Skip to content

Minor improvements in blog posts #724

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jun 12, 2023
Merged

Minor improvements in blog posts #724

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 7 additions & 5 deletions ...content/blog/generating-llm-embeddings-with-open-source-models-in-postgresml.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -127,8 +127,10 @@ Since our corpus of documents (movie reviews) are all relatively short and simil

It takes a couple of minutes to download and cache the `intfloat/e5-small` model to generate the first embedding. After that, it's pretty fast.

Note how we prefix the text we want to embed with either `passage: ` or `query: `, the e5 model requires us to prefix our data with `passage: ` if we're generating embeddings for our corpus and `query: ` if we want to find semantically similar content.

```postgresql
SELECT pgml.embed('intfloat/e5-small', 'hi mom');
SELECT pgml.embed('intfloat/e5-small', 'passage: hi mom');
```

This is a pretty powerful function, because we can pass any arbitrary text to any open source model, and it will generate an embedding for us. We can benchmark how long it takes to generate an embedding for a single review, using client-side timings in Postgres:
Expand All @@ -147,7 +149,7 @@ Aside from using this function with strings passed from a client, we can use it
```postgresql
SELECT
review_body,
pgml.embed('intfloat/e5-small', review_body)
pgml.embed('intfloat/e5-small', 'passage: ' || review_body)
FROM pgml.amazon_us_reviews
LIMIT 1;
```
Expand All @@ -171,7 +173,7 @@ Time to generate an embedding increases with the length of the input text, and v
```postgresql
SELECT
review_body,
pgml.embed('intfloat/e5-small', review_body) AS embedding
pgml.embed('intfloat/e5-small', 'passage: ' || review_body) AS embedding
FROM pgml.amazon_us_reviews
LIMIT 1000;
```
Expand All @@ -191,7 +193,7 @@ SELECT
reviqew_body,
pgml.embed(
'intfloat/e5-small',
review_body,
'passage: ' || review_body,
'{"device": "cpu"}'
) AS embedding
FROM pgml.amazon_us_reviews
Expand Down Expand Up @@ -328,7 +330,7 @@ BEGIN
UPDATE pgml.amazon_us_reviews
SET review_embedding_e5_large = pgml.embed(
'intfloat/e5-large',
review_body
'passage: ' || review_body
)
WHERE id BETWEEN i AND i + 10
AND review_embedding_e5_large IS NULL;
Expand Down
28 changes: 14 additions & 14 deletions ...og/personalize-embedding-vector-search-results-with-huggingface-and-pgvector.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -137,7 +137,7 @@ We can find a customer that our embeddings model feels is close to the sentiment
WITH request AS (
SELECT pgml.embed(
'intfloat/e5-large',
'I love all Star Wars, but Empire Strikes Back is particularly amazing'
'query: I love all Star Wars, but Empire Strikes Back is particularly amazing'
)::vector(1024) AS embedding
)

Expand All @@ -147,17 +147,17 @@ SELECT
star_rating_avg,
1 - (
movie_embedding_e5_large <=> (SELECT embedding FROM request)
) AS cosine_similiarity
) AS cosine_similarity
FROM customers
ORDER BY cosine_similiarity DESC
ORDER BY cosine_similarity DESC
LIMIT 1;
```

!!!

!!! results

| id | total_reviews | star_rating_avg | cosine_similiarity |
| id | total_reviews | star_rating_avg | cosine_similarity |
|----------|---------------|--------------------|--------------------|
| 44366773 | 1 | 2.0000000000000000 | 0.8831349398621555 |

Expand Down Expand Up @@ -215,7 +215,7 @@ Now we can write our personalized SQL query. It's nearly the same as our query f
WITH request AS (
SELECT pgml.embed(
'intfloat/e5-large',
'Best 1980''s scifi movie'
'query: Best 1980''s scifi movie'
)::vector(1024) AS embedding
),

Expand All @@ -226,18 +226,18 @@ customer AS (
WHERE id = '44366773'
),

-- vector similarity search for movies and calculate a customer_cosine_similiarity at the same time
-- vector similarity search for movies and calculate a customer_cosine_similarity at the same time
first_pass AS (
SELECT
title,
total_reviews,
star_rating_avg,
1 - (
review_embedding_e5_large <=> (SELECT embedding FROM request)
) AS request_cosine_similiarity,
) AS request_cosine_similarity,
(1 - (
review_embedding_e5_large <=> (SELECT embedding FROM customer)
) - 0.9) * 10 AS customer_cosine_similiarity,
) - 0.9) * 10 AS customer_cosine_similarity,
star_rating_avg / 5 AS star_rating_score
FROM movies
WHERE total_reviews > 10
Expand All @@ -251,9 +251,9 @@ SELECT
total_reviews,
round(star_rating_avg, 2) as star_rating_avg,
star_rating_score,
request_cosine_similiarity,
customer_cosine_similiarity,
request_cosine_similiarity + customer_cosine_similiarity + star_rating_score AS final_score
request_cosine_similarity,
customer_cosine_similarity,
request_cosine_similarity + customer_cosine_similarity + star_rating_score AS final_score
FROM first_pass
ORDER BY final_score DESC
LIMIT 10;
Expand All @@ -263,7 +263,7 @@ LIMIT 10;

!!! results

| title | total_reviews | star_rating_avg | star_rating_score | request_cosine_similiarity | customer_cosine_similiarity | final_score |
| title | total_reviews | star_rating_avg | star_rating_score | request_cosine_similarity | customer_cosine_similarity | final_score |
|----------------------------------------------------------------------|---------------|-----------------|------------------------|----------------------------|-----------------------------|--------------------|
| Star Wars, Episode V: The Empire Strikes Back (Widescreen Edition) | 78 | 4.44 | 0.88717948717948718000 | 0.8295302273865711 | 0.9999999999999998 | 2.716709714566058 |
| Star Wars, Episode IV: A New Hope (Widescreen Edition) | 80 | 4.36 | 0.87250000000000000000 | 0.8339361274771777 | 0.9336656923446551 | 2.640101819821833 |
Expand All @@ -280,15 +280,15 @@ LIMIT 10;

!!!

Bingo. Now we're boosting movies by `(customer_cosine_similiarity - 0.9) * 10`, and we've kept our previous boost for movies with a high average star rating. Not only does Episode V top the list as expected, Episode IV is a close second. This query has gotten fairly complex! But the results are perfect for me, I mean our hypothetical customer who is searching for "Best 1980's scifi movie" but has already revealed to us with their one movie review that they think like the comment "I love all Star Wars, but Empire Strikes Back is particularly amazing". I promise I'm not just doing all of this to find a new movie to watch tonight.
Bingo. Now we're boosting movies by `(customer_cosine_similarity - 0.9) * 10`, and we've kept our previous boost for movies with a high average star rating. Not only does Episode V top the list as expected, Episode IV is a close second. This query has gotten fairly complex! But the results are perfect for me, I mean our hypothetical customer who is searching for "Best 1980's scifi movie" but has already revealed to us with their one movie review that they think like the comment "I love all Star Wars, but Empire Strikes Back is particularly amazing". I promise I'm not just doing all of this to find a new movie to watch tonight.

You can compare this to our non-personalized results from the previous article for reference Forbidden Planet used to be the top result, but now it's #3.

!!! code_block time="124.119 ms"

!!! results

| title | total_reviews | star_rating_avg | final_score | star_rating_score | cosine_similiarity |
| title | total_reviews | star_rating_avg | final_score | star_rating_score | cosine_similarity |
|:-----------------------------------------------------|--------------:|----------------:|-------------------:|-----------------------:|-------------------:|
| Forbidden Planet (Two-Disc 50th Anniversary Edition) | 255 | 4.82 | 1.8216832158805154 | 0.96392156862745098000 | 0.8577616472530644 |
| Back to the Future | 31 | 4.94 | 1.82090702765472 | 0.98709677419354838000 | 0.8338102534611714 |
Expand Down
32 changes: 16 additions & 16 deletions .../blog/tuning-vector-recall-while-generating-query-embeddings-in-the-database.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -129,7 +129,7 @@ We'll start with semantic search. Given a user query, e.g. "Best 1980's scifi mo
WITH request AS (
SELECT pgml.embed(
'intfloat/e5-large',
'Best 1980''s scifi movie'
'query: Best 1980''s scifi movie'
)::vector(1024) AS embedding
)

Expand All @@ -142,17 +142,17 @@ SELECT
review_embedding_e5_large <=> (
SELECT embedding FROM request
)
) AS cosine_similiarity
) AS cosine_similarity
FROM pgml.amazon_us_reviews
ORDER BY review_embedding_e5_large <=> (SELECT embedding FROM request)
ORDER BY cosine_similarity
LIMIT 5;
```

!!!

!!! results

| review_body | product_title | star_rating | total_votes | cosine_similiarity |
| review_body | product_title | star_rating | total_votes | cosine_similarity |
|-----------------------------------------------------|---------------------------------------------------------------|-------------|-------------|--------------------|
| best 80s SciFi movie ever | The Adventures of Buckaroo Banzai Across the Eighth Dimension | 5 | 1 | 0.956207707312679 |
| One of the best 80's sci-fi movies, beyond a doubt! | Close Encounters of the Third Kind [Blu-ray] | 5 | 1 | 0.9298004258989776 |
Expand Down Expand Up @@ -270,7 +270,7 @@ SELECT
title,
1 - (
review_embedding_e5_large <=> (SELECT embedding FROM request)
) AS cosine_similiarity
) AS cosine_similarity
FROM movies
ORDER BY review_embedding_e5_large <=> (SELECT embedding FROM request)
LIMIT 10;
Expand All @@ -280,7 +280,7 @@ LIMIT 10;

!!! results

| title | cosine_similiarity |
| title | cosine_similarity |
|--------------------------------------------------------------------|--------------------|
| THX 1138 (The George Lucas Director's Cut Special Edition/ 2-Disc) | 0.8652007733744973 |
| 2010: The Year We Make Contact | 0.8621574666546908 |
Expand Down Expand Up @@ -328,7 +328,7 @@ SELECT
title,
1 - (
review_embedding_e5_large <=> (SELECT embedding FROM request)
) AS cosine_similiarity
) AS cosine_similarity
FROM movies
ORDER BY review_embedding_e5_large <=> (SELECT embedding FROM request)
LIMIT 10;
Expand All @@ -338,7 +338,7 @@ LIMIT 10;

!!! results

| title | cosine_similiarity |
| title | cosine_similarity |
|--------------------------------------------------------------------|--------------------|
| THX 1138 (The George Lucas Director's Cut Special Edition/ 2-Disc) | 0.8652007733744973 |
| Big Trouble in Little China [UMD for PSP] | 0.8649691870870362 |
Expand Down Expand Up @@ -411,7 +411,7 @@ SET ivfflat.probes = 1;
WITH request AS (
SELECT pgml.embed(
'intfloat/e5-large',
'Best 1980''s scifi movie'
'query: Best 1980''s scifi movie'
)::vector(1024) AS embedding
)

Expand All @@ -420,7 +420,7 @@ SELECT
total_reviews,
1 - (
review_embedding_e5_large <=> (SELECT embedding FROM request)
) AS cosine_similiarity
) AS cosine_similarity
FROM movies
WHERE <strong>total_reviews > 10</strong>
ORDER BY review_embedding_e5_large <=> (SELECT embedding FROM request)
Expand All @@ -431,7 +431,7 @@ LIMIT 10;

!!! results

| title | total_reviews | cosine_similiarity |
| title | total_reviews | cosine_similarity |
|------------------------------------------------------|---------------|--------------------|
| 2010: The Year We Make Contact | 29 | 0.8621574666546908 |
| Forbidden Planet | 202 | 0.861032948199611 |
Expand Down Expand Up @@ -467,7 +467,7 @@ SQL is a very expressive language that can handle a lot of complexity. To keep t
WITH request AS (
SELECT pgml.embed(
'intfloat/e5-large',
'Best 1980''s scifi movie'
'query: Best 1980''s scifi movie'
)::vector(1024) AS embedding
),

Expand All @@ -479,7 +479,7 @@ first_pass AS (
star_rating_avg,
1 - (
review_embedding_e5_large <=> (SELECT embedding FROM request)
) AS cosine_similiarity,
) AS cosine_similarity,
star_rating_avg / 5 AS star_rating_score
FROM movies
WHERE total_reviews > 10
Expand All @@ -493,8 +493,8 @@ SELECT
total_reviews,
round(star_rating_avg, 2) as star_rating_avg,
star_rating_score,
cosine_similiarity,
cosine_similiarity + star_rating_score AS final_score
cosine_similarity,
cosine_similarity + star_rating_score AS final_score
FROM first_pass
ORDER BY final_score DESC
LIMIT 10;
Expand All @@ -504,7 +504,7 @@ LIMIT 10;

!!! results

| title | total_reviews | star_rating_avg | final_score | star_rating_score | cosine_similiarity |
| title | total_reviews | star_rating_avg | final_score | star_rating_score | cosine_similarity |
|:-----------------------------------------------------|--------------:|----------------:|-------------------:|-----------------------:|-------------------:|
| Forbidden Planet (Two-Disc 50th Anniversary Edition) | 255 | 4.82 | 1.8216832158805154 | 0.96392156862745098000 | 0.8577616472530644 |
| Back to the Future | 31 | 4.94 | 1.82090702765472 | 0.98709677419354838000 | 0.8338102534611714 |
Expand Down
pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy