44 items tagged “embeddings”
2024
Is async Django ready for prime time? (via) Jonathan Adly reports on his experience using Django to build ColiVara, a hosted RAG API that uses ColQwen2 visual embeddings, inspired by the ColPali paper.
In a breach of Betteridge's law of headlines the answer to the question posed by this headline is “yes”.
We believe async Django is ready for production. In theory, there should be no performance loss when using async Django instead of FastAPI for the same tasks.
The ColiVara application is itself open source, and you can see how it makes use of Django’s relatively new asynchronous ORM features in the api/views.py module.
I also picked up a useful trick from their Dockerfile: if you want uv in a container you can install it with this one-liner:
COPY --from=ghcr.io/astral-sh/uv:latest /uv /bin/uv
Weeknotes: asynchronous LLMs, synchronous embeddings, and I kind of started a podcast
These past few weeks I’ve been bringing Datasette and LLM together and distracting myself with a new sort-of-podcast crossed with a live streaming experiment.
[... 896 words]llm-gguf 0.2, now with embeddings. This new release of my llm-gguf plugin - which provides support for locally hosted GGUF LLMs - adds a new feature: it now supports embedding models distributed as GGUFs as well.
This means you can use models like the bafflingly small (30.8MB in its smallest quantization) mxbai-embed-xsmall-v1 with LLM like this:
llm install llm-gguf
llm gguf download-embed-model \
'https://huggingface.co/mixedbread-ai/mxbai-embed-xsmall-v1/resolve/main/gguf/mxbai-embed-xsmall-v1-q8_0.gguf'
Then to embed a string:
llm embed -m gguf/mxbai-embed-xsmall-v1-q8_0 -c 'hello'
The LLM docs have extensive coverage of things you can then do with this model, like embedding every row in a CSV file / file in a directory / record in a SQLite database table and running similarity and semantic search against them.
Under the hood this takes advantage of the create_embedding() method provided by the llama-cpp-python wrapper around llama.cpp.
Binary vector embeddings are so cool (via) Evan Schwartz:
Vector embeddings by themselves are pretty neat. Binary quantized vector embeddings are extra impressive. In short, they can retain 95+% retrieval accuracy with 32x compression and ~25x retrieval speedup.
It's so unintuitive how well this trick works: take a vector of 1024x4 byte floating point numbers (4096 bytes = 32,768 bits), turn that into an array of single bits for > 0 or <= 0 which reduces it to just 1024 bits or 128 bytes - a 1/32 reduction.
Now you can compare vectors using a simple Hamming distance - a count of the number of bits that differ - and yet still get embedding similarity scores that are only around 10% less accurate than if you had used the much larger floating point numbers.
Evan digs into models that this works for, which include OpenAI's text-embedding-3-large and the small but powerful all-MiniLM-L6-v2.
Bridging Language Gaps in Multilingual Embeddings via Contrastive Learning (via) Most text embeddings models suffer from a "language gap", where phrases in different languages with the same semantic meaning end up with embedding vectors that aren't clustered together.
Jina claim their new jina-embeddings-v3 (CC BY-NC 4.0, which means you need to license it for commercial use if you're not using their API) is much better on this front, thanks to a training technique called "contrastive learning".
There are 30 languages represented in our contrastive learning dataset, but 97% of pairs and triplets are in just one language, with only 3% involving cross-language pairs or triplets. But this 3% is enough to produce a dramatic result: Embeddings show very little language clustering and semantically similar texts produce close embeddings regardless of their language
Hybrid full-text search and vector search with SQLite. As part of Alex’s work on his sqlite-vec SQLite extension - adding fast vector lookups to SQLite - he’s been investigating hybrid search, where search results from both vector similarity and traditional full-text search are combined together.
The most promising approach looks to be Reciprocal Rank Fusion, which combines the top ranked items from both approaches. Here’s Alex’s SQL query:
-- the sqlite-vec KNN vector search results
with vec_matches as (
select
article_id,
row_number() over (order by distance) as rank_number,
distance
from vec_articles
where
headline_embedding match lembed(:query)
and k = :k
),
-- the FTS5 search results
fts_matches as (
select
rowid,
row_number() over (order by rank) as rank_number,
rank as score
from fts_articles
where headline match :query
limit :k
),
-- combine FTS5 + vector search results with RRF
final as (
select
articles.id,
articles.headline,
vec_matches.rank_number as vec_rank,
fts_matches.rank_number as fts_rank,
-- RRF algorithm
(
coalesce(1.0 / (:rrf_k + fts_matches.rank_number), 0.0) * :weight_fts +
coalesce(1.0 / (:rrf_k + vec_matches.rank_number), 0.0) * :weight_vec
) as combined_rank,
vec_matches.distance as vec_distance,
fts_matches.score as fts_score
from fts_matches
full outer join vec_matches on vec_matches.article_id = fts_matches.rowid
join articles on articles.rowid = coalesce(fts_matches.rowid, vec_matches.article_id)
order by combined_rank desc
)
select * from final;I’ve been puzzled in the past over how to best do that because the distance scores from vector similarity and the relevance scores from FTS are meaningless in comparison to each other. RRF doesn’t even attempt to compare them - it uses them purely for row_number() ranking within each set and combines the results based on that.
Conflating Overture Places Using DuckDB, Ollama, Embeddings, and More.
Drew Breunig's detailed tutorial on "conflation" - combining different geospatial data sources by de-duplicating address strings such as RESTAURANT LOS ARCOS,3359 FOOTHILL BLVD,OAKLAND,94601 and LOS ARCOS TAQUERIA,3359 FOOTHILL BLVD,OAKLAND,94601.
Drew uses an entirely offline stack based around Python, DuckDB and Ollama and finds that a combination of H3 geospatial tiles and mxbai-embed-large embeddings (though other embedding models should work equally well) gets really good results.
Introducing Contextual Retrieval (via) Here's an interesting new embedding/RAG technique, described by Anthropic but it should work for any embedding model against any other LLM.
One of the big challenges in implementing semantic search against vector embeddings - often used as part of a RAG system - is creating "chunks" of documents that are most likely to semantically match queries from users.
Anthropic provide this solid example where semantic chunks might let you down:
Imagine you had a collection of financial information (say, U.S. SEC filings) embedded in your knowledge base, and you received the following question: "What was the revenue growth for ACME Corp in Q2 2023?"
A relevant chunk might contain the text: "The company's revenue grew by 3% over the previous quarter." However, this chunk on its own doesn't specify which company it's referring to or the relevant time period, making it difficult to retrieve the right information or use the information effectively.
Their proposed solution is to take each chunk at indexing time and expand it using an LLM - so the above sentence would become this instead:
This chunk is from an SEC filing on ACME corp's performance in Q2 2023; the previous quarter's revenue was $314 million. The company's revenue grew by 3% over the previous quarter."
This chunk was created by Claude 3 Haiku (their least expensive model) using the following prompt template:
<document>
{{WHOLE_DOCUMENT}}
</document>
Here is the chunk we want to situate within the whole document
<chunk>
{{CHUNK_CONTENT}}
</chunk>
Please give a short succinct context to situate this chunk within the overall document for the purposes of improving search retrieval of the chunk. Answer only with the succinct context and nothing else.
Here's the really clever bit: running the above prompt for every chunk in a document could get really expensive thanks to the inclusion of the entire document in each prompt. Claude added context caching last month, which allows you to pay around 1/10th of the cost for tokens cached up to your specified beakpoint.
By Anthropic's calculations:
Assuming 800 token chunks, 8k token documents, 50 token context instructions, and 100 tokens of context per chunk, the one-time cost to generate contextualized chunks is $1.02 per million document tokens.
Anthropic provide a detailed notebook demonstrating an implementation of this pattern. Their eventual solution combines cosine similarity and BM25 indexing, uses embeddings from Voyage AI and adds a reranking step powered by Cohere.
The notebook also includes an evaluation set using JSONL - here's that evaluation data in Datasette Lite.
OpenAI: Improve file search result relevance with chunk ranking (via) I've mostly been ignoring OpenAI's Assistants API. It provides an alternative to their standard messages API where you construct "assistants", chatbots with optional access to additional tools and that store full conversation threads on the server so you don't need to pass the previous conversation with every call to their API.
I'm pretty comfortable with their existing API and I found the assistants API to be quite a bit more complicated. So far the only thing I've used it for is a script to scrape OpenAI Code Interpreter to keep track of updates to their enviroment's Python packages.
Code Interpreter aside, the other interesting assistants feature is File Search. You can upload files in a wide variety of formats and OpenAI will chunk them, store the chunks in a vector store and make them available to help answer questions posed to your assistant - it's their version of hosted RAG.
Prior to today OpenAI had kept the details of how this worked undocumented. I found this infuriating, because when I'm building a RAG system the details of how files are chunked and scored for relevance is the whole game - without understanding that I can't make effective decisions about what kind of documents to use and how to build on top of the tool.
This has finally changed! You can now run a "step" (a round of conversation in the chat) and then retrieve details of exactly which chunks of the file were used in the response and how they were scored using the following incantation:
run_step = client.beta.threads.runs.steps.retrieve( thread_id="thread_abc123", run_id="run_abc123", step_id="step_abc123", include=[ "step_details.tool_calls[*].file_search.results[*].content" ] )
(See what I mean about the API being a little obtuse?)
I tried this out today and the results were very promising. Here's a chat transcript with an assistant I created against an old PDF copy of the Datasette documentation - I used the above new API to dump out the full list of snippets used to answer the question "tell me about ways to use spatialite".
It pulled in a lot of content! 57,017 characters by my count, spread across 20 search results (customizable), for a total of 15,021 tokens as measured by ttok. At current GPT-4o-mini prices that would cost 0.225 cents (less than a quarter of a cent), but with regular GPT-4o it would cost 7.5 cents.
OpenAI provide up to 1GB of vector storage for free, then charge $0.10/GB/day for vector storage beyond that. My 173 page PDF seems to have taken up 728KB after being chunked and stored, so that GB should stretch a pretty long way.
Confession: I couldn't be bothered to work through the OpenAI code examples myself, so I hit Ctrl+A on that web page and copied the whole lot into Claude 3.5 Sonnet, then prompted it:
Based on this documentation, write me a Python CLI app (using the Click CLi library) with the following features:
openai-file-chat add-files name-of-vector-store *.pdf *.txt
This creates a new vector store called name-of-vector-store and adds all the files passed to the command to that store.
openai-file-chat name-of-vector-store1 name-of-vector-store2 ...
This starts an interactive chat with the user, where any time they hit enter the question is answered by a chat assistant using the specified vector stores.
We iterated on this a few times to build me a one-off CLI app for trying out the new features. It's got a few bugs that I haven't fixed yet, but it was a very productive way of prototyping against the new API.
Using sqlite-vec with embeddings in sqlite-utils and Datasette. My notes on trying out Alex Garcia's newly released sqlite-vec SQLite extension, including how to use it with OpenAI embeddings in both Datasette and sqlite-utils.
Introducing sqlite-lembed: A SQLite extension for generating text embeddings locally (via) Alex Garcia's latest SQLite extension is a C wrapper around the llama.cpp that exposes just its embedding support, allowing you to register a GGUF file containing an embedding model:
INSERT INTO temp.lembed_models(name, model)
select 'all-MiniLM-L6-v2',
lembed_model_from_file('all-MiniLM-L6-v2.e4ce9877.q8_0.gguf');
And then use it to calculate embeddings as part of a SQL query:
select lembed(
'all-MiniLM-L6-v2',
'The United States Postal Service is an independent agency...'
); -- X'A402...09C3' (1536 bytes)
all-MiniLM-L6-v2.e4ce9877.q8_0.gguf here is a 24MB file, so this should run quite happily even on machines without much available RAM.
What if you don't want to run the models locally at all? Alex has another new extension for that, described in Introducing sqlite-rembed: A SQLite extension for generating text embeddings from remote APIs. The rembed is for remote embeddings, and this extension uses Rust to call multiple remotely-hosted embeddings APIs, registered like this:
INSERT INTO temp.rembed_clients(name, options)
VALUES ('text-embedding-3-small', 'openai');
select rembed(
'text-embedding-3-small',
'The United States Postal Service is an independent agency...'
); -- X'A452...01FC', Blob<6144 bytes>
Here's the Rust code that implements Rust wrapper functions for HTTP JSON APIs from OpenAI, Nomic, Cohere, Jina, Mixedbread and localhost servers provided by Ollama and Llamafile.
Both of these extensions are designed to complement Alex's sqlite-vec extension, which is nearing a first stable release.
Searching an aerial photo with text queries. Robin Wilson built a demo that lets you search a large aerial photograph of Southampton for things like "roundabout" or "tennis court". He explains how it works in detail: he used the SkyCLIP model, which is trained on "5.2 million remote sensing image-text pairs in total, covering more than 29K distinct semantic tags" to generate embeddings for 200x200 image segments (with 100px of overlap), then stored them in Pinecone.
The Super Effectiveness of Pokémon Embeddings Using Only Raw JSON and Images.
A deep dive into embeddings from Max Woolf, exploring 1,000 different Pokémon (loaded from PokéAPI using this epic GraphQL query) and then embedding the cleaned up JSON data using nomic-embed-text-v1.5 and the official Pokémon image representations using nomic-embed-vision-v1.5.
I hadn't seen nomic-embed-vision-v1.5 before: it brings multimodality to Nomic embeddings and operates in the same embedding space as nomic-embed-text-v1.5 which means you can use it to perform CLIP-style tricks comparing text and images. Here's their announcement from June 5th:
Together, Nomic Embed is the only unified embedding space that outperforms OpenAI CLIP and OpenAI Text Embedding 3 Small on multimodal and text tasks respectively.
Sadly the new vision weights are available under a non-commercial Creative Commons license (unlike the text weights which are Apache 2), so if you want to use the vision weights commercially you'll need to access them via Nomic's paid API.
Nomic do say this though:
As Nomic releases future models, we intend to re-license less recent models in our catalogue under the Apache-2.0 license.
Val Vibes: Semantic search in Val Town. A neat case-study by JP Posma on how Val Town's developers can use Val Town Vals to build prototypes of new features that later make it into Val Town core.
This one explores building out semantic search against Vals using OpenAI embeddings and the PostgreSQL pgvector extension.
Using DuckDB for Embeddings and Vector Search (via) Sören Brunk's comprehensive tutorial combining DuckDB 1.0, a subset of German Wikipedia from Hugging Face (loaded using Parquet), the BGE M3 embedding model and DuckDB's new vss extension for implementing an HNSW vector index.
My Twitter thread figuring out the AI features in Microsoft’s Recall. I posed this question on Twitter about why Microsoft Recall (previously) is being described as "AI":
Is it just that the OCR uses a machine learning model, or are there other AI components in the mix here?
I learned that Recall works by taking full desktop screenshots and then applying both OCR and some sort of CLIP-style embeddings model to their content. Both the OCRd text and the vector embeddings are stored in SQLite databases (schema here, thanks Daniel Feldman) which can then be used to search your past computer activity both by text but also by semantic vision terms - "blue dress" to find blue dresses in screenshots, for example. The si_diskann_graph table names hint at Microsoft's DiskANN vector indexing library
A Microsoft engineer confirmed on Hacker News that Recall uses on-disk vector databases to provide local semantic search for both text and images, and that they aren't using Microsoft's Phi-3 or Phi-3 Vision models. As far as I can tell there's no LLM used by the Recall system at all at the moment, just embeddings.
Exploring Hacker News by mapping and analyzing 40 million posts and comments for fun
(via)
A real tour de force of data engineering. Wilson Lin fetched 40 million posts and comments from the Hacker News API (using Node.js with a custom multi-process worker pool) and then ran them all through the BGE-M3 embedding model using RunPod, which let him fire up ~150 GPU instances to get the whole run done in a few hours, using a custom RocksDB and Rust queue he built to save on Amazon SQS costs.
Then he crawled 4 million linked pages, embedded that content using the faster and cheaper jina-embeddings-v2-small-en model, ran UMAP dimensionality reduction to render a 2D map and did a whole lot of follow-on work to identify topic areas and make the map look good.
That's not even half the project - Wilson built several interactive features on top of the resulting data, and experimented with custom rendering techniques on top of canvas to get everything to render quickly.
There's so much in here, and both the code and data (multiple GBs of arrow files) are available if you want to dig in and try some of this out for yourself.
In the Hacker News comments Wilson shares that the total cost of the project was a couple of hundred dollars.
One tiny detail I particularly enjoyed - unrelated to the embeddings - was this trick for testing which edge location is closest to a user using JavaScript:
const edge = await Promise.race(
EDGES.map(async (edge) => {
// Run a few times to avoid potential cold start biases.
for (let i = 0; i < 3; i++) {
await fetch(`https://${edge}.edge-hndr.wilsonl.in/healthz`);
}
return edge;
}),
);
I’m writing a new vector search SQLite Extension.
Alex Garcia is working on sqlite-vec, a spiritual successor to his sqlite-vss project. The new SQLite C extension will have zero other dependencies (sqlite-vss used some tricky C++ libraries) and will work using virtual tables, storing chunks of vectors in shadow tables to avoid needing to load everything into memory at once.
llm-nomic-api-embed. My new plugin for LLM which adds API access to the Nomic series of embedding models. Nomic models can be run locally too, which makes them a great long-term commitment as there’s no risk of the models being retired in a way that damages the value of your previously calculated embedding vectors.
Cohere int8 & binary Embeddings—Scale Your Vector Database to Large Datasets (via) Jo Kristian Bergum told me “The accuracy retention [of binary embedding vectors] is sensitive to whether the model has been using this binarization as part of the loss function.”
Cohere provide an API for embeddings, and last week added support for returning binary vectors specifically tuned in this way.
250M embeddings (Cohere provide a downloadable dataset of 250M embedded documents from Wikipedia) at float32 (4 bytes) is 954GB.
Cohere claim that reducing to 1 bit per dimension knocks that down to 30 GB (954/32) while keeping “90-98% of the original search quality”.
My binary vector search is better than your FP32 vectors. I’m still trying to get my head around this, but here’s what I understand so far.
Embedding vectors as calculated by models such as OpenAI text-embedding-3-small are arrays of floating point values, which look something like this:
[0.0051681744, 0.017187592, -0.018685209, -0.01855924, -0.04725188...]—1356 elements long
Different embedding models have different lengths, but they tend to be hundreds up to low thousands of numbers. If each float is 32 bits that’s 4 bytes per float, which can add up to a lot of memory if you have millions of embedding vectors to compare.
If you look at those numbers you’ll note that they are all pretty small positive or negative numbers, close to 0.
Binary vector search is a trick where you take that sequence of floating point numbers and turn it into a binary vector—just a list of 1s and 0s, where you store a 1 if the corresponding float was greater than 0 and a 0 otherwise.
For the above example, this would start [1, 1, 0, 0, 0...]
Incredibly, it looks like the cosine distance between these 0 and 1 vectors captures much of the semantic relevant meaning present in the distance between the much more accurate vectors. This means you can use 1/32nd of the space and still get useful results!
Ce Gao here suggests a further optimization: use the binary vectors for a fast brute-force lookup of the top 200 matches, then run a more expensive re-ranking against those filtered values using the full floating point vectors.
Adaptive Retrieval with Matryoshka Embeddings (via) Nomic Embed v1 only came out two weeks ago, but the same team just released Nomic Embed v1.5 trained using a new technique called Matryoshka Representation.
This means that unlike v1 the v1.5 embeddings are resizable—instead of a fixed 768 dimension embedding vector you can trade size for quality and drop that size all the way down to 64, while still maintaining strong semantically relevant results.
Joshua Lochner build this interactive demo on top of Transformers.js which illustrates quite how well this works: it lets you embed a query, embed a series of potentially matching text sentences and then adjust the number of dimensions and see what impact it has on the results.
Announcing DuckDB 0.10.0.
Somewhat buried in this announcement: DuckDB has Fixed-Length Arrays now, along with array_cross_product(a1, a2), array_cosine_similarity(a1, a2) and array_inner_product(a1, a2) functions.
This means you can now use DuckDB to find related content (and other tricks) using vector embeddings!
Also notable:
DuckDB can now attach MySQL, Postgres, and SQLite databases in addition to databases stored in its own format. This allows data to be read into DuckDB and moved between these systems in a convenient manner, as attached databases are fully functional, appear just as regular tables, and can be updated in a safe, transactional manner.
llm-sentence-transformers 0.2. I added a new --trust-remote-code option when registering an embedding model, which means LLM can now run embeddings through the new Nomic AI nomic-embed-text-v1 model.
Introducing Nomic Embed: A Truly Open Embedding Model. A new text embedding model from Nomic AI which supports 8192 length sequences, claims better scores than many other models (including OpenAI’s new text-embedding-3-small) and is available as both a hosted API and a run-yourself model. The model is Apache 2 licensed and Nomic have released the full set of training data and code.
From the accompanying paper: “Full training of nomic-embed-text-v1 can be conducted in a single week on one 8xH100 node.”
ChunkViz (via) Handy tool by Greg Kamradt to help understand how different text chunking mechanisms work by visualizing them. Chunking is an important part of preparing text to be embedded for semantic search, and thanks to this tool I’ve finally got a solid mental model of what recursive character text splitting does.
ColBERT query-passage scoring interpretability (via) Neat interactive visualization tool for understanding what the ColBERT embedding model does—this works by loading around 50MB of model files directly into your browser and running them with WebAssembly.
Text Embeddings Reveal (Almost) As Much As Text. Embeddings of text—where a text string is converted into a fixed-number length array of floating point numbers—are demonstrably reversible: “a multi-step method that iteratively corrects and re-embeds text is able to recover 92% of 32-token text inputs exactly”.
This means that if you’re using a vector database for embeddings of private data you need to treat those embedding vectors with the same level of protection as the original text.
2023
Fleet Context. This project took the source code and documentation for 1221 popular Python libraries and ran them through the OpenAI text-embedding-ada-002 embedding model, then made those pre-calculated embedding vectors available as Parquet files for download from S3 or via a custom Python CLI tool.
I haven’t seen many projects release pre-calculated embeddings like this, it’s an interesting initiative.
Execute Jina embeddings with a CLI using llm-embed-jina
Berlin-based Jina AI just released a new family of embedding models, boasting that they are the “world’s first open-source 8K text embedding model” and that they rival OpenAI’s text-embedding-ada-002 in quality.