H-3614: Introduce `chonky` PDF Embeddings #5673

JesusFileto · 2024-11-20T06:41:07Z

🌟 What is the purpose of this PR?

This PR outlines the work for the embeddings that will be done in the PDF and preliminary structs with how they will be stored. The main goal is to split embeddings into multiple levels of textual information such as table embeddings. Future PRs will focus on implementing the XML metadata that will accompany these embeddings to be used for entity extraction.

🚫 Blocked by

Requires Setup of HASH API keys for Google Vertex API and including the HuggingFaceToken

...

🔍 What does this change?

...

Pre-Merge Checklist 🚀

🚢 Has this modified a publishable library?

This PR:

does not modify any publishable blocks or libraries, or modifications do not need publishing

📜 Does this require a change to the docs?

The changes in this PR:

require changes to docs which are not made in this PR
- Further modifications of current format for API calls and setup may change

🕸️ Does this require a change to the Turbo Graph?

The changes in this PR:

do not affect the execution graph

⚠️ Known issues

When first running the API calls for HuggingFace, the model is initially "cold" and takes 20 seconds to boot up the model for use. Will be implementing a redrive mechanism for this scenario later.

🐾 Next steps

Add XML Metadata from the pdf to further enrich the embeddings
Investigate how we wish to store all this data

🛡 What tests cover this?

New tests were added

TimDiekmann

Thanks @JesusFileto!
I noticed that you have used curl which means there is no asynchronous execution of requests possible. In particular for AI calls, asynchronous calls are preferred. In addition, curl pulls in libcurl (and OpenSSL), while reqwest is written in Rust and can use rustls.

As async backend we use tokio. If you add the "macros" and "rt-multi-thread" to tokio you can use #[tokio::main] on fn main() and #[tokio::test] instead of #[test] which makes the function async.
As this effectively makes the code async we can also use tokio to read files (by using the fs feature on tokio and tokio::fs instead of std::fs)

Cargo.toml

libs/chonky/src/embedding.rs

codecov · 2024-11-26T17:49:26Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 21.72%. Comparing base (76d9ea5) to head (2d1a4a1).
Report is 139 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #5673   +/-   ##
=======================================
  Coverage   21.72%   21.72%           
=======================================
  Files         566      566           
  Lines       19156    19156           
  Branches     2755     2755           
=======================================
  Hits         4162     4162           
  Misses      14942    14942           
  Partials       52       52

Flag	Coverage Δ
apps.hash-ai-worker-ts	`1.32% <ø> (ø)`
apps.hash-api	`1.16% <ø> (ø)`
blockprotocol.type-system	`46.42% <ø> (ø)`
local.harpc-client	`50.17% <ø> (ø)`
local.hash-backend-utils	`8.81% <ø> (ø)`
local.hash-graph-sdk	`58.62% <ø> (ø)`
local.hash-isomorphic-utils	`0.98% <ø> (ø)`
local.hash-subgraph	`24.54% <ø> (ø)`
rust.deer	`6.66% <ø> (ø)`
rust.error-stack	`72.51% <ø> (ø)`
rust.sarif	`88.66% <ø> (ø)`

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

TimDiekmann

Great work moving to async 🎉
I did a more in-depth review. I think it's a good time to improve the code quality a bit 🙂

libs/chonky/src/embedding/hugging_face_api.rs

TimDiekmann · 2024-11-28T09:56:46Z

libs/chonky/src/embedding/multi_modal_embedding.rs

+    .to_owned())
+}
+
+fn base64_json(image_data: Vec<u8>) -> Result<String, Report<ChonkyError>> {


It's not needed to pass a vector, a slice is enough

Suggested change

fn base64_json(image_data: Vec<u8>) -> Result<String, Report<ChonkyError>> {

fn base64_json(image_data: &[u8]) -> Result<String, Report<ChonkyError>> {

If you want to be even more generic you can use the same approach as above:

Suggested change

fn base64_json(image_data: Vec<u8>) -> Result<String, Report<ChonkyError>> {

fn base64_json(image_data: impl AsRef<[u8]>) -> Result<String, Report<ChonkyError>> {

I see, will add this in next PR.

The second suggestion is a more flexible way without side effects (the calling site doesn't need to be updated).
However, you can change the caller to

- payload = base64_json(image_payload); + payload = base64_json(&image_payload);

libs/chonky/src/embedding/multi_modal_embedding.rs

TimDiekmann · 2024-11-28T10:49:58Z

libs/chonky/src/embedding/multi_modal_embedding.rs

+}
+
+// Parses the response to extract the image embedding vector
+fn extract_embedding(response_text: &str) -> Result<Vec<f64>, Report<ChonkyError>> {


If we have a struct for embedding, such as TextEmbedding and ImageEmbedding we probably want to return an enum here

enum Embedding { Text(TextEmbedding), Image(ImageEmbedding), }

Was implementing this but the ImageEmbedding struct requires the image in the struct, which is not accessible when just requesting the embedding, since the calling function should have knowledge on the type of vector it is receiving, it may be easier to not implement this (to avoid having to handling if let branches upstream where the compiler does not know what datatype we are expecting but we do)

libs/chonky/src/embedding/multi_modal_embedding.rs

Cargo.toml

TimDiekmann

I had a re-review and the changes look very good. I added a few more comments.
Also, let's make sure to consistently use the new Embedding structs as they really contributes to the readability of the code.

TimDiekmann · 2024-12-04T11:32:56Z

libs/chonky/src/lib.rs

+                let xmin: i32 = num_traits::cast(bbox.xmin).ok_or(ChonkyError::Pdfium)?;
+                let ymin: i32 = num_traits::cast(bbox.ymin).ok_or(ChonkyError::Pdfium)?;
+                let xmax: i32 = num_traits::cast(bbox.xmax).ok_or(ChonkyError::Pdfium)?;
+                let ymax: i32 = num_traits::cast(bbox.ymax).ok_or(ChonkyError::Pdfium)?;


What is the exact behavior here we expect? Because num_traits::cast simply discards the decimals which is the behavior of floor. We should add to the comment that flooring it is the expected behavior.

TimDiekmann · 2024-12-04T11:33:50Z

libs/chonky/src/lib.rs

+pub mod embedding;
+
+pub use embedding::{hugging_face_api, multi_modal_embedding};


We don't need both to be pub. Either is sufficient.

TimDiekmann · 2024-12-04T11:36:10Z

libs/chonky/src/lib.rs

+
+#[derive(Debug, Clone)]
+pub struct Embedding {
+    _model_used: String,            //model name reveals image or text embedding model


It's often the case that the model is statically non

Suggested change

_model_used: String, //model name reveals image or text embedding model

_model_used: Cow<'static, str>, //model name reveals image or text embedding model

Also, is there a specific reason why you hide this detail and not making this pub?

The only reason the field is not pub is since it is not used yet so clippy complains.

libs/chonky/src/embedding/multi_modal_embedding.rs

TimDiekmann · 2024-12-04T11:55:47Z

libs/chonky/src/embedding/multi_modal_embedding.rs

+pub async fn embed_pdf_object_images(
+    pdf_image_extract: Vec<Vec<DynamicImage>>,
+    project_id: &str,
+) -> Result<Vec<Vec<Vec<f64>>>, Report<ChonkyError>> {


I definitely see the improvement already. I'll keep this comment open for the Box<[f64]> approach.

TimDiekmann · 2024-12-04T12:21:40Z

libs/chonky/src/embedding/multi_modal_embedding.rs

+///
+/// [`ChonkyError::VertexAPI`] when there are HTTP request errors
+pub async fn embed_screenshots(
+    pdf_image_extract: Vec<DynamicImage>,


The actual idiomatic version (but also quite hard to start with) Is using streams.
This way the function would become

pub fn embed_pdf_object_images( pdf_image_extract: impl IntoIterator<Item = PageImageObjects>, project_id: &str, ) -> impl Stream<Item = Result<PageImageObjectsEmbeddings, Report<ChonkyError>>> { stream::iter(pdf_image_extract) .then(|page_images| embed_screenshots(page_images, project_id)) .map(|embeddings| { Ok(PageImageObjectsEmbeddings { _embeddings: embeddings?, }) }) }

Note, that this function is not async, because it returns a Stream (which is pretty much an async iterator).
I don't say we need to take this route because streams are not yet well matured in Rust (and a more built-in feature is on the way), but it avoids allocations.

TimDiekmann · 2024-12-04T12:22:22Z

libs/chonky/src/embedding/multi_modal_embedding.rs

+pub async fn embed_tables(
+    pdf_table_bounds: Vec<Vec<ExtractedTable>>,
+    project_id: &str,
+) -> Result<Vec<Vec<Vec<f64>>>, Report<ChonkyError>> {


Actually, we still have this signature 😅

libs/chonky/src/embedding/hugging_face_api.rs

TimDiekmann · 2024-12-04T12:39:18Z

libs/chonky/src/embedding/hugging_face_api.rs

+async fn get_binary_image_data(
+    image_path: impl AsRef<Path> + core::marker::Send + core::marker::Sync,
+) -> Result<Vec<u8>, Report<ChonkyError>> {
+    fs::read(image_path)
+        .await
+        .change_context(ChonkyError::ImageError)
+}


This really doesn't more than reading a file. We should probably inline it. Also, the error ImageError seems to be inaccurate as this is really a file operation. (Btw, tokio::fs::read only wraps the reading operation in an another task, nothing else).

If we keep it, we should use Send only and directly:

Suggested change

async fn get_binary_image_data(

image_path: impl AsRef<Path> + core::marker::Send + core::marker::Sync,

) -> Result<Vec<u8>, Report<ChonkyError>> {

fs::read(image_path)

.await

.change_context(ChonkyError::ImageError)

}

async fn get_binary_image_data(

image_path: impl AsRef<Path> + Send,

) -> Result<Vec<u8>, Report<ChonkyError>> {

fs::read(image_path)

.await

.change_context(ChonkyError::ImageError)

}

In trying to move the function inline, found that the explicit references to Send + Sync bounds avoid compiler errors when using references to image_path further downstream when retrying the request if the hugging face model was initially cold. For this reason it may be wise to not inline this function.

TimDiekmann · 2024-12-04T12:41:38Z

libs/chonky/src/embedding/multi_modal_embedding.rs

+    .to_owned())
+}
+
+fn base64_json(image_data: Vec<u8>) -> Result<String, Report<ChonkyError>> {


The second suggestion is a more flexible way without side effects (the calling site doesn't need to be updated).
However, you can change the caller to

- payload = base64_json(image_payload); + payload = base64_json(&image_payload);

… `chonky` tests (#5687) Co-authored-by: JesusFileto <91223391+JesusFileto@users.noreply.github.com> Co-authored-by: Jesus Fileto <jesus_filetojr@icloud.com>

TimDiekmann · 2024-12-05T10:24:19Z

libs/chonky/src/embedding/multi_modal_embedding.rs

+}
+
+/// A function that performs authentication with Google Vertex API and performs
+/// a curl request to obtain multimodal embeddings given an image path


Suggested change

/// a curl request to obtain multimodal embeddings given an image path

/// a request to obtain multimodal embeddings given an image path

libs/chonky/src/embedding/multi_modal_embedding.rs

Cargo.toml

TimDiekmann · 2024-12-12T18:53:46Z

Cargo.lock

+[[package]]
+name = "curl"
+version = "0.4.47"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d9fb4d13a1be2b58f14d60adba57c9834b78c62fd86c3e76a148f732686e9265"
+dependencies = [
+ "curl-sys",
+ "libc",
+ "openssl-probe",
+ "openssl-sys",
+ "schannel",
+ "socket2",
+ "windows-sys 0.52.0",
+]


It looks like you need to update this file, it seems outdated

TimDiekmann · 2024-12-12T18:58:00Z

libs/chonky/src/snapshots/google_test_embedding_page_1.json

Could you add json files to .gitattributes as well?

vilkinsons · 2025-01-10T12:23:00Z

Tagging @TimDiekmann in to get this PR over the line as we'd like to unblock merge. Thanks @JesusFileto for your work in getting it this far!

# Conflicts: # .markdownlintignore # Cargo.lock # Cargo.toml

github-actions · 2025-01-13T09:34:27Z

Benchmark results

@rust/hash-graph-benches – Integrations

representative_read_entity

Function	Value	Mean	Flame graphs
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/song/v/1`	$$16.8 \mathrm{ms} \pm 222 \mathrm{μs}\left({\color{lightgreen}-32.196 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/organization/v/1`	$$16.2 \mathrm{ms} \pm 198 \mathrm{μs}\left({\color{lightgreen}-6.706 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/building/v/1`	$$17.5 \mathrm{ms} \pm 175 \mathrm{μs}\left({\color{gray}-1.654 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/playlist/v/1`	$$16.8 \mathrm{ms} \pm 218 \mathrm{μs}\left({\color{gray}-2.193 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/uk-address/v/1`	$$16.1 \mathrm{ms} \pm 176 \mathrm{μs}\left({\color{lightgreen}-8.209 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/book/v/1`	$$15.9 \mathrm{ms} \pm 178 \mathrm{μs}\left({\color{lightgreen}-7.216 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/block/v/1`	$$15.7 \mathrm{ms} \pm 167 \mathrm{μs}\left({\color{lightgreen}-13.764 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/person/v/1`	$$17.0 \mathrm{ms} \pm 208 \mathrm{μs}\left({\color{lightgreen}-31.089 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	entity type ID: `https://blockprotocol.org/@alice/types/entity-type/page/v/2`	$$17.3 \mathrm{ms} \pm 271 \mathrm{μs}\left({\color{lightgreen}-8.024 \mathrm{\%}}\right) $$	Flame Graph

representative_read_multiple_entities

Function	Value	Mean	Flame graphs
entity_by_property	depths: DT=255, PT=255, ET=255, E=255	$$66.5 \mathrm{ms} \pm 363 \mathrm{μs}\left({\color{gray}-1.459 \mathrm{\%}}\right) $$	Flame Graph
entity_by_property	depths: DT=0, PT=0, ET=0, E=0	$$37.9 \mathrm{ms} \pm 152 \mathrm{μs}\left({\color{lightgreen}-8.109 \mathrm{\%}}\right) $$	Flame Graph
entity_by_property	depths: DT=2, PT=2, ET=2, E=2	$$56.3 \mathrm{ms} \pm 278 \mathrm{μs}\left({\color{gray}-3.743 \mathrm{\%}}\right) $$	Flame Graph
entity_by_property	depths: DT=0, PT=0, ET=0, E=2	$$42.8 \mathrm{ms} \pm 304 \mathrm{μs}\left({\color{lightgreen}-6.631 \mathrm{\%}}\right) $$	Flame Graph
entity_by_property	depths: DT=0, PT=0, ET=2, E=2	$$48.3 \mathrm{ms} \pm 257 \mathrm{μs}\left({\color{lightgreen}-5.262 \mathrm{\%}}\right) $$	Flame Graph
entity_by_property	depths: DT=0, PT=2, ET=2, E=2	$$52.3 \mathrm{ms} \pm 281 \mathrm{μs}\left({\color{lightgreen}-6.440 \mathrm{\%}}\right) $$	Flame Graph
link_by_source_by_property	depths: DT=255, PT=255, ET=255, E=255	$$100 \mathrm{ms} \pm 441 \mathrm{μs}\left({\color{gray}-0.973 \mathrm{\%}}\right) $$	Flame Graph
link_by_source_by_property	depths: DT=0, PT=0, ET=0, E=0	$$38.7 \mathrm{ms} \pm 254 \mathrm{μs}\left({\color{gray}-4.240 \mathrm{\%}}\right) $$	Flame Graph
link_by_source_by_property	depths: DT=2, PT=2, ET=2, E=2	$$89.3 \mathrm{ms} \pm 217 \mathrm{μs}\left({\color{gray}-4.556 \mathrm{\%}}\right) $$	Flame Graph
link_by_source_by_property	depths: DT=0, PT=0, ET=0, E=2	$$72.7 \mathrm{ms} \pm 460 \mathrm{μs}\left({\color{lightgreen}-6.719 \mathrm{\%}}\right) $$	Flame Graph
link_by_source_by_property	depths: DT=0, PT=0, ET=2, E=2	$$81.1 \mathrm{ms} \pm 276 \mathrm{μs}\left({\color{lightgreen}-5.233 \mathrm{\%}}\right) $$	Flame Graph
link_by_source_by_property	depths: DT=0, PT=2, ET=2, E=2	$$88.4 \mathrm{ms} \pm 752 \mathrm{μs}\left({\color{gray}-2.244 \mathrm{\%}}\right) $$	Flame Graph

representative_read_entity_type

Function	Value	Mean	Flame graphs
get_entity_type_by_id	Account ID: `d4e16033-c281-4cde-aa35-9085bf2e7579`	$$2.14 \mathrm{ms} \pm 5.91 \mathrm{μs}\left({\color{gray}-3.231 \mathrm{\%}}\right) $$	Flame Graph

scaling_read_entity_complete_one_depth

Function	Value	Mean	Flame graphs
entity_by_id	50 entities	$$271 \mathrm{ms} \pm 1.96 \mathrm{ms}\left({\color{lightgreen}-94.977 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	5 entities	$$28.0 \mathrm{ms} \pm 313 \mathrm{μs}\left({\color{gray}1.95 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	1 entities	$$21.0 \mathrm{ms} \pm 92.8 \mathrm{μs}\left({\color{gray}-1.660 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	10 entities	$$57.8 \mathrm{ms} \pm 177 \mathrm{μs}\left({\color{red}76.0 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	25 entities	$$169 \mathrm{ms} \pm 9.32 \mathrm{ms}\left({\color{gray}-4.958 \mathrm{\%}}\right) $$	Flame Graph

scaling_read_entity_linkless

Function	Value	Mean	Flame graphs
entity_by_id	1 entities	$$2.00 \mathrm{ms} \pm 6.16 \mathrm{μs}\left({\color{gray}-0.547 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	100 entities	$$2.23 \mathrm{ms} \pm 9.76 \mathrm{μs}\left({\color{gray}4.35 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	10 entities	$$2.04 \mathrm{ms} \pm 7.36 \mathrm{μs}\left({\color{gray}1.40 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	1000 entities	$$3.16 \mathrm{ms} \pm 9.34 \mathrm{μs}\left({\color{red}7.63 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	10000 entities	$$13.8 \mathrm{ms} \pm 70.0 \mathrm{μs}\left({\color{gray}-2.461 \mathrm{\%}}\right) $$	Flame Graph

scaling_read_entity_complete_zero_depth

Function	Value	Mean	Flame graphs
entity_by_id	50 entities	$$4.14 \mathrm{ms} \pm 20.7 \mathrm{μs}\left({\color{gray}-1.040 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	5 entities	$$2.04 \mathrm{ms} \pm 4.89 \mathrm{μs}\left({\color{gray}0.632 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	1 entities	$$2.03 \mathrm{ms} \pm 15.5 \mathrm{μs}\left({\color{gray}0.261 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	10 entities	$$2.24 \mathrm{ms} \pm 11.0 \mathrm{μs}\left({\color{gray}-0.437 \mathrm{\%}}\right) $$	Flame Graph
entity_by_id	25 entities	$$3.42 \mathrm{ms} \pm 22.5 \mathrm{μs}\left({\color{gray}0.646 \mathrm{\%}}\right) $$	Flame Graph

JesusFileto added 2 commits November 17, 2024 16:02

Add Multi-Modal Embeddings

08857d5

Implement Preliminary Embedding Model

5359c42

github-actions bot added area/deps Relates to third-party dependencies (area) area/libs Relates to first-party libraries/crates/packages (area) area/tests New or updated tests area/libs > chonky Affects the `chonky` crate (library) labels Nov 20, 2024

CiaranMn requested a review from TimDiekmann November 20, 2024 09:47

TimDiekmann requested changes Nov 20, 2024

View reviewed changes

Cargo.toml Outdated Show resolved Hide resolved

libs/chonky/src/embedding.rs Outdated Show resolved Hide resolved

libs/chonky/src/embedding.rs Outdated Show resolved Hide resolved

vilkinsons changed the title ~~Introduce PDF Embeddings~~ H-3614: Introduce chonky PDF Embeddings Nov 20, 2024

vilkinsons assigned JesusFileto Nov 20, 2024

github-advanced-security bot found potential problems Nov 26, 2024

View reviewed changes

libs/chonky/src/embedding.rs Fixed Show fixed Hide fixed

Prevent Cold HuggingFace Models from giving testing errors

3aaf4f7

JesusFileto force-pushed the Chonky branch from d1da5a1 to 3aaf4f7 Compare November 26, 2024 17:53

Begin Implementing Async through Reqwest

5d9d336

vercel bot had a problem deploying to Preview – hashdotdesign November 27, 2024 07:59 Failure

vercel bot deployed to Preview – hash November 27, 2024 08:05 View deployment

vercel bot deployed to Preview – hashdotdev November 27, 2024 08:06 View deployment

TimDiekmann requested changes Nov 28, 2024

View reviewed changes

Struct organization and restructuring + cleanup

2652f84

TimDiekmann requested changes Dec 4, 2024

View reviewed changes

H-3618: Add GitHub action step to authenticate with Google Cloud, for…

c999477

… `chonky` tests (#5687) Co-authored-by: JesusFileto <91223391+JesusFileto@users.noreply.github.com> Co-authored-by: Jesus Fileto <jesus_filetojr@icloud.com>

github-actions bot added the area/infra Relates to version control, CI, CD or IaC (area) label Dec 4, 2024

vercel bot deployed to Preview – hashdotdesign December 4, 2024 15:14 View deployment

vercel bot deployed to Preview – hash December 4, 2024 15:16 View deployment

vercel bot deployed to Preview – hashdotdev December 4, 2024 15:20 View deployment

TimDiekmann reviewed Dec 5, 2024

View reviewed changes

JesusFileto and others added 3 commits December 12, 2024 04:39

Add More Structs + Slice Cleanup

60fbe82

Merge branch 'Chonky' of https://github.com/hashintel/hash into Chonky

346091b

Merge branch 'main' into Chonky

b23b910

github-advanced-security bot found potential problems Dec 12, 2024

View reviewed changes

libs/chonky/src/embedding/multi_modal_embedding.rs Fixed Show fixed Hide fixed

vercel bot had a problem deploying to Preview – hashdotdesign December 12, 2024 15:34 Failure

vercel bot deployed to Preview – hash December 12, 2024 15:39 View deployment

vercel bot deployed to Preview – hashdotdev December 12, 2024 15:40 View deployment

Move Image Pairing with embedding outside of embedding function

5b3ffcd

vercel bot deployed to Preview – hashdotdesign December 12, 2024 16:11 View deployment

Merge branch 'main' into Chonky

af2e960

vercel bot deployed to Preview – hashdotdesign December 12, 2024 16:32 View deployment

vercel bot deployed to Preview – hash December 12, 2024 16:34 View deployment

TimDiekmann reviewed Dec 12, 2024

View reviewed changes

vilkinsons assigned TimDiekmann Jan 10, 2025

vilkinsons requested a review from TimDiekmann January 10, 2025 12:22

TimDiekmann added 3 commits January 13, 2025 09:58

Merge branch 'refs/heads/main' into Chonky

db4a160

# Conflicts: # .markdownlintignore # Cargo.lock # Cargo.toml

Fix merge conflicts

0f4103b

Update libs/chonky/src/embedding/hugging_face_api.rs

2d1a4a1

vercel bot deployed to Preview – hashdotdev January 13, 2025 09:10 View deployment

vercel bot deployed to Preview – hashdotdesign January 13, 2025 09:15 View deployment

vercel bot deployed to Preview – hash January 13, 2025 09:18 View deployment

vilkinsons enabled auto-merge January 13, 2025 10:00

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

H-3614: Introduce `chonky` PDF Embeddings #5673

H-3614: Introduce `chonky` PDF Embeddings #5673

JesusFileto commented Nov 20, 2024

TimDiekmann left a comment

codecov bot commented Nov 26, 2024 •

edited

Loading

TimDiekmann left a comment

TimDiekmann Nov 28, 2024

JesusFileto Dec 4, 2024

TimDiekmann Dec 4, 2024

TimDiekmann Nov 28, 2024

JesusFileto Dec 4, 2024

TimDiekmann left a comment

TimDiekmann Dec 4, 2024

TimDiekmann Dec 4, 2024

TimDiekmann Dec 4, 2024

JesusFileto Dec 12, 2024

TimDiekmann Dec 4, 2024

TimDiekmann Dec 4, 2024

TimDiekmann Dec 4, 2024

TimDiekmann Dec 4, 2024

JesusFileto Dec 12, 2024 •

edited

Loading

TimDiekmann Dec 4, 2024

TimDiekmann Dec 5, 2024

TimDiekmann Dec 12, 2024

TimDiekmann Dec 12, 2024

vilkinsons commented Jan 10, 2025 •

edited

Loading

github-actions bot commented Jan 13, 2025

@rust/hash-graph-benches – Integrations

representative_read_entity

representative_read_multiple_entities

representative_read_entity_type

scaling_read_entity_complete_one_depth

scaling_read_entity_linkless

scaling_read_entity_complete_zero_depth

	fn base64_json(image_data: Vec<u8>) -> Result<String, Report<ChonkyError>> {
	fn base64_json(image_data: &[u8]) -> Result<String, Report<ChonkyError>> {

	fn base64_json(image_data: Vec<u8>) -> Result<String, Report<ChonkyError>> {
	fn base64_json(image_data: impl AsRef<[u8]>) -> Result<String, Report<ChonkyError>> {

		pub mod embedding;

		pub use embedding::{hugging_face_api, multi_modal_embedding};

	_model_used: String, //model name reveals image or text embedding model
	_model_used: Cow<'static, str>, //model name reveals image or text embedding model

	/// a curl request to obtain multimodal embeddings given an image path
	/// a request to obtain multimodal embeddings given an image path

H-3614: Introduce chonky PDF Embeddings #5673

Are you sure you want to change the base?

H-3614: Introduce chonky PDF Embeddings #5673

Conversation

JesusFileto commented Nov 20, 2024

🌟 What is the purpose of this PR?

🚫 Blocked by

🔍 What does this change?

Pre-Merge Checklist 🚀

🚢 Has this modified a publishable library?

📜 Does this require a change to the docs?

🕸️ Does this require a change to the Turbo Graph?

⚠️ Known issues

🐾 Next steps

🛡 What tests cover this?

TimDiekmann left a comment

Choose a reason for hiding this comment

codecov bot commented Nov 26, 2024 • edited Loading

Codecov Report

TimDiekmann left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

TimDiekmann left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JesusFileto Dec 12, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

vilkinsons commented Jan 10, 2025 • edited Loading

github-actions bot commented Jan 13, 2025

Benchmark results

@rust/hash-graph-benches – Integrations

representative_read_entity

representative_read_multiple_entities

representative_read_entity_type

scaling_read_entity_complete_one_depth

scaling_read_entity_linkless

scaling_read_entity_complete_zero_depth

H-3614: Introduce `chonky` PDF Embeddings #5673

H-3614: Introduce `chonky` PDF Embeddings #5673

codecov bot commented Nov 26, 2024 •

edited

Loading

JesusFileto Dec 12, 2024 •

edited

Loading

vilkinsons commented Jan 10, 2025 •

edited

Loading