Efficient Fine-Tuning for Llama-v2-7b on a Single GPU

DeepLearningAI · Intermediate ·🧠 Large Language Models ·2y ago

The first problem you’re likely to encounter when fine-tuning an LLM is the “host out of memory” error. It’s more difficult for fine-tuning the 7B parameter Llama-2 model which requires more memory. In this talk, we are having Piero Molino and Travis Addair from the open-source Ludwig project to show you how to tackle this problem. The good news is that, with an optimized LLM training framework like Ludwig.ai, you can get the host memory overhead back down to a more reasonable host memory even when training on multiple GPUs. In this hands-on workshop, we‘ll discuss the unique challenges in…

Watch on YouTube ↗ (saves to browser)

Playlist

Uploads from DeepLearningAI · DeepLearningAI · 0 of 60

← Previous Next →

Forward and Backward Propagation (C1W4L06)

Forward and Backward Propagation (C1W4L06)

deeplearning.ai's Heroes of Deep Learning: Yuanqing Lin

deeplearning.ai's Heroes of Deep Learning: Yuanqing Lin

deeplearning.ai's Heroes of Deep Learning: Ruslan Salakhutdinov

deeplearning.ai's Heroes of Deep Learning: Ruslan Salakhutdinov

deeplearning.ai's Heroes of Deep Learning: Yoshua Bengio

deeplearning.ai's Heroes of Deep Learning: Yoshua Bengio

deeplearning.ai's Heroes of Deep Learning: Pieter Abbeel

deeplearning.ai's Heroes of Deep Learning: Pieter Abbeel

deeplearning.ai's Heroes of Deep Learning: Ian Goodfellow

deeplearning.ai's Heroes of Deep Learning: Ian Goodfellow

deeplearning.ai's Heroes of Deep Learning: Andrej Karpathy

deeplearning.ai's Heroes of Deep Learning: Andrej Karpathy

Using an Appropriate Scale (C2W3L02)

Using an Appropriate Scale (C2W3L02)

Gradient Checking (C2W1L13)

Gradient Checking (C2W1L13)

Gradient Checking Implementation Notes (C2W1L14)

Gradient Checking Implementation Notes (C2W1L14)

Learning Rate Decay (C2W2L09)

Learning Rate Decay (C2W2L09)

Understanding Mini-Batch Gradient Dexcent (C2W2L02)

Understanding Mini-Batch Gradient Dexcent (C2W2L02)

Mini Batch Gradient Descent (C2W2L01)

Mini Batch Gradient Descent (C2W2L01)

The Problem of Local Optima (C2W3L10)

The Problem of Local Optima (C2W3L10)

Exponentially Weighted Averages (C2W2L03)

Exponentially Weighted Averages (C2W2L03)

Tuning Process (C2W3L01)

Tuning Process (C2W3L01)

Understanding Exponentially Weighted Averages (C2W2L04)

Understanding Exponentially Weighted Averages (C2W2L04)

Bias Correction of Exponentially Weighted Averages (C2W2L05)

Bias Correction of Exponentially Weighted Averages (C2W2L05)

Gradient Descent With Momentum (C2W2L06)

Gradient Descent With Momentum (C2W2L06)

Normalizing Activations in a Network (C2W3L04)

Normalizing Activations in a Network (C2W3L04)

Hyperparameter Tuning in Practice (C2W3L03)

Hyperparameter Tuning in Practice (C2W3L03)

Adam Optimization Algorithm (C2W2L08)

Adam Optimization Algorithm (C2W2L08)

RMSProp (C2W2L07)

RMSProp (C2W2L07)

Fitting Batch Norm Into Neural Networks (C2W3L05)

Fitting Batch Norm Into Neural Networks (C2W3L05)

Why Does Batch Norm Work? (C2W3L06)

Why Does Batch Norm Work? (C2W3L06)

Batch Norm At Test Time (C2W3L07)

Batch Norm At Test Time (C2W3L07)

Softmax Regression (C2W3L08)

Softmax Regression (C2W3L08)

Deep Learning Frameworks (C2W3L10)

Deep Learning Frameworks (C2W3L10)

Neural Network Overview (C1W3L01)

Neural Network Overview (C1W3L01)

Training Softmax Classifier (C2W3L09)

Training Softmax Classifier (C2W3L09)

Why Deep Representations? (C1W4L04)

Why Deep Representations? (C1W4L04)

Gradient Descent For Neural Networks (C1W3L09)

Gradient Descent For Neural Networks (C1W3L09)

Neural Network Representations (C1W3L02)

Neural Network Representations (C1W3L02)

TensorFlow (C2W3L11)

TensorFlow (C2W3L11)

Activation Functions (C1W3L06)

Activation Functions (C1W3L06)

Explanation For Vectorized Implementation (C1W3L05)

Explanation For Vectorized Implementation (C1W3L05)

Getting Matrix Dimensions Right (C1W4L03)

Getting Matrix Dimensions Right (C1W4L03)

Understanding Dropout (C2W1L07)

Understanding Dropout (C2W1L07)

Building Blocks of a Deep Neural Network (C1W4L05)

Building Blocks of a Deep Neural Network (C1W4L05)

Why Non-linear Activation Functions (C1W3L07)

Why Non-linear Activation Functions (C1W3L07)

Computing Neural Network Output (C1W3L03)

Computing Neural Network Output (C1W3L03)

Backpropagation Intuition (C1W3L10)

Backpropagation Intuition (C1W3L10)

Train/Dev/Test Sets (C2W1L01)

Train/Dev/Test Sets (C2W1L01)

Deep L-Layer Neural Network (C1W4L01)

Deep L-Layer Neural Network (C1W4L01)

Random Initialization (C1W3L11)

Random Initialization (C1W3L11)

Other Regularization Methods (C2W1L08)

Other Regularization Methods (C2W1L08)

Normalizing Inputs (C2W1L09)

Normalizing Inputs (C2W1L09)

Derivatives Of Activation Functions (C1W3L08)

Derivatives Of Activation Functions (C1W3L08)

Parameters vs Hyperparameters (C1W4L07)

Parameters vs Hyperparameters (C1W4L07)

Vectorizing Across Multiple Examples (C1W3L04)

Vectorizing Across Multiple Examples (C1W3L04)

What does this have to do with the brain? (C1W4L08)

What does this have to do with the brain? (C1W4L08)

Dropout Regularization (C2W1L06)

Dropout Regularization (C2W1L06)

Vanishing/Exploding Gradients (C2W1L10)

Vanishing/Exploding Gradients (C2W1L10)

Basic Recipe for Machine Learning (C2W1L03)

Basic Recipe for Machine Learning (C2W1L03)

Bias/Variance (C2W1L02)

Bias/Variance (C2W1L02)

Forward Propagation in a Deep Network (C1W4L02)

Forward Propagation in a Deep Network (C1W4L02)

Weight Initialization in a Deep Network (C2W1L11)

Weight Initialization in a Deep Network (C2W1L11)

Numerical Approximations of Gradients (C2W1L12)

Numerical Approximations of Gradients (C2W1L12)

Regularization (C2W1L04)

Regularization (C2W1L04)

Why Regularization Reduces Overfitting (C2W1L05)

Why Regularization Reduces Overfitting (C2W1L05)

5 Levels of AI Agents - From Simple LLM Calls to Multi-Agent Systems

5 Levels of AI Agents - From Simple LLM Calls to Multi-Agent Systems

Dave Ebbelaar (LLM Eng)