval - Sparse Embedding
        • (5๊ฐ•) Passage Retrieval - Dense Embedding
        • ์˜คํ”ผ์Šค์•„์›Œ
        • (6๊ฐ•) Scaling up with FAISS
        • (7๊ฐ•) Linking MRC and Retrieval
        • (8๊ฐ•) Reducing Training Bias
        • (9๊ฐ•) Closed-book QA with T5
        • (10๊ฐ•) QA with Phrase Retrieval
        • ๋งˆ์Šคํ„ฐํด๋ž˜์Šค
      • [P Stage] - ๋ฐ์ดํ„ฐ์ œ์ž‘(NLP)
        • (1๊ฐ•) ๋ฐ์ดํ„ฐ ์ œ์ž‘์˜ A to Z
        • (2๊ฐ•) ์ž์—ฐ์–ด์ฒ˜๋ฆฌ ๋ฐ์ดํ„ฐ ๊ธฐ์ดˆ
        • (3๊ฐ•) ์ž์—ฐ์–ด์ฒ˜๋ฆฌ ๋ฐ์ดํ„ฐ ์†Œ๊ฐœ 1
        • (4๊ฐ•) ์ž์—ฐ์–ด์ฒ˜๋ฆฌ ๋ฐ์ดํ„ฐ ์†Œ๊ฐœ 2
        • (5๊ฐ•) ์›์‹œ ๋ฐ์ดํ„ฐ์˜ ์ˆ˜์ง‘๊ณผ ๊ฐ€๊ณต
        • ์˜คํ”ผ์Šค์•„์›Œ (11.10, ์ˆ˜)
        • (6๊ฐ•) ๋ฐ์ดํ„ฐ ๊ตฌ์ถ• ์ž‘์—… ์„ค๊ณ„
        • (7๊ฐ•) ๋ฐ์ดํ„ฐ ๊ตฌ์ถ• ๊ฐ€์ด๋“œ๋ผ์ธ ์ž‘์„ฑ ๊ธฐ์ดˆ
        • (8๊ฐ•) ๊ด€๊ณ„ ์ถ”์ถœ ๊ณผ์ œ์˜ ์ดํ•ด
        • (9๊ฐ•) ๊ด€๊ณ„ ์ถ”์ถœ ๊ด€๋ จ ๋…ผ๋ฌธ ์ฝ๊ธฐ
        • (10๊ฐ•) ๊ด€๊ณ„ ์ถ”์ถœ ๋ฐ์ดํ„ฐ ๊ตฌ์ถ• ์‹ค์Šต
      • [P Stage] - ๋ชจ๋ธ ์ตœ์ ํ™”
        • (1๊ฐ•) ์ตœ์ ํ™” ์†Œ๊ฐœ ๋ฐ ๊ฐ•์˜ ๊ฐœ์š”
        • (2๊ฐ•) ๋Œ€ํšŒ ๋ฐ ๋ฐ์ดํ„ฐ์…‹ ์†Œ๊ฐœ
        • (3๊ฐ•) ์ž‘์€ ๋ชจ๋ธ, ์ข‹์€ ํŒŒ๋ผ๋ฏธํ„ฐ ์ฐพ๊ธฐ: AutoML ์ด๋ก 
        • ๐Ÿ”จ(4๊ฐ•) ์ž‘์€ ๋ชจ๋ธ, ์ข‹์€ ํŒŒ๋ผ๋ฏธํ„ฐ ์ฐพ๊ธฐ: AutoML ์‹ค์Šต
        • (5๊ฐ•) ์ž‘์€ ๋ชจ๋ธ, ์ข‹์€ ํŒŒ๋ผ๋ฏธํ„ฐ ์ฐพ๊ธฐ: Data Augmentation & AutoML ๊ฒฐ๊ณผ ๋ถ„์„
        • ๐Ÿ”จ์˜คํ”ผ์Šค์•„์›Œ -Baseline ์ฝ”๋“œ์— ๋ชจ๋“ˆ ์ž‘์„ฑํ•˜๊ธฐ(์‹ ์ข…์„  ๋ฉ˜ํ† ๋‹˜)
      • [P Stage] - Product Serving
        • Part 1: Product Serving ๊ฐœ๋ก 
          • 1.1 ๊ฐ•์˜ ์ง„ํ–‰ ๋ฐฉ์‹
          • 1.2 MLOps ๊ฐœ๋ก 
          • 1.3 Model Serving
          • 1.4 ๋จธ์‹ ๋Ÿฌ๋‹ ํ”„๋กœ์ ํŠธ ๋ผ์ดํ”„ ์‚ฌ์ดํด
        • Part 2: ํ”„๋กœํ† ํƒ€์ž…๋ถ€ํ„ฐ ์ ์ง„์ ์œผ๋กœ ๊ฐœ์„ ํ•˜๊ธฐ
          • 2.1 ํ”„๋กœํ† ํƒ€์ดํ•‘ - Notebook ๋ฒ ์ด์Šค(Voila)
          • 2.2 ํ”„๋กœํ† ํƒ€์ดํ•‘ - ์›น ์„œ๋น„์Šค ํ˜•ํƒœ(Streamlit)
          • 2.3 Linux & Shell Command
          • 2.4 Cloud
          • 2.5 Github Action์„ ํ™œ์šฉํ•œ CI/CD
        • Part 3: ๋” ์™„์„ฑํ™”๋œ ์ œํ’ˆ์œผ๋กœ
          • 3.1.1 FastAPI
          • 3.1.2 Fast API
          • 3.1.3 Fast API
          • 3.2 Docker
          • 3.3 Logging
          • 3.4 MLFlow
        • Part 4: ์‹ฌํ™” ์†Œ์žฌ
          • 4.1 BentoML
          • 4.2 Airflow
          • 4.3 ๋จธ์‹ ๋Ÿฌ๋‹ ๋””์ž์ธ ํŒจํ„ด
          • 4.4 ์•ž์œผ๋กœ ๋” ๊ณต๋ถ€ํ•˜๋ฉด ์ข‹์„ ๋‚ด์šฉ
      • ํŠน๊ฐ•
        • (ํŠน๊ฐ•) ๊น€์ƒํ›ˆ - ์บ๊ธ€ ๊ทธ๋žœ๋“œ๋งˆ์Šคํ„ฐ์˜ ๋…ธํ•˜์šฐ ๋Œ€๋ฐฉ์ถœ
        • (ํŠน๊ฐ•) ์ดํ™œ์„ - ์„œ๋น„์Šค ํ–ฅ AI ๋ชจ๋ธ ๊ฐœ๋ฐœํ•˜๊ธฐ
        • (ํŠน๊ฐ•) ๊ตฌ์ข…๋งŒ - AI + ML๊ณผ Quant Trading
        • (ํŠน๊ฐ•) ๋ฌธ์ง€ํ˜• - ๋‚ด๊ฐ€ ๋งŒ๋“  AI ๋ชจ๋ธ์€ ํ•ฉ๋ฒ•์ผ๊นŒ, ๋ถˆ๋ฒ•์ผ๊นŒ
        • (ํŠน๊ฐ•) ์ด์ค€์—ฝ - Full Stack ML Engineer
        • (ํŠน๊ฐ•) ๋ฐ•์€์ • - AI ์‹œ๋Œ€์˜ ์ปค๋ฆฌ์–ด ๋นŒ๋”ฉ
        • (ํŠน๊ฐ•) ์˜คํ˜œ์—ฐ - AI Ethics
    • Competition
      • (DACON)ํ•œ๊ตญ์–ด ๋ฌธ์žฅ ๊ด€๊ณ„ ๋ถ„๋ฅ˜ ๊ฒฝ์ง„๋Œ€ํšŒ
        • Day1(2.14, ์›”)
        • Day2(2.15, ํ™”)
        • Day3(2.16, ์ˆ˜)
        • Day4(2.17, ๋ชฉ)
      • 2021 ์ธ๊ณต์ง€๋Šฅ ๋ฐ์ดํ„ฐ ํ™œ์šฉ ๊ฒฝ์ง„๋Œ€ํšŒ
        • ์—ญ๋Ÿ‰ํ‰๊ฐ€
          • Day1 (9.28, ํ™”)
          • Day2 (9.29, ์ˆ˜)
          • Day3 (9.30, ๋ชฉ)
        • ์˜ˆ์„ 
          • Data ๋ถ„์„
          • NSML
          • What We Have Done?
    • ETC
      • ์ธํ„ฐ๋ทฐ ๋Œ€๋น„
        • Computer Science
        • ML/DL
      • Poetry๋กœ dependency ๊ด€๋ฆฌ
        • window์—์„œ ์„ค์น˜ํ•˜๊ธฐ
      • code block
      • ๊ณต๋ถ€ํ•  ๊ฒƒ ์ž„์‹œ๋ณด๊ด€
      • Transformer to T5
      • Hugging Face Tutorial
        • Ch1. Transformer models
        • Ch2. Using Transformers
        • Ch3. Fine-tuning a model with the Trainer API
      • KLUE
      • Pandas
  • TIL : Ops
    • AWS
      • SageMaker
  • TIL : Computer (CS)
    • Error
      • TextEncodeInput must be Union[TextInputSequence, Tuple[InputSequence, InputSequence]]
    • Algorithm
      • Theory
      • Programmers
        • ๊ธฐ๋Šฅ๊ฐœ๋ฐœ
    • ETC
      • Github ์ปค๋ฐ‹ ํžˆ์Šคํ† ๋ฆฌ ์‚ญ์ œ
Powered by GitBook
On this page
  • 1. Fire Module from Squeezenet
  • 1.1. Squeezenet
  • 1.2. Fire module
  • 2. Baseline ์ฝ”๋“œ ์ดํ•ดํ•˜๊ธฐ
  • 2.1. Model Parser ํ๋ฆ„ ์ดํ•ดํ•˜๊ธฐ
  • 3. Custom module ์ž‘์„ฑํ•˜๊ธฐ

Was this helpful?

  1. TIL : ML
  2. Boostcamp
  3. [P Stage] - ๋ชจ๋ธ ์ตœ์ ํ™”

๐Ÿ”จ์˜คํ”ผ์Šค์•„์›Œ -Baseline ์ฝ”๋“œ์— ๋ชจ๋“ˆ ์ž‘์„ฑํ•˜๊ธฐ(์‹ ์ข…์„  ๋ฉ˜ํ† ๋‹˜)

Previous(5๊ฐ•) ์ž‘์€ ๋ชจ๋ธ, ์ข‹์€ ํŒŒ๋ผ๋ฏธํ„ฐ ์ฐพ๊ธฐ: Data Augmentation & AutoML ๊ฒฐ๊ณผ ๋ถ„์„Next[P Stage] - Product Serving

Last updated 3 years ago

Was this helpful?

1. Fire Module from Squeezenet

1.1. Squeezenet

  • Alexnet ์ˆ˜์ค€์˜ ์ •ํ™•๋„

  • 50๋ฐฐ ์ž‘์€ Parameters, 0.5Mb ๋ฏธ๋งŒ์˜ ํฌ๊ธฐ

  • ์ž‘์€ CNN์˜ ์ด์ 

    • ๋ถ„์‚ฐํ•™์Šต์‹œ ๋ชจ๋ธ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ ์ˆ˜์™€ ์„œ๋ฒ„๊ฐ„ ํ†ต์‹  overhead๊ฐ€ ๋น„๋ก€ํ•˜๋ฏ€๋กœ, ๋น ๋ฅด๊ฒŒ ๋ถ„์‚ฐํ•™์Šต๊ฐ€๋Šฅ

    • ํ•™์Šต ์™„๋ฃŒ๋œ ์„œ๋ฒ„์—์„œ ์ž์œจ์ฃผํ–‰์ฐจ์™€ ๊ฐ™์€ ๋‹ค๋ฅธ client๋กœ ๋ชจ๋ธ์„ ์ „์†กํ•  ๋•Œ overhead๊ฐ์†Œ

    • ์ œํ•œ๋œ ๋ฉ”๋ชจ๋ฆฌ์—๋„ ์ž„๋ฒ ๋”ฉ์ด ๊ฐ€๋Šฅํ•˜๋‹ค.

์ „์ฒด ๊ตฌ์กฐ

1.2. Fire module

2. Baseline ์ฝ”๋“œ ์ดํ•ดํ•˜๊ธฐ

2.1. Model Parser ํ๋ฆ„ ์ดํ•ดํ•˜๊ธฐ

3. Custom module ์ž‘์„ฑํ•˜๊ธฐ