dynamo_async_openai/types/
embedding.rs

1// SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2// SPDX-License-Identifier: Apache-2.0
3//
4// Based on https://github.com/64bit/async-openai/ by Himanshu Neema
5// Original Copyright (c) 2022 Himanshu Neema
6// Licensed under MIT License (see ATTRIBUTIONS-Rust.md)
7//
8// Modifications Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES.
9// Licensed under Apache 2.0
10
11use base64::engine::{Engine, general_purpose};
12use derive_builder::Builder;
13use serde::{Deserialize, Serialize};
14use utoipa::ToSchema;
15
16use crate::error::OpenAIError;
17
18#[derive(ToSchema, Debug, Serialize, Clone, PartialEq, Deserialize)]
19#[serde(untagged)]
20pub enum EmbeddingInput {
21    String(String),
22    StringArray(Vec<String>),
23    // Minimum value is 0, maximum value is 100257 (inclusive).
24    IntegerArray(Vec<u32>),
25    ArrayOfIntegerArray(Vec<Vec<u32>>),
26}
27
28#[derive(ToSchema, Debug, Serialize, Default, Clone, PartialEq, Deserialize)]
29#[serde(rename_all = "lowercase")]
30pub enum EncodingFormat {
31    #[default]
32    Float,
33    Base64,
34}
35
36#[derive(ToSchema, Debug, Serialize, Default, Clone, Builder, PartialEq, Deserialize)]
37#[builder(name = "CreateEmbeddingRequestArgs")]
38#[builder(pattern = "mutable")]
39#[builder(setter(into, strip_option), default)]
40#[builder(derive(Debug))]
41#[builder(build_fn(error = "OpenAIError"))]
42pub struct CreateEmbeddingRequest {
43    /// ID of the model to use. You can use the
44    /// [List models](https://platform.openai.com/docs/api-reference/models/list)
45    /// API to see all of your available models, or see our
46    /// [Model overview](https://platform.openai.com/docs/models/overview)
47    /// for descriptions of them.
48    pub model: String,
49
50    ///  Input text to embed, encoded as a string or array of tokens. To embed multiple inputs in a single request, pass an array of strings or array of token arrays. The input must not exceed the max input tokens for the model (8192 tokens for `text-embedding-ada-002`), cannot be an empty string, and any array must be 2048 dimensions or less. [Example Python code](https://cookbook.openai.com/examples/how_to_count_tokens_with_tiktoken) for counting tokens.
51    pub input: EmbeddingInput,
52
53    /// The format to return the embeddings in. Can be either `float` or [`base64`](https://pypi.org/project/pybase64/). Defaults to float
54    #[serde(skip_serializing_if = "Option::is_none")]
55    pub encoding_format: Option<EncodingFormat>,
56
57    /// A unique identifier representing your end-user, which will help OpenAI
58    ///  to monitor and detect abuse. [Learn more](https://platform.openai.com/docs/usage-policies/end-user-ids).
59    #[serde(skip_serializing_if = "Option::is_none")]
60    pub user: Option<String>,
61
62    /// The number of dimensions the resulting output embeddings should have. Only supported in `text-embedding-3` and later models.
63    #[serde(skip_serializing_if = "Option::is_none")]
64    pub dimensions: Option<u32>,
65}
66
67/// Represents an embedding vector returned by embedding endpoint.
68#[derive(ToSchema, Debug, Deserialize, Serialize, Clone, PartialEq)]
69pub struct Embedding {
70    /// The index of the embedding in the list of embeddings.
71    pub index: u32,
72    /// The object type, which is always "embedding".
73    pub object: String,
74    /// The embedding vector, which is a list of floats. The length of vector
75    /// depends on the model as listed in the [embedding guide](https://platform.openai.com/docs/guides/embeddings).
76    pub embedding: Vec<f32>,
77}
78
79#[derive(ToSchema, Debug, Deserialize, Serialize, Clone, PartialEq)]
80pub struct Base64EmbeddingVector(pub String);
81
82impl From<Base64EmbeddingVector> for Vec<f32> {
83    fn from(value: Base64EmbeddingVector) -> Self {
84        let bytes = general_purpose::STANDARD
85            .decode(value.0)
86            .expect("openai base64 encoding to be valid");
87        let chunks = bytes.chunks_exact(4);
88        chunks
89            .map(|chunk| f32::from_le_bytes([chunk[0], chunk[1], chunk[2], chunk[3]]))
90            .collect()
91    }
92}
93
94/// Represents an base64-encoded embedding vector returned by embedding endpoint.
95#[derive(ToSchema, Debug, Deserialize, Serialize, Clone, PartialEq)]
96pub struct Base64Embedding {
97    /// The index of the embedding in the list of embeddings.
98    pub index: u32,
99    /// The object type, which is always "embedding".
100    pub object: String,
101    /// The embedding vector, encoded in base64.
102    pub embedding: Base64EmbeddingVector,
103}
104
105#[derive(ToSchema, Debug, Deserialize, Serialize, Clone, PartialEq)]
106pub struct EmbeddingUsage {
107    /// The number of tokens used by the prompt.
108    pub prompt_tokens: u32,
109    /// The total number of tokens used by the request.
110    pub total_tokens: u32,
111}
112
113#[derive(ToSchema, Debug, Deserialize, Clone, PartialEq, Serialize)]
114pub struct CreateEmbeddingResponse {
115    pub object: String,
116    /// The name of the model used to generate the embedding.
117    pub model: String,
118    /// The list of embeddings generated by the model.
119    pub data: Vec<Embedding>,
120    /// The usage information for the request.
121    pub usage: EmbeddingUsage,
122}
123
124#[derive(ToSchema, Debug, Deserialize, Clone, PartialEq, Serialize)]
125pub struct CreateBase64EmbeddingResponse {
126    pub object: String,
127    /// The name of the model used to generate the embedding.
128    pub model: String,
129    /// The list of embeddings generated by the model.
130    pub data: Vec<Base64Embedding>,
131    /// The usage information for the request.
132    pub usage: EmbeddingUsage,
133}
dynamo_async_openai/types/embedding.rs

dynamo_async_openai/types/
embedding.rs
