Domain 3

Question 1

T or F: LLMs are cost-effective and easy to maintain

Answer 1

False. The duration and cost of training a model are important considerations because it can be expensive for hardware storage and more.

Question 2

Name five considerations of using foundational models

Answer 2

latency constraints, inference speed, and real-time requirements, Architecture and complexity

Question 3

T or F: Accuracy is recommended with datasets that are not evenly distributed or imbalanced.

Answer 3

Accuracy is not recommended with datasets that are not evenly distributed or imbalanced.

Question 4

Name some metrics you can use to evaluate model performance

Answer 4

Such metrics might include accuracy, precision, recall, F1 score, root mean squared error or RMSE, mean average precision or MAP, and mean absolute error, MAE.

Question 5

biases that might be present in the training data. It’s important to understand how to mitigate risks, address ethical concerns, and make informed decisions about model selection and fine-tuning

Question 6

Another consideration is the availability and compatibility of the pre-trained model

Answer 6

you should check whether the model is compatible with your framework, language, and environment, and check to ensure it has a license and documentation. You should also check whether the model has been updated and maintained regularly and whether it has any known issues or limitations.

Question 7

interpretability

Answer 7

he ability to interpret and explain model outcomes is important. Being transparent refers to interpretability. it means being able to explain mathematically through coefficients and formulas why a model makes a certain prediction. This interpretability is possible if the model is simple enough, but foundation models are not interpretable by design because they are extremely complex. if interpretability is a requirement, then pre-trained foundation models might not be the best choice.

Question 8

How is explainability different from intertability

Answer 8

Explainability attempts to explain this black box, by approximating it locally with a simpler model that is interpretable.

Question 9

Greater complexity might lead to enhanced performance…

Answer 9

, but can increase costs. The more complicated the model is, the harder it is to explain the outputs of the model. And there are more considerations too, such as hardware constraints, maintenance updates, data privacy, transfer learning, and more.

Question 10

What is inference?

Answer 10

he inference is where you process new data through the model to make predictions. It is the process of generating an output from an input that you provided to model

Question 11

Amazon Bedrock foundation models supports what inference parameters?

Answer 11

Temperature, Top K, Top P to control randomness and diversity in the response. Amazon Bedrock also supports parameters such as response length, penalties, and stop sequences to limit the length of the response.

Question 12

What is a prerequisite to creating a vector database?

Answer 12

A vector database is filled with dense vectors by processing input data, generally text data, and using an ML model, generally an embedding model. So it’s important to understand that a machine learning model is a prerequisite to create a vector database and the indexing technology itself. Vector databases are the factual reference of foundation model based applications, helping the model retrieve trustworthy data. Foundation models use vector databases as an external data source to improve their capabilities by enhancing search recommendations and text generation use cases. Vector databases add additional capabilities for efficient and fast lookup, and to provide data management, fault tolerance, authentication, and access control and query engine.

Question 13

knowledge bases for Amazon Bedrock

Answer 13

give you the ability of collecting data sources into a repository of information. This way, you build an application that takes advantage of retrieval augmented generation, RAG.

Question 14

What two components does RAG have?

Answer 14

RAG combines two components, a retriever component, which searches through a knowledge base and a generator component, which produces outputs based on the retrieved information. This combination helps the model access up-to-date and domain-specific knowledge beyond their training data.

Question 15

AWS services that help store embeddings within vector databases.

Answer 15

Examples include Amazon OpenSearch Service, Amazon Aurora, Redis, Amazon Neptune, Amazon DocumentDB with MongoDB compatibility, and Amazon RDS with PostgreSQL.

Question 16

Amazon RDS for PostgreSQL also supports the pgvector extension

Answer 16

to store embeddings and perform efficient searches.

Question 17

Agents for Amazon Bedrock

Answer 17

is a fully managed AI capability from AWS to help you build applications foundation models. Agents can automatically break down tasks and generate the required orchestration logic or write custom code, and agents can securely connect to your databases through APIs, they can ingest and structure the data for machine consumption and augment it with contextual details to produce more accurate responses and fulfill requests.

Question 18

Temperature:

Answer 18

Adjusts the randomness of the model’s response. A lower temperature results in more focused responses, while a higher temperature leads to more diverse outputs.

Question 19

Top-k

Answer 19

defines the cut-off for the number of words (tokens) for each completion to choose from, ordered by their probabilities. A lower Top K value reduces the chance of an unusual word being selected.

Question 20

Top-p

Answer 20

Top P works similarly to Top K The percentage of most-likely candidates that the model considers for the next token.

Choose a lower value to decrease the size of the pool and limit the options to more likely outputs.

Choose a higher value to increase the size of the pool and allow the model to consider less likely outputs.

Question 21

The BERT score is a metric developed to assess the quality of generated responses compared to a set of reference responses. It uses pre-trained models to calculate semantic similarity between the generated responses and reference answers. Here’s how it works:

Question 22

ROUGE, or Recall-Oriented Understudy for Gisting Evaluation, is a set of metrics and a software package used for evaluating automatic summarization and machine translation software in natural language processing. The metrics compare an automatically produced summary or translation against a reference or a set of references (human-produced) summary or translation.

Question 23

What is few-shot prompting? I

Answer 23

t is when you provide a few examples to help the LLM models better perform and calibrate their output to meet your expectations.

Question 24

What is zero-shot prompting?

Answer 24

It is a sentiment classification prompt with no examples provided to the prompt.

Question 25

You can also use a prompt template. .

Answer 25

Templates might include instructions, few-shot examples, and specific content and questions for different use cases

Question 26

chain-of-thought prompting to break down the reasoning process into intermediate steps. This kind of prompting can improve the quality and coherence of the final output.

Question 27

prompt tuning,

Answer 27

where the actual prompt text is replaced with a continuous embedding backer that is optimized during training. This technique helps the prompt to be fine-tuned for a specific task. At the same time, it keeps the rest of the model parameters frozen, which can be more efficient than full fine-tuning

Question 28

What are common tasks supported by LLMs on Amazon Bedrock?

Answer 28

A few are classification, question and answer with and without context, summarization, open-ended text generation, code generation, math, and reasoning or logical thinking.

Question 29

Latent space

Answer 29

is the encoded knowledge of language in a large language model. It’s the stored patterns of data that capture relationships and, when prompted, reconstruct language from those patterns.

Question 30

f you prompt a language model and get dissatisfactory or a negative response, then your prompt might be insufficient for the model.

Answer 30

But it’s also possible, especially if your model is smaller, that the model’s latent space might not have enough information about the topic of your prompt. That situation can cause the model to hallucinate a model that doesn’t know the exact specifics of a prompt because the knowledge isn’t in its latent space will choose the closest match. This result might be interpreted as a mistake, but the model is actually functioning correctly.

Question 31

Prompt engineering has several key techniques. First, be specific and provide clear instructions or specifications for the task at hand. For example, include the desired format, examples, comparison, style, tone, output length, and detailed context. Second, include examples of the desired behavior and direction, such as sample texts, data formats, templates, code, graphs, charts, and more. Third, experiment and use an iterative process to test prompts and understand how the modifications alter the responses. Fourth, know the strengths and weaknesses of your model. Fifth, balance simplicity and complexity in your prompts to avoid vague, unrelated, or unexpected answers. Six, specifically for your prompt engineers, use multiple comments to offer more context without cluttering your prompt. Seventh, add guardrails.

Question 32

hat is prompt injection?

Answer 32

It describes attacks of prompt manipulation.

Question 33

jailbreaking

Answer 33

When an attacker tries to bypass the guardrails that you have established, this is called jailbreaking. In this case, it is different because jailbreaking targets the safety measures put in place such as guardrails

Question 34

Hijacking

Answer 34

is an attempt to change or manipulate the original prompt with new instructions.

Question 35

Poisoning

Answer 35

is another risk of prompt engineering where harmful instructions are embedded in messages, emails, web pages, and more

Question 36

Guardrails

Answer 36

uardrails provide safety and privacy controls to manage interactions in your generative AI applications. You can define topics within the context of your application that are not desirable. You can set words to be blocked. You can configure thresholds to filter across categories that might be harmful and prompt attacks such as jailbreak and prompt injections. You can also filter inputs that might contain sensitive data.

Question 37

What are the key elements of training a foundation model? They include pre-training, fine-tuning, and continuous pre-training

Answer 37

They include pre-training, fine-tuning, and continuous pre-training

Question 38

pre-training,

Answer 38

which is a complex process. It requires millions of graphic processing units, GPUs, compute hours, terabytes and petabytes of data, trillions of tokens, trial and error, and time.

Question 39

Fine-tuning

Answer 39

is a process that extends the training of the model to improve the generation of completions for a specific task. It is a supervised learning process and you use a dataset of labeled examples to update the weights of the LLM

Question 40

happens when the whole fine-tuning process modifies the weights of the original LLM. This can improve the performance of the single task fine-tuning, but it can degrade performance on other tasks.

Answer 40

Catastrophic forgetting

Question 41

Parameter-efficient fine-tuning, PEFT,

Answer 41

is a process and set of techniques that freeze or preserve the parameters and weights of the original LLM and fine-tune or train a small number of task-specific adaptor layers and parameters. PEFT reduces the compute and memory that’s needed because it’s fine-tuning a small set of model parameters

Question 42

Low-rank adaptation or LoRA,

Answer 42

is a popular PEFT technique that also preserves or freezes the original weights of the foundation model and creates new trainable low-rank matrices into each layer of a transformer architecture.

Question 43

T or false PEFT and LoRA modify the weights of your model, but not the representations.

Answer 43

True

Question 44

Multitask fine-tuning

Answer 44

is an extension of fine-tuning a single task. Multitask fine-tuning requires a lot of data. For example, the dataset might contain examples that instruct a model to complete multiple tasks such as reviews or ratings, summarization, translating code, and more. This produces an instruction tuned model that has learned how to complete many different tasks simultaneously.

Question 45

What fine-tuning process modifies the weights of the model to adapt to domain-specific data?

Answer 45

Domain adaptation fine-tuning gives you the ability to use the pre-trained foundation models and adapt them to specific tasks by using limited domain-specific data. You can use domain adaptation fine-tuning to help your model work with domain-specific language such as industry jargon, technical terms, or other specialized data

Question 46

mazon SageMaker JumpStart provides the capability to fine-tune a large language model, particularly a text generation model on the domain- specific dataset. You can fine-tune models with their custom dataset to improve performance in specific domains.

Question 47

reinforcement learning from human feedback, or RLHF

Answer 47

RLHF uses reinforcement learning to fine-tune the LLM with human feedback data to better align the model with human preferences

Question 48

If you have low-code data preparation, you can use this to create data flows that define your ML data pre-processing. These data flows feature engineering workflows that use little to no coding.

Answer 48

Amazon SageMaker Canvas

Question 49

Suppose that you have data preparation that needs to detect bias in your data. You can use this to analyze your data and detect potential biases across multiple facets.

Answer 49

Amazon SageMaker Clarify

Question 50

One optimization technique is to improve application performance by reducing the size of the LLMs. This action can reduce the inference latency because the smaller size model loads more quickly. However,

Answer 50

remember that reducing the size of the model might decrease its performance.

Question 51

Other optimization techniques include

Answer 51

making a more concise prompt, reducing the size of the retrieved snippets and their number, and reducing generation through inference parameters and prompt.

Question 52

metrics, such as accuracy or root mean square error, RMSE, are more straightforward to calculate. This is because the predictions are deterministic and can be compared against the labels. Can you use these for Gen AI?

Answer 52

No. The output of generative AI models is non-deterministic and makes the evaluation more difficult to determine.

Question 53

Recall Oriented Understudy for Gisting Evaluation, or ROUGE,

Answer 53

is a set of metrics and a software package. It is used to evaluate automatic summarization tasks and machine translation software in natural language processing. It evaluates how well the input compares to the generated output.

Question 54

Bilingual Evaluation Understudy, or BLEU,

Answer 54

is an algorithm that is used for translation tasks. It evaluates the quality of text which has been machine translated from one natural language to another.

Question 55

You can use this to evaluate large language models, LLMs, and create model evaluation jobs. A model evaluation job helps to evaluate and compare model quality and metrics for text-based foundation models from SageMaker JumpStart.

Answer 55

Amazon SageMaker Clarify

Question 56

Amazon Bedrock provides this that can automatically compare generated responses and calculate a semantic similarity base score, BERTscore, against a human reference. It is suitable to evaluate faithfulness and hallucinations in text-generation tasks.

Answer 56

an evaluation module

Question 57

General Language Understanding Evaluation, GLUE,

Answer 57

It is a collection of natural language tasks, such as sentiment analysis and question answering. You can use these tasks to evaluate and compare model performance across a set of language tasks. Then, you can use the benchmark to measure and compare the model performance

Question 58

Massive Multitask Language Understanding, MMLU,

Answer 58

evaluates the knowledge and problem-solving capabilities of the model. To perform well, models must have extensive world knowledge and problem-solving ability. The models are tested on more than basic language understanding, such as history, mathematics, laws, computer science, and more.

Question 59

The Beyond the Imitation Game Benchmark, BIG-bench,

Answer 59

focuses on tasks that are beyond the capabilities of the current language models. It contains tasks such as math, biology, physics, bias, linguistics, reasoning, childhood development, software development, and more.

Question 60

Holistic Evaluation of Language Models, HELM,

Answer 60

which is a benchmark to help improve model transparency. It offers users guidance on which model performs well for a given task. HELM is a combination of metrics for tasks such as summarization, question and answer, sentiment analysis, and bias detection

Question 61

Instruction tuning

Answer 61

refers to the process of providing specific labeled examples to train a model on a specific task. Instruction tuning can help models follow specific tasks or responses to prompts in a specific way

Question 62

Length

Answer 62

Foundation models typically support parameters that limit the length of the response. Examples of these parameters are provided below.

Response length – An exact value to specify the minimum or maximum number of tokens to return in the generated response.

Penalties – Specify the degree to which to penalize outputs in a response. Examples include the following.

The length of the response.

Repeated tokens in a response.

Frequency of tokens in a response.

Types of tokens in a response.

Stop sequences – Specify sequences of characters that stop the model from generating further tokens. If the model generates a stop sequence that you specify, it will stop generating after that sequence.