liboxen 0.53.0

Oxen is a fast data version control system, built with machine learning training data in mind. Designed to handle terabytes of data with ease, using a workflow similar to git. Version both structured and unstructured data of any modality: text, images, video, audio, CSV, Parquet, JSONL, model checkpoints, and more. liboxen is the embeddable core library behind the oxen CLI and server, which power fine tuning and inference pipelines for multimodal LLMs, image models, and video models on Oxen.ai.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137

// This format is deprecated favor of the new JSON format

use std::io::BufWriter;
use std::str;

use polars::prelude::*;
use serde::{Deserialize, Serialize};
use std::io::Cursor;
use utoipa::ToSchema;

use crate::core::df::tabular;
use crate::model::DataFrameSize;
use crate::{model::Schema, opts::DFOpts};

#[derive(Serialize, Deserialize, Debug, Clone, ToSchema)]
pub struct JsonDataFrame {
    pub schema: Schema,
    pub view_schema: Schema,
    pub view_size: DataFrameSize,
    pub full_size: DataFrameSize,
    pub data: serde_json::Value,
}

impl JsonDataFrame {
    pub fn empty(schema: &Schema) -> JsonDataFrame {
        JsonDataFrame {
            schema: schema.to_owned(),
            view_schema: schema.to_owned(),
            view_size: DataFrameSize {
                height: 0,
                width: 0,
            },
            full_size: DataFrameSize {
                height: 0,
                width: 0,
            },
            data: serde_json::Value::Null,
        }
    }

    pub async fn from_df_opts(df: DataFrame, opts: DFOpts) -> JsonDataFrame {
        let full_height = df.height();
        let full_width = df.width();

        let schema = Schema::from_polars(df.schema());
        let mut view_df = tabular::transform(df, opts).await.unwrap();
        let view_schema = Schema::from_polars(view_df.schema());

        JsonDataFrame {
            schema,
            view_schema,
            view_size: DataFrameSize {
                height: view_df.height(),
                width: view_df.width(),
            },
            full_size: DataFrameSize {
                height: full_height,
                width: full_width,
            },
            data: JsonDataFrame::json_data(&mut view_df),
        }
    }

    pub async fn to_df(&self) -> DataFrame {
        if self.data == serde_json::Value::Null {
            DataFrame::empty()
        } else {
            // The fields were coming out of order, so we need to reorder them
            let columns = self.schema.fields_names();
            log::debug!("Got columns: {columns:?}");

            match &self.data {
                serde_json::Value::Array(arr) => {
                    if !arr.is_empty() {
                        let data = self.data.to_string();
                        let content = Cursor::new(data.as_bytes());
                        log::debug!("Deserializing df: [{data}]");
                        let df = JsonReader::new(content).finish().unwrap();

                        let opts = DFOpts::from_column_names(columns);
                        tabular::transform(df, opts).await.unwrap()
                    } else {
                        let cols = columns
                            .iter()
                            .map(|name| {
                                Column::Series(
                                    Series::new(PlSmallStr::from_str(name), Vec::<&str>::new())
                                        .into(),
                                )
                            })
                            .collect::<Vec<Column>>();
                        DataFrame::new(cols).unwrap()
                    }
                }
                _ => {
                    log::error!("Could not parse non-array json data: {:?}", self.data);
                    DataFrame::empty()
                }
            }
        }
    }

    pub fn from_slice(
        df: &mut DataFrame,
        og_schema: Schema,
        og_size: DataFrameSize,
        view_schema: Schema,
    ) -> JsonDataFrame {
        JsonDataFrame {
            schema: og_schema,
            view_schema,
            view_size: DataFrameSize {
                height: df.height(),
                width: df.width(),
            },
            full_size: og_size,
            data: JsonDataFrame::json_data(df),
        }
    }

    fn json_data(df: &mut DataFrame) -> serde_json::Value {
        log::debug!("Serializing df: [{df}]");

        // TODO: serialize to json
        let data: Vec<u8> = Vec::new();
        let mut buf = BufWriter::new(data);

        let mut writer = JsonWriter::new(&mut buf).with_json_format(JsonFormat::Json);
        writer.finish(df).expect("Could not write df json buffer");

        let buffer = buf.into_inner().expect("Could not get buffer");

        let json_str = str::from_utf8(&buffer).unwrap();

        serde_json::from_str(json_str).unwrap()
    }
}