onwards 0.19.1

A flexible LLM proxy library
Documentation 
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
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218

//! Legacy Completions API schemas
//!
//! These schemas match the OpenAI Completions API specification (legacy text completions).
//! See: https://platform.openai.com/docs/api-reference/completions
//!
//! In strict mode, completions requests are validated against the typed schema,
//! forwarded to the upstream `/v1/completions` endpoint, and the response is
//! sanitized (unknown fields stripped, model field rewritten).

use serde::{Deserialize, Serialize};

use super::chat_completions::{StopSequence, Usage};

/// Prompt input — matches the OpenAI spec `oneOf`: string | string[] | integer[] | integer[][]
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(untagged)]
pub enum CompletionPrompt {
    Single(String),
    Multiple(Vec<String>),
    /// Pre-tokenized prompt as a flat array of token IDs
    Tokens(Vec<u32>),
    /// Pre-tokenized prompt as an array of token ID arrays (batch)
    TokenArrays(Vec<Vec<u32>>),
}

/// Request body for POST /v1/completions
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct CompletionRequest {
    /// The model to use for completion
    pub model: String,

    /// The prompt to generate completions for (optional; defaults to `<|endoftext|>` server-side)
    #[serde(skip_serializing_if = "Option::is_none")]
    pub prompt: Option<CompletionPrompt>,

    /// Text to append after the completion (fill-in-the-middle)
    #[serde(skip_serializing_if = "Option::is_none")]
    pub suffix: Option<String>,

    /// Maximum tokens to generate
    #[serde(skip_serializing_if = "Option::is_none")]
    pub max_tokens: Option<u32>,

    /// Sampling temperature (0–2)
    #[serde(skip_serializing_if = "Option::is_none")]
    pub temperature: Option<f32>,

    /// Nucleus sampling parameter
    #[serde(skip_serializing_if = "Option::is_none")]
    pub top_p: Option<f32>,

    /// Number of completions to generate
    #[serde(skip_serializing_if = "Option::is_none")]
    pub n: Option<u32>,

    /// Whether to stream the response
    #[serde(skip_serializing_if = "Option::is_none")]
    pub stream: Option<bool>,

    /// Include log probabilities (0–5)
    #[serde(skip_serializing_if = "Option::is_none")]
    pub logprobs: Option<u32>,

    /// Echo the prompt in the response
    #[serde(skip_serializing_if = "Option::is_none")]
    pub echo: Option<bool>,

    /// Stop sequences
    #[serde(skip_serializing_if = "Option::is_none")]
    pub stop: Option<StopSequence>,

    /// Presence penalty (−2.0 to 2.0)
    #[serde(skip_serializing_if = "Option::is_none")]
    pub presence_penalty: Option<f32>,

    /// Frequency penalty (−2.0 to 2.0)
    #[serde(skip_serializing_if = "Option::is_none")]
    pub frequency_penalty: Option<f32>,

    /// Generate best_of completions server-side and return the best
    #[serde(skip_serializing_if = "Option::is_none")]
    pub best_of: Option<u32>,

    /// Logit bias for tokens
    #[serde(skip_serializing_if = "Option::is_none")]
    pub logit_bias: Option<serde_json::Value>,

    /// User identifier for abuse tracking
    #[serde(skip_serializing_if = "Option::is_none")]
    pub user: Option<String>,

    /// Random seed for deterministic sampling
    #[serde(skip_serializing_if = "Option::is_none")]
    pub seed: Option<i64>,
}

/// Response from POST /v1/completions (non-streaming)
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct CompletionResponse {
    pub id: String,
    pub object: String,
    pub created: u64,
    pub model: String,
    pub choices: Vec<CompletionChoice>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub usage: Option<Usage>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub system_fingerprint: Option<String>,
}

/// A single completion choice
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct CompletionChoice {
    pub text: String,
    pub index: u32,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub logprobs: Option<serde_json::Value>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub finish_reason: Option<String>,
}

/// Streaming chunk from POST /v1/completions with stream=true
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct CompletionChunk {
    pub id: String,
    pub object: String,
    pub created: u64,
    pub model: String,
    pub choices: Vec<CompletionChunkChoice>,
}

/// A single choice within a streaming completion chunk
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct CompletionChunkChoice {
    pub text: String,
    pub index: u32,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub logprobs: Option<serde_json::Value>,
    #[serde(skip_serializing_if = "Option::is_none")]
    pub finish_reason: Option<String>,
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_deserialize_string_prompt() {
        let json = r#"{"model": "gpt-3.5-turbo-instruct", "prompt": "Say hello"}"#;
        let req: CompletionRequest = serde_json::from_str(json).unwrap();
        assert!(matches!(req.prompt, Some(CompletionPrompt::Single(_))));
    }

    #[test]
    fn test_deserialize_array_of_strings_prompt() {
        let json = r#"{"model": "gpt-3.5-turbo-instruct", "prompt": ["Hello", "World"]}"#;
        let req: CompletionRequest = serde_json::from_str(json).unwrap();
        assert!(matches!(req.prompt, Some(CompletionPrompt::Multiple(_))));
    }

    #[test]
    fn test_deserialize_token_array_prompt() {
        let json = r#"{"model": "gpt-3.5-turbo-instruct", "prompt": [1, 2, 3]}"#;
        let req: CompletionRequest = serde_json::from_str(json).unwrap();
        assert!(matches!(req.prompt, Some(CompletionPrompt::Tokens(_))));
    }

    #[test]
    fn test_deserialize_token_array_of_arrays_prompt() {
        let json = r#"{"model": "gpt-3.5-turbo-instruct", "prompt": [[1, 2], [3, 4]]}"#;
        let req: CompletionRequest = serde_json::from_str(json).unwrap();
        assert!(matches!(req.prompt, Some(CompletionPrompt::TokenArrays(_))));
    }

    #[test]
    fn test_reject_mixed_token_array_prompt() {
        // Mixed arrays (e.g. integers and strings) don't match any variant
        let json = r#"{"model": "gpt-3.5-turbo-instruct", "prompt": [1, "hello"]}"#;
        assert!(serde_json::from_str::<CompletionRequest>(json).is_err());
    }

    #[test]
    fn test_reject_float_token_array_prompt() {
        // Floats are not integers per the spec
        let json = r#"{"model": "gpt-3.5-turbo-instruct", "prompt": [1.5, 2.5]}"#;
        assert!(serde_json::from_str::<CompletionRequest>(json).is_err());
    }

    #[test]
    fn test_deserialize_with_all_fields() {
        let json = r#"{
            "model": "gpt-3.5-turbo-instruct",
            "prompt": "Complete this",
            "suffix": "end",
            "max_tokens": 100,
            "temperature": 0.7,
            "top_p": 0.9,
            "n": 1,
            "stream": false,
            "logprobs": 3,
            "echo": true,
            "stop": "\n",
            "presence_penalty": 0.1,
            "frequency_penalty": 0.2,
            "best_of": 3,
            "user": "user-123",
            "seed": 42
        }"#;
        let req: CompletionRequest = serde_json::from_str(json).unwrap();
        assert_eq!(req.model, "gpt-3.5-turbo-instruct");
        assert_eq!(req.max_tokens, Some(100));
        assert_eq!(req.temperature, Some(0.7));
        assert_eq!(req.logprobs, Some(3));
        assert_eq!(req.echo, Some(true));
        assert_eq!(req.best_of, Some(3));
        assert_eq!(req.seed, Some(42));
    }
}