A Comprehensive Guide to LLM Training: Overview of Different Methods to Train an LLM

February 16, 2024


With the continuous evolution of large language models (LLMs), researchers are exploring and pushing the boundaries of what these models can achieve. The demand for enhanced performance has increased with each new architecture and expanded parameter set, which is creating the necessity for advancement in LLM training methods.

The size and complexity of modern LLMs demand significant computational resources. This is where cloud-based GPUs are required. These cloud GPUs can execute multiple complex mathematical operations in parallel, which makes them very fast and efficient.

But apart from computational resources, latency and accuracy emerge as important factors affecting the performance of LLMs. The demand for fast, accurate model responses has increased, which motivates researchers to explore new and different training methods for LLMs.

Let’s dive deeper into this subject.

Overview of LLM Training Methods

There are three main methods to train large language models. Let’s look at them one by one.

  1. Unsupervised Pre-training: This is the initial phase of training which focuses on exposing the model to a vast amount of text data. The input quantity is high, but the quality of the response is low. In this phase, the model is trained to predict the next probable token for a trillion sequence of texts. The output of this phase generally gives the foundation or base model.
  2. Supervised Fine-Tuning: After the unsupervised pre-training, the model goes under supervised finetuning which is aimed at tailoring its capabilities to specific tasks or domains. In this phase, the model is provided with a smaller but high-quality dataset which consists of pairs of prompts and corresponding responses. By training on this dataset, the model learns to generate responses in line with the given prompts, which effectively transforms the base model into a specialized tool. This step is often referred to as ‘instruction tuning’, as it involves the fine-tuning of the model based on the explicit instructions provided by the training data.
  3. Reinforcement Learning from Human Feedback (RLHF):  This is the final phase which introduces reinforcement learning techniques. It leverages human feedback to further refine the model’s performance. This process unfolds in two stages:
  1. Training a Reward Model: Initially, a reward model is trained to serve as a scoring function that assesses the quality of the generated responses. Human labelers evaluate responses and provide feedback, which is used to train the reward model.
  1. Optimizing the LLM: After the reward model training, the LLM undergoes iterative optimization to generate responses that elicit high scores from the reward model. This optimization process involves adjusting the model's parameters to enhance the quality of the generated text while it adheres to constraints such as maintaining consistency and avoiding deterioration in performance. Essentially, the goal is to learn an optimal strategy or policy, for generating text that aligns with human preferences and constraints.

This final phase requires careful orchestration; it involves balancing the pursuit of higher-quality outputs with the need to maintain coherence and effectiveness in text completion. Through a combination of sophisticated algorithms and human-guided feedback, the LLM evolves into a powerful tool that is capable of producing contextually relevant and coherent text across various applications and domains.

Accelerating with E2E Cloud GPUs

In the context of the outlined LLM training process, cloud GPUs play an important role in accelerating the computational tasks involved in each stage. During the unsupervised pre-training phase, the LLM is exposed to massive amounts of data, and it requires extensive computational resources to process and analyze this data efficiently. Cloud GPUs excel in parallel computing, which enables the model to train on vast datasets in a fraction of the time it would take with traditional CPUs. The parallel processing capabilities of GPUs allow for simultaneous computation of multiple operations, which significantly speeds up the training process and reduces time to completion.

In the supervised fine-tuning stage, the model is further refined based on specific tasks or domains, which necessitates iterative training on high-quality datasets. Cloud GPUs enable fast experimentation and iteration by providing scalable computing resources on demand. Researchers and developers can quickly deploy GPU instances in the cloud, which allows them to train and evaluate different versions of the model in parallel, by accelerating the optimization process and facilitating faster convergence to optimal performance.

In the RLHF phase, the iterative optimization process relies heavily on computational resources to train the reward model and optimize the LLM based on human feedback. Cloud GPUs enable researchers and developers to scale up the training of complex reinforcement learning algorithms, such as policy gradient methods, which require extensive computational power for training neural networks. By harnessing the parallel processing capabilities of cloud GPUs, researchers and developers can accelerate the training of the reward model and expedite the optimization of the LLM, which leads to more efficient learning and improved performance.

E2E Networks provides a variety of advanced cloud GPUs that can accelerate the LLM training process. For more information about the products of E2E Networks, visit here

To get started with E2E Cloud GPUs, login into your E2E account. Set up your SSH keys by visiting Settings.

 After creating the SSH keys, visit Compute to create a node instance.

Open your Visual Studio code, and download the extension Remote Explorer and Remote SSH. Open a new terminal. Login into your local system with the following code:

ssh root@

With this, you’ll be logged in to your node. 

To see the performance of a model with different training methods, here we are using the Llama 2 model.

A Brief Introduction to Llama 2

Llama 2 represents a significant advancement in large language models; it was introduced by Meta AI in 2023 to democratize access to powerful AI capabilities for both research and commercial endeavors. Llama 2 is offered free of charge and is designed to excel in various natural language processing tasks, from text generation to programming code comprehension. 

Unlike its predecessor, Llama 1, which was initially accessible only to select research institutions under specific licensing conditions, Llama 2 is available to any organization with fewer than 700 million active users. Llama 2 adopts a strategy of prioritizing model performance enhancement over sheer parameter count, which enables smaller organizations and research communities to leverage its capabilities without exorbitant computational costs by featuring models with varying parameters.

Llama 2 comprises a series of transformer-based autoregressive causal language models. These models operate by taking a sequence of words as input and recursively predicting the next word. During self-supervised pre-training, LLMs are fed the beginnings of sample sentences drawn from an extensive corpus of unlabeled data. Their task is to predict the subsequent word, training the model to minimize the discrepancy between the actual next word and its own predictions. Base foundation models are fundamentally not pre-trained to directly respond to prompts; instead, they append text to prompts in a grammatically coherent manner. Further refinement can be achieved through techniques like supervised learning and reinforcement learning, which is necessary to tailor a foundation model for specific applications such as dialogue systems, instruction following, or creative writing. Llama 2 models serve as a foundational framework for building purpose-specific models. 

Unsupervised Pre-training with Base Model Llama 2

Let’s see how a foundation model under unsupervised pre-training goes with the Llama 2 model.

To get started, load the model.

from transformers import (
from peft import LoraConfig
from trl import SFTTrainer

model = AutoModelForCausalLM.from_pretrained("NousResearch/Llama-2-7b-chat-hf", device_map="auto")

There is one flag while loading the model, i.e., device_map. It ensures that the model is moved to your GPU.

Then, we will load the tokenizer with the same model that we loaded before.

import torch
tokenizer = AutoTokenizer.from_pretrained("NousResearch/Llama-2-7b-chat-hf", padding_side="left")
Tokenizer.pad_token = tokenizer.eos_token

Now, we’ll preprocess the text input with the tokenizer. The model_inputs variable holds the tokenized text input as well as the attention masks. After tokenizing the inputs, the generate method will return the generated tokens.

model_inputs = tokenizer(["India is", "The future of AI is"], return_tensors="pt", padding=True).to("cuda")
generated_ids = model.generate(**model_inputs,max_new_tokens=50)
tokenizer.batch_decode(generated_ids, skip_special_tokens=True)

The following will be the result:

['India is a country with a diverse culture and a long history. It is home to many different ethnic groups, each with their own unique traditions and customs. In this essay, we will explore some of the main ethnic groups in India and',
 'The future of AI is exciting and uncertain\n\nThe future of AI is exciting and uncertain. On one hand, AI has the potential to revolutionize numerous industries and aspects of society, from healthcare and education to transportation and entertainment. On']

As the maximum number of new tokens is 50, it has generated text according to that.

Supervised Fine-Tuning with Llama 2

As we noticed, the unsupervised pre-trained model is performing well, but to train the particular model on a specific dataset, so that it could generate responses accordingly, we will move to supervised fine-tuning.

To get started, we will first load the dataset. I have picked a Llama dataset from Hugging Face, which can be found here

from datasets import load_dataset
dataset = load_dataset("mlabonne/guanaco-llama2-1k", split="train")

Now, we will use 4-bit quantization and set the quantization configuration to the model to improve the speed.

bnb_4bit_compute_dtype = "float16"
compute_dtype = getattr(torch, bnb_4bit_compute_dtype)

bnb_config = BitsAndBytesConfig(

Now, load the model.

model = AutoModelForCausalLM.from_pretrained(
    device_map={"": 0},
model.config.use_cache = False

Load the tokenizer and set the padding side to the right to overcome the issues with FP16.

tokenizer = AutoTokenizer.from_pretrained("NousResearch/Llama-2-7b-chat-hf", trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = "right"

Now, define the PEFT parameters. We’ll be using LoRA and setting the configurations according to that.

peft_params = LoraConfig(

Then, we’ll define the training parameters. But, before that, we’ll make a directory to store the output.

%mkdir /root/results

The training results can be reported to the TensorBoard or Weights & Biases, but I have used TensorBoard here. Before passing the TrainingArguments, make sure that TensorFlow and TensorBoard are installed in your node.

The training parameters will look like this:

training_args = TrainingArguments(

Now to fine-tune the model under supervised training, we’ll pass the model, training dataset, tokenizer, PEFT configurations, dataset text field, maximum sequence length, and training parameters to the SFT trainer, which is a module under the TRL library. TRL library is a Hugging Face library, especially for reinforcement learning.

trainer = SFTTrainer(

Then, we’ll train the model and save the model. 


Using the text generation pipeline, we’ll pass the prompt and get the responses accordingly.

pipe = pipeline(task="text-generation", model=model, tokenizer=tokenizer, max_length=200)

Reinforcement Learning from Human Feedback

To refine the model’s performance and assess the quality of the response, we move to the RLHF training.

First, we’ll train a reward model and then optimize the LLM.

Reward Modeling

To create a reward model, you can create a reward model class, or simply use the RewardTrainer from TRL Hugging Face library. 

Reward model class can be defined as:

class RewardTrainer(Trainer):
    def compute_loss(self, model, inputs, return_outputs=False):
        # Compute rewards for inputs 'j' and 'k' using the provided model
        rewards_j = model(input_ids=inputs["input_ids_j"], attention_mask=inputs["attention_mask_j"])[0]
        rewards_k = model(input_ids=inputs["input_ids_k"], attention_mask=inputs["attention_mask_k"])[0]
        # Compute the negative log-sigmoid loss between the rewards
        loss = -nn.functional.logsigmoid(rewards_j - rewards_k).mean()
        if return_outputs:
            # Return loss and rewards for debugging or monitoring purposes
            return loss, {"rewards_j": rewards_j, "rewards_k": rewards_k}
        return loss

If you’re choosing to use RewardTrainer, then it expects a very specific format for the dataset. It uses four entries, which are:

  • input_ids_chosen
  • attention_mask_chosen
  • input_ids_rejected
  • attention_mask_rejected

For reward modeling, I have used this dataset, which meets the requirements of RewardTrainer already.

train_dataset = load_dataset("Anthropic/hh-rlhf", split="train")

Then, we’ll define the preprocessing function in which all four entries will be there.

def preprocess_function(examples):
    new_examples = {
        "input_ids_chosen": [],
        "attention_mask_chosen": [],
        "input_ids_rejected": [],
        "attention_mask_rejected": [],
    for chosen, rejected in zip(examples["chosen"], examples["rejected"]):
        tokenized_j = tokenizer(chosen, truncation=True)
        tokenized_k = tokenizer(rejected, truncation=True)


    return new_examples

We’ll map the dataset to the preprocessing function in batches.

# Apply preprocess_function to the training dataset in parallel using multiple processes
train_dataset = train_dataset.map(
# Filter the dataset to remove examples where the length of input_ids_chosen and input_ids_rejected exceeds 512 tokens
train_dataset = train_dataset.filter(
    lambda x: len(x["input_ids_chosen"]) <= 512
    and len(x["input_ids_rejected"]) <= 512

Then, we’ll pass the quantized model, tokenizer, training arguments, dataset, and PEFTconfiguration to the RewardTrainer.

trainer = RewardTrainer(

After that, we’ll train and save the model.


LLM Optimization

There are many optimization techniques to optimize an LLM, but Direct Preference Optimization and Proximal Policy Optimization are mostly used with the reward model. DPO is used when the reward function is not defined, while PPO works with the reward function. Here, we defined the Reward function using RewardTrainer. 

Proximal Policy Optimization

The PPOTrainer has the requirement to align a generated response with a query where the rewards are obtained from the reward model. In each step of the PPO Trainer, a batch of prompts is sampled from the dataset, and then these prompts are used to generate the responses from the SFT model. 

After that, the reward model is used to compute the rewards for the generated response. At last, these computed rewards are used to optimize the SFT model using the PPO Trainer.

PPO trainer expects the dataset to have the text column so that it can be renamed to query later.  In the Reward training model, we used this dataset, where the columns are ‘chosen’ and ‘rejected’. But here in PPO training, we have to select one column which we can rename to ‘query’, which is a PPO trainer’s requirement.

train_dataset = train_dataset.rename_column("chosen", "query")
train_dataset = train_dataset.remove_columns(["rejected"])

Now, initialize the PPO trainer by setting its configuration.

from trl import PPOConfig

config = PPOConfig(

After that, we will initialize our model. PPO requires a reference model which is generated by the PPOTrainer automatically.

from trl import AutoModelForCausalLMWithValueHead, PPOConfig, PPOTrainer

model = AutoModelForCausalLMWithValueHead.from_pretrained(config.model_name)
tokenizer = AutoTokenizer.from_pretrained(config.model_name)

tokenizer.pad_token = tokenizer.eos_token

Then we will train the PPO trainer with the SFT model, PPO configurations, dataset, and tokenizer. After that, we’ll save the model.

from trl import PPOTrainer

ppo_trainer = PPOTrainer(

Direct Preference Optimization

The DPO trainer also expects a very specific format for the dataset. Using the DPO trainer, the model will be trained to directly optimize preference for the sentence that is most relevant, given two sentences. 

The DPO trainer requires these three entries:

  • Prompt
  • Chosen
  • Rejected

In our dataset, chosen and rejected are already present, but the prompt column is not there. Just for an experiment, I generated the relevant prompt manually and added the prompt column to the dataset. 

def generate_prompt(chosen_text, rejected_text):
    # Combine the chosen and rejected text to form a prompt
    prompt = f"Given the chosen text: '{chosen_text}', and the rejected text: '{rejected_text}', please provide a response."
    return prompt

for example in train_dataset:
    # Generate prompt based on 'chosen' and 'rejected' columns
    prompt = generate_prompt(example['chosen'], example['rejected'])
    # Add the generated prompt to the dataset
    example['prompt'] = prompt

After this, my dataset meets the expectation of a DPO Trainer. We’ll initialize the DPO Trainer by passing the SFT model, training arguments, dataset, and tokenizer. We’ll train the model with a DPO trainer and save the model.

dpo_trainer = DPOTrainer(


From unsupervised pretrained models to optimized models using reinforcement learning, we have explored various training methods for large language models. We utilized E2E Network’s Cloud GPUs to train different models for efficacy and efficient performance. Since all the models are saved in your directory, it’s your time to experiment with them and monitor their responses with respect to your datasets.

Thanks for reading!

Latest Blogs
This is a decorative image for: A Complete Guide To Customer Acquisition For Startups
October 18, 2022

A Complete Guide To Customer Acquisition For Startups

Any business is enlivened by its customers. Therefore, a strategy to constantly bring in new clients is an ongoing requirement. In this regard, having a proper customer acquisition strategy can be of great importance.

So, if you are just starting your business, or planning to expand it, read on to learn more about this concept.

The problem with customer acquisition

As an organization, when working in a diverse and competitive market like India, you need to have a well-defined customer acquisition strategy to attain success. However, this is where most startups struggle. Now, you may have a great product or service, but if you are not in the right place targeting the right demographic, you are not likely to get the results you want.

To resolve this, typically, companies invest, but if that is not channelized properly, it will be futile.

So, the best way out of this dilemma is to have a clear customer acquisition strategy in place.

How can you create the ideal customer acquisition strategy for your business?

  • Define what your goals are

You need to define your goals so that you can meet the revenue expectations you have for the current fiscal year. You need to find a value for the metrics –

  • MRR – Monthly recurring revenue, which tells you all the income that can be generated from all your income channels.
  • CLV – Customer lifetime value tells you how much a customer is willing to spend on your business during your mutual relationship duration.  
  • CAC – Customer acquisition costs, which tells how much your organization needs to spend to acquire customers constantly.
  • Churn rate – It tells you the rate at which customers stop doing business.

All these metrics tell you how well you will be able to grow your business and revenue.

  • Identify your ideal customers

You need to understand who your current customers are and who your target customers are. Once you are aware of your customer base, you can focus your energies in that direction and get the maximum sale of your products or services. You can also understand what your customers require through various analytics and markers and address them to leverage your products/services towards them.

  • Choose your channels for customer acquisition

How will you acquire customers who will eventually tell at what scale and at what rate you need to expand your business? You could market and sell your products on social media channels like Instagram, Facebook and YouTube, or invest in paid marketing like Google Ads. You need to develop a unique strategy for each of these channels. 

  • Communicate with your customers

If you know exactly what your customers have in mind, then you will be able to develop your customer strategy with a clear perspective in mind. You can do it through surveys or customer opinion forms, email contact forms, blog posts and social media posts. After that, you just need to measure the analytics, clearly understand the insights, and improve your strategy accordingly.

Combining these strategies with your long-term business plan will bring results. However, there will be challenges on the way, where you need to adapt as per the requirements to make the most of it. At the same time, introducing new technologies like AI and ML can also solve such issues easily. To learn more about the use of AI and ML and how they are transforming businesses, keep referring to the blog section of E2E Networks.

Reference Links




This is a decorative image for: Constructing 3D objects through Deep Learning
October 18, 2022

Image-based 3D Object Reconstruction State-of-the-Art and trends in the Deep Learning Era

3D reconstruction is one of the most complex issues of deep learning systems. There have been multiple types of research in this field, and almost everything has been tried on it — computer vision, computer graphics and machine learning, but to no avail. However, that has resulted in CNN or convolutional neural networks foraying into this field, which has yielded some success.

The Main Objective of the 3D Object Reconstruction

Developing this deep learning technology aims to infer the shape of 3D objects from 2D images. So, to conduct the experiment, you need the following:

  • Highly calibrated cameras that take a photograph of the image from various angles.
  • Large training datasets can predict the geometry of the object whose 3D image reconstruction needs to be done. These datasets can be collected from a database of images, or they can be collected and sampled from a video.

By using the apparatus and datasets, you will be able to proceed with the 3D reconstruction from 2D datasets.

State-of-the-art Technology Used by the Datasets for the Reconstruction of 3D Objects

The technology used for this purpose needs to stick to the following parameters:

  • Input

Training with the help of one or multiple RGB images, where the segmentation of the 3D ground truth needs to be done. It could be one image, multiple images or even a video stream.

The testing will also be done on the same parameters, which will also help to create a uniform, cluttered background, or both.

  • Output

The volumetric output will be done in both high and low resolution, and the surface output will be generated through parameterisation, template deformation and point cloud. Moreover, the direct and intermediate outputs will be calculated this way.

  • Network architecture used

The architecture used in training is 3D-VAE-GAN, which has an encoder and a decoder, with TL-Net and conditional GAN. At the same time, the testing architecture is 3D-VAE, which has an encoder and a decoder.

  • Training used

The degree of supervision used in 2D vs 3D supervision, weak supervision along with loss functions have to be included in this system. The training procedure is adversarial training with joint 2D and 3D embeddings. Also, the network architecture is extremely important for the speed and processing quality of the output images.

  • Practical applications and use cases

Volumetric representations and surface representations can do the reconstruction. Powerful computer systems need to be used for reconstruction.

Given below are some of the places where 3D Object Reconstruction Deep Learning Systems are used:

  • 3D reconstruction technology can be used in the Police Department for drawing the faces of criminals whose images have been procured from a crime site where their faces are not completely revealed.
  • It can be used for re-modelling ruins at ancient architectural sites. The rubble or the debris stubs of structures can be used to recreate the entire building structure and get an idea of how it looked in the past.
  • They can be used in plastic surgery where the organs, face, limbs or any other portion of the body has been damaged and needs to be rebuilt.
  • It can be used in airport security, where concealed shapes can be used for guessing whether a person is armed or is carrying explosives or not.
  • It can also help in completing DNA sequences.

So, if you are planning to implement this technology, then you can rent the required infrastructure from E2E Networks and avoid investing in it. And if you plan to learn more about such topics, then keep a tab on the blog section of the website

Reference Links



This is a decorative image for: Comprehensive Guide to Deep Q-Learning for Data Science Enthusiasts
October 18, 2022

A Comprehensive Guide To Deep Q-Learning For Data Science Enthusiasts

For all data science enthusiasts who would love to dig deep, we have composed a write-up about Q-Learning specifically for you all. Deep Q-Learning and Reinforcement learning (RL) are extremely popular these days. These two data science methodologies use Python libraries like TensorFlow 2 and openAI’s Gym environment.

So, read on to know more.

What is Deep Q-Learning?

Deep Q-Learning utilizes the principles of Q-learning, but instead of using the Q-table, it uses the neural network. The algorithm of deep Q-Learning uses the states as input and the optimal Q-value of every action possible as the output. The agent gathers and stores all the previous experiences in the memory of the trained tuple in the following order:

State> Next state> Action> Reward

The neural network training stability increases using a random batch of previous data by using the experience replay. Experience replay also means the previous experiences stocking, and the target network uses it for training and calculation of the Q-network and the predicted Q-Value. This neural network uses openAI Gym, which is provided by taxi-v3 environments.

Now, any understanding of Deep Q-Learning   is incomplete without talking about Reinforcement Learning.

What is Reinforcement Learning?

Reinforcement is a subsection of ML. This part of ML is related to the action in which an environmental agent participates in a reward-based system and uses Reinforcement Learning to maximize the rewards. Reinforcement Learning is a different technique from unsupervised learning or supervised learning because it does not require a supervised input/output pair. The number of corrections is also less, so it is a highly efficient technique.

Now, the understanding of reinforcement learning is incomplete without knowing about Markov Decision Process (MDP). MDP is involved with each state that has been presented in the results of the environment, derived from the state previously there. The information which composes both states is gathered and transferred to the decision process. The task of the chosen agent is to maximize the awards. The MDP optimizes the actions and helps construct the optimal policy.

For developing the MDP, you need to follow the Q-Learning Algorithm, which is an extremely important part of data science and machine learning.

What is Q-Learning Algorithm?

The process of Q-Learning is important for understanding the data from scratch. It involves defining the parameters, choosing the actions from the current state and also choosing the actions from the previous state and then developing a Q-table for maximizing the results or output rewards.

The 4 steps that are involved in Q-Learning:

  1. Initializing parameters – The RL (reinforcement learning) model learns the set of actions that the agent requires in the state, environment and time.
  2. Identifying current state – The model stores the prior records for optimal action definition for maximizing the results. For acting in the present state, the state needs to be identified and perform an action combination for it.
  3. Choosing the optimal action set and gaining the relevant experience – A Q-table is generated from the data with a set of specific states and actions, and the weight of this data is calculated for updating the Q-Table to the following step.
  4. Updating Q-table rewards and next state determination – After the relevant experience is gained and agents start getting environmental records. The reward amplitude helps to present the subsequent step.  

In case the Q-table size is huge, then the generation of the model is a time-consuming process. This situation requires Deep Q-learning.

Hopefully, this write-up has provided an outline of Deep Q-Learning and its related concepts. If you wish to learn more about such topics, then keep a tab on the blog section of the E2E Networks website.

Reference Links



This is a decorative image for: GAUDI: A Neural Architect for Immersive 3D Scene Generation
October 13, 2022

GAUDI: A Neural Architect for Immersive 3D Scene Generation

The evolution of artificial intelligence in the past decade has been staggering, and now the focus is shifting towards AI and ML systems to understand and generate 3D spaces. As a result, there has been extensive research on manipulating 3D generative models. In this regard, Apple’s AI and ML scientists have developed GAUDI, a method specifically for this job.

An introduction to GAUDI

The GAUDI 3D immersive technique founders named it after the famous architect Antoni Gaudi. This AI model takes the help of a camera pose decoder, which enables it to guess the possible camera angles of a scene. Hence, the decoder then makes it possible to predict the 3D canvas from almost every angle.

What does GAUDI do?

GAUDI can perform multiple functions –

  • The extensions of these generative models have a tremendous effect on ML and computer vision. Pragmatically, such models are highly useful. They are applied in model-based reinforcement learning and planning world models, SLAM is s, or 3D content creation.
  • Generative modelling for 3D objects has been used for generating scenes using graf, pigan, and gsn, which incorporate a GAN (Generative Adversarial Network). The generator codes radiance fields exclusively. Using the 3D space in the scene along with the camera pose generates the 3D image from that point. This point has a density scalar and RGB value for that specific point in 3D space. This can be done from a 2D camera view. It does this by imposing 3D datasets on those 2D shots. It isolates various objects and scenes and combines them to render a new scene altogether.
  • GAUDI also removes GANs pathologies like mode collapse and improved GAN.
  • GAUDI also uses this to train data on a canonical coordinate system. You can compare it by looking at the trajectory of the scenes.

How is GAUDI applied to the content?

The steps of application for GAUDI have been given below:

  • Each trajectory is created, which consists of a sequence of posed images (These images are from a 3D scene) encoded into a latent representation. This representation which has a radiance field or what we refer to as the 3D scene and the camera path is created in a disentangled way. The results are interpreted as free parameters. The problem is optimized by and formulation of a reconstruction objective.
  • This simple training process is then scaled to trajectories, thousands of them creating a large number of views. The model samples the radiance fields totally from the previous distribution that the model has learned.
  • The scenes are thus synthesized by interpolation within the hidden space.
  • The scaling of 3D scenes generates many scenes that contain thousands of images. During training, there is no issue related to canonical orientation or mode collapse.
  • A novel de-noising optimization technique is used to find hidden representations that collaborate in modelling the camera poses and the radiance field to create multiple datasets with state-of-the-art performance in generating 3D scenes by building a setup that uses images and text.

To conclude, GAUDI has more capabilities and can also be used for sampling various images and video datasets. Furthermore, this will make a foray into AR (augmented reality) and VR (virtual reality). With GAUDI in hand, the sky is only the limit in the field of media creation. So, if you enjoy reading about the latest development in the field of AI and ML, then keep a tab on the blog section of the E2E Networks website.

Reference Links




Build on the most powerful infrastructure cloud

A vector illustration of a tech city using latest cloud technologies & infrastructure