https://github.com/wdoppenberg/polars-candle
A text embedding extension for the Polars Dataframe library.
https://github.com/wdoppenberg/polars-candle
Last synced: 4 months ago
JSON representation
A text embedding extension for the Polars Dataframe library.
- Host: GitHub
- URL: https://github.com/wdoppenberg/polars-candle
- Owner: wdoppenberg
- Created: 2024-04-19T12:12:33.000Z (over 1 year ago)
- Default Branch: master
- Last Pushed: 2024-11-21T07:31:26.000Z (12 months ago)
- Last Synced: 2025-06-29T18:03:00.567Z (4 months ago)
- Language: Python
- Size: 116 KB
- Stars: 24
- Watchers: 1
- Forks: 0
- Open Issues: 3
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- awesome-polars - polars-candle - Polars plugin for running [candle](https://github.com/huggingface/candle) ML models on Polars DataFrames by [@wdoppenberg](https://github.com/wdoppenberg). (Libraries/Packages/Scripts / Polars plugins)
README
# `polars-candle`
A [`polars`](https://pola.rs/) extension for running [`candle`](https://github.com/huggingface/candle) ML
models on `polars` DataFrames.
# Example
Pull any applicable model from Huggingface, such as the recently released
[Snowflake model](https://huggingface.co/Snowflake/snowflake-arctic-embed-xs), and embed text using a simple API.
```python
import polars as pl
import polars_candle # ignore: F401
df = pl.DataFrame({"s": ["This is a sentence", "This is another sentence"]})
embed_kwargs = {
"model_repo": "Snowflake/snowflake-arctic-embed-xs",
"pooling": "mean",
}
df = df.with_columns(
pl.col("s").candle.embed_text(**embed_kwargs).alias("s_embedding")
)
print(df)
# ┌──────────────────────────┬───────────────────────────────────┐
# │ s ┆ s_embedding │
# │ --- ┆ --- │
# │ str ┆ array[f32, 384] │
# ╞══════════════════════════╪═══════════════════════════════════╡
# │ This is a sentence ┆ [-0.056457, 0.559411, … -0.20403… │
# │ This is another sentence ┆ [-0.117206, 0.336827, … 0.174078… │
# └──────────────────────────┴───────────────────────────────────┘
```
Currently, Bert, JinaBert, and Distilbert models are supported. More models will be added in the future. Check
my other repository [`wdoppenberg/glowrs`](https://github.com/wdoppenberg/glowrs) to learn more about the underlying
implementation for sentence embedding.
# Installation
Make sure you have `polars` installed. If not, install it using `pip install polars`. Then, install `polars-candle` using
```bash
pip install polars-candle
```
_Note:_ The macOS ARM wheels of this library come with Metal support out of the box. For CUDA, check the below
instructions on how to build from source.
If you want to install the latest version from the repository, you can use:
```bash
pip install git+https://github.com/wdoppenberg/polars-candle.git
```
_Note:_ You need to have the Rust toolchain installed on your system to compile the library. See
[here](https://www.rust-lang.org/tools/install) for instructions on how to install Rust.
You can set build features using `maturin`:
```bash
maturin develop --release -F
```
Where `` can be one of the following:
* `metal` Install with Metal acceleration.
* `cuda` Install with CUDA acceleration. Might require additional setup such as installing CUDA libraries.
* `accelerate` Install with the Accelerate framework.
# Roadmap
- [x] Embed text using Bert, JinaBert, and Distilbert models.
- [ ] Add more models.
- [ ] More configuration options for embedding (e.g. pooling strategy, device selection, etc.).
- [ ] Support & test streaming workloads.
# Credits
- Massive thanks to [`polars`](https://pola.rs/) & their contributors for providing a blazing fast DataFrame library
with the ability to extend it with custom functions using [`pyo3-polars`](https://github.com/pola-rs/pyo3-polars).
- Great work so far by Huggingface on [`candle`](https://github.com/huggingface/candle) for providing a simple
interface to run ML models.
# Note
This is a work in progress and the API might change in the future. Feel free to open an issue if you have any
suggestions or improvements.