Embeddings in Depth - Part of the Ollama Course

Показать описание

Dive into the world of embeddings and their crucial role in modern AI applications, particularly in enhancing search capabilities and information retrieval. This video, part of our comprehensive Ollama course, explains:

- What embeddings are and how they differ from traditional text-matching searches
- The importance of embeddings in Retrieval Augmented Generation (RAG)
- How to create and use embeddings with Ollama's API
- A practical comparison of different embedding models, including:
- nomic-embed-text
- mxbai-embed-large
- all-minilm
- snowflake-arctic-embed
- bge-m3
- bge-large
- llama3.1

We'll demonstrate real-world applications, discuss performance considerations, and explore the nuances of working with embeddings. Whether you're new to AI or looking to deepen your understanding, this video provides valuable insights into this powerful technology.

Join us as we uncover how embeddings are transforming the way we interact with and retrieve information in the age of AI.

#AI #MachineLearning #Embeddings #Ollama #InformationRetrieval

(they have a pretty url because they are paying at least $100 per month for Discord. You help get more viewers to this channel and I can afford that too.)

Join this channel to get access to perks:

00:00 - Start
00:36 - There is another way
00:47 - Welcome to the course
01:25 - How do embeddings fit in
01:45 - What does the actual embedding
02:12 - Dimensions
02:39 - Similarity Search
03:39 - How to create the embedding
03:58 - The 3 endpoints
04:43 - The right endpoint to use
05:25 - Python and JS/TS libraries
05:54 - Let's look at a simple example
06:09 - The sample I used to embed
06:50 - Which is faster
07:34 - Let's look at the answers
08:55 - Where to find the example code
09:08 - Some of the variables to play with
09:32 - Frustrations

Рекомендации по теме

Комментарии

Matt, thanks for the instruction. Best explanations I've found thus far, so many mile-wide and one inch deep overviews out there. Any other sources you find useful for someone climbing the learning curve?

Another question, it seems like just taking a corpus and building the vector database and then just returning the results to the user would be a pretty powerful search tool. My project is building an expert system from a history of user group messages on a specific technical topic.

MTom

Hi Matt. Love your series. I am running on a decent spec Asus M15 but this script is taking more than 30mins to run. I am running deno out of powershell. Would I do better in WSL2? Thx

jcohen

Great video. Thank you for sharing. Quick question Matt, which is your primary, go-to embedding model? I was using mxbai-e-l-v1 until July, but switched to My people are now hinting for me to switch to all-MiniLM-L6-v2, but i didn't know it was an embedding model. I'll have to do some research if i can find the time...or delegate. Your thoughts? 😁

trystianfx

Out of interest for the llama3.1 embedding test - what quantisation was that? Just wondering if you’re using a q4_k_m if it improves at all with something like q6_k_l which has less quantised embedding heads.

sammcj

Cool, this channel is really great. With your help, I am able to train my models very easily and implemented RAG with the backend, but I need to ask you one thing.
I have created embeddings on mongoDB to perform vector search, but to use it, we have to assign limit inside the vector search aggregation pipeline which always returns me the non-essential data. If the limit is too high, and if it is low, it will definitely miss some of the important data. I can't really fix this limit to get a particular number of matches because of the backend complexity. Can you please suggest me some alternatives or any solution to get the result which matches the user query. I also can't rely on the vector search score because they all are so close depending on the user query.

basantrajshakti

Are there any benchmarks for embedding models? Are these different from the LLM benchmark s?

narulavarun

are there any correlation between language used within the text that is converted into vectors and the embedding modle that is used? Are there any non english embedding models or how could one create an own embedding model?

conneyk

CPUs or GPUs for the vectors? I thought ai used mainly the speed of gpu. Just a question. Sorry.

anthony-fi

Embeddings in Depth - Part of the Ollama Course

Embeddings in Depth - Part of the Ollama Course

Vectoring Words (Word Embeddings) - Computerphile

Word Embeddings & Positional Encoding in NLP Transformer model explained - Part 1

Deep Dive into Multimodal Embeddings Part 1&2

Word Embeddings, Word2Vec And CBOW Indepth Intuition And Working- Part 1 | NLP For Machine Learning

SMLTAR: Word Embeddings Part 1 (smltar01 5)

How large language models work, a visual intro to transformers | Chapter 5, Deep Learning

Part 1: Dependency parsing with structure preserving embeddings

Multi Head Attention Part 2: Entire mathematics explained

what are embeddings | vector databases | part 0 | all about embeddings

Lecture 2 | Word Vector Representations: word2vec

Mastering Embeddings with OpenAI: Part 1 - Understanding, Generating and Visualising Embeddings

Embeddings vs Fine Tuning - Part 1, Embeddings

Machine Learning from Square One 2020 - Part 2: Word Embeddings and Language Modelling

Part 1 | Training Word Embeddings | Word2Vec

#012 - Monetising Embeddings and ChatGPT: Architecture Deep Dive - Part I

Exploring Matryoshka Embedding Models | PART I

Transformer Embeddings - Part 3 | Visualization of Embeddings

Part 2 | Python | Training Word Embeddings | Word2Vec |

Comparison of Embedding Models for RAG | Part 1

SMLTAR: Word Embeddings Part 2 (smltar01 5)

Choosing Embedding Models for RAG Applications Part 1

Choosing Embedding Models for RAG Applications Part 2

Histology: Embedding Process