lib.rs 43.4 KB
Newer Older
1
/// Text Generation Inference Webserver
OlivierDehaene's avatar
OlivierDehaene committed
2
pub mod config;
3
mod infer;
Olivier Dehaene's avatar
Olivier Dehaene committed
4
pub mod server;
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
5
mod validation;
Olivier Dehaene's avatar
Olivier Dehaene committed
6

7
8
9
#[cfg(feature = "kserve")]
mod kserve;

Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
10
use serde::{Deserialize, Serialize};
Nicolas Patry's avatar
Nicolas Patry committed
11
use tracing::warn;
12
use utoipa::ToSchema;
Olivier Dehaene's avatar
Olivier Dehaene committed
13
use validation::Validation;
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
14

drbh's avatar
drbh committed
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
#[derive(Clone, Deserialize, ToSchema)]
pub(crate) struct VertexInstance {
    #[schema(example = "What is Deep Learning?")]
    pub inputs: String,
    #[schema(nullable = true, default = "null", example = "null")]
    pub parameters: Option<GenerateParameters>,
}

#[derive(Deserialize, ToSchema)]
pub(crate) struct VertexRequest {
    #[serde(rename = "instances")]
    pub instances: Vec<VertexInstance>,
}

#[derive(Clone, Deserialize, ToSchema, Serialize)]
pub(crate) struct VertexResponse {
    pub predictions: Vec<String>,
}

34
35
/// Hub type
#[derive(Clone, Debug, Deserialize)]
36
pub struct HubModelInfo {
37
38
39
40
41
42
    #[serde(rename(deserialize = "id"))]
    pub model_id: String,
    pub sha: Option<String>,
    pub pipeline_tag: Option<String>,
}

43
44
45
46
47
48
49
50
51
52
53
54
55
#[derive(Debug, Clone, Deserialize, PartialEq)]
pub struct ChatTemplate {
    name: String,
    template: String,
}

#[derive(Debug, Clone, Deserialize, PartialEq)]
#[serde(untagged)]
pub enum ChatTemplateVersions {
    Single(String),
    Multiple(Vec<ChatTemplate>),
}

56
57
use std::path::Path;

58
#[derive(Debug, Clone, Deserialize, Default)]
59
pub struct HubTokenizerConfig {
60
    pub chat_template: Option<ChatTemplateVersions>,
61
    pub completion_template: Option<String>,
62
63
    pub bos_token: Option<TokenizerConfigToken>,
    pub eos_token: Option<TokenizerConfigToken>,
64
65
66
    pub tokenizer_class: Option<String>,
    pub add_bos_token: Option<bool>,
    pub add_eos_token: Option<bool>,
67
68
69
}

impl HubTokenizerConfig {
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
    pub fn from_file<P: AsRef<Path>>(filename: P) -> Option<Self> {
        std::fs::read_to_string(filename)
            .ok()
            .and_then(|content| serde_json::from_str(&content).ok())
    }
}

#[derive(Debug, Clone, Deserialize, Serialize, PartialEq)]
#[serde(untagged)]
pub enum TokenizerConfigToken {
    String(String),
    Object { content: String },
}

impl TokenizerConfigToken {
    pub fn as_str(&self) -> &str {
        match self {
            TokenizerConfigToken::String(s) => s,
            TokenizerConfigToken::Object { content } => content,
        }
90
91
92
    }
}

93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(tag = "processor_class")]
pub enum HubPreprocessorConfig {
    Idefics2Processor(Idefics2Preprocessor),
}

impl HubPreprocessorConfig {
    pub fn from_file<P: AsRef<std::path::Path>>(filename: P) -> Option<Self> {
        let content = std::fs::read_to_string(filename).ok()?;
        serde_json::from_str(&content).ok()
    }
}

#[derive(Clone, Debug, Serialize, Deserialize)]
pub struct Idefics2Preprocessor {
    #[serde(default)]
    do_image_splitting: bool,
}

drbh's avatar
drbh committed
112
113
114
115
116
117
118
119
#[derive(Debug, Clone, Deserialize, Default)]
pub struct HubProcessorConfig {
    pub chat_template: Option<ChatTemplateVersions>,
    pub image_seq_len: usize,
    pub processor_class: Option<String>,
}

impl HubProcessorConfig {
120
121
122
123
    pub fn from_file<P: AsRef<Path>>(filename: P) -> Option<Self> {
        std::fs::read_to_string(filename)
            .ok()
            .and_then(|content| serde_json::from_str(&content).ok())
drbh's avatar
drbh committed
124
125
126
    }
}

127
#[derive(Clone, Debug, Deserialize, ToSchema, Serialize)]
drbh's avatar
drbh committed
128
129
#[serde(tag = "type", content = "value")]
pub(crate) enum GrammarType {
130
131
132
133
134
    /// A string that represents a [JSON Schema](https://json-schema.org/).
    ///
    /// JSON Schema is a declarative language that allows to annotate JSON documents
    /// with types and descriptions.
    #[serde(rename = "json")]
drbh's avatar
drbh committed
135
    #[serde(alias = "json_object")]
136
137
    #[schema(example = json ! ({"properties": {"location":{"type": "string"}}}))]
    Json(serde_json::Value),
drbh's avatar
drbh committed
138
139
140
141
    #[serde(rename = "regex")]
    Regex(String),
}

142
143
#[derive(Clone, Debug, Serialize, ToSchema)]
pub struct Info {
144
    /// Model info
145
146
147
148
    #[schema(example = "bigscience/blomm-560m")]
    pub model_id: String,
    #[schema(nullable = true, example = "e985a63cdc139290c5f700ff1929f0b5942cced2")]
    pub model_sha: Option<String>,
149
150
151
152
    #[schema(example = "torch.float16")]
    pub model_dtype: String,
    #[schema(example = "cuda")]
    pub model_device_type: String,
153
154
    #[schema(nullable = true, example = "text-generation")]
    pub model_pipeline_tag: Option<String>,
155
156
157
158
159
160
161
162
    /// Router Parameters
    #[schema(example = "128")]
    pub max_concurrent_requests: usize,
    #[schema(example = "2")]
    pub max_best_of: usize,
    #[schema(example = "4")]
    pub max_stop_sequences: usize,
    #[schema(example = "1024")]
OlivierDehaene's avatar
OlivierDehaene committed
163
    pub max_input_tokens: usize,
164
165
166
167
168
169
170
171
    #[schema(example = "2048")]
    pub max_total_tokens: usize,
    #[schema(example = "1.2")]
    pub waiting_served_ratio: f32,
    #[schema(example = "32000")]
    pub max_batch_total_tokens: u32,
    #[schema(example = "20")]
    pub max_waiting_tokens: usize,
172
173
    #[schema(nullable = true, example = "null")]
    pub max_batch_size: Option<usize>,
174
175
    #[schema(example = "2")]
    pub validation_workers: usize,
176
177
    #[schema(example = "32")]
    pub max_client_batch_size: usize,
178
    /// Router Info
179
180
    #[schema(example = "text-generation-router")]
    pub router: &'static str,
181
182
183
184
    #[schema(example = "0.5.0")]
    pub version: &'static str,
    #[schema(nullable = true, example = "null")]
    pub sha: Option<&'static str>,
185
186
    #[schema(nullable = true, example = "null")]
    pub docker_label: Option<&'static str>,
187
188
}

drbh's avatar
drbh committed
189
#[derive(Clone, Debug, Deserialize, ToSchema, Default)]
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
190
pub(crate) struct GenerateParameters {
191
    /// Generate best_of sequences and return the one if the highest token logprobs.
192
193
194
    #[serde(default)]
    #[schema(exclusive_minimum = 0, nullable = true, default = "null", example = 1)]
    pub best_of: Option<usize>,
195
196

    /// The value used to module the logits distribution.
197
198
199
200
201
202
203
204
    #[serde(default)]
    #[schema(
        exclusive_minimum = 0.0,
        nullable = true,
        default = "null",
        example = 0.5
    )]
    pub temperature: Option<f32>,
205
206
207

    /// The parameter for repetition penalty. 1.0 means no penalty.
    /// See [this paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.
208
209
210
211
212
213
214
215
    #[serde(default)]
    #[schema(
        exclusive_minimum = 0.0,
        nullable = true,
        default = "null",
        example = 1.03
    )]
    pub repetition_penalty: Option<f32>,
216
217
218
219

    /// The parameter for frequency penalty. 1.0 means no penalty
    /// Penalize new tokens based on their existing frequency in the text so far,
    /// decreasing the model's likelihood to repeat the same line verbatim.
220
    #[serde(default)]
221
222
223
224
225
226
227
    #[schema(
        exclusive_minimum = -2.0,
        nullable = true,
        default = "null",
        example = 0.1
    )]
    pub frequency_penalty: Option<f32>,
228
229

    /// The number of highest probability vocabulary tokens to keep for top-k-filtering.
230
    #[serde(default)]
231
232
    #[schema(exclusive_minimum = 0, nullable = true, default = "null", example = 10)]
    pub top_k: Option<i32>,
233
234

    /// Top-p value for nucleus sampling.
235
236
237
238
239
240
241
242
243
    #[serde(default)]
    #[schema(
        exclusive_minimum = 0.0,
        maximum = 1.0,
        nullable = true,
        default = "null",
        example = 0.95
    )]
    pub top_p: Option<f32>,
244
245
246

    /// Typical Decoding mass
    /// See [Typical Decoding for Natural Language Generation](https://arxiv.org/abs/2202.00666) for more information.
247
    #[serde(default)]
248
249
250
251
252
253
254
255
    #[schema(
        exclusive_minimum = 0.0,
        maximum = 1.0,
        nullable = true,
        default = "null",
        example = 0.95
    )]
    pub typical_p: Option<f32>,
256
257

    /// Activate logits sampling.
258
    #[serde(default)]
259
    #[schema(default = "false", example = true)]
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
260
    pub do_sample: bool,
261
262

    /// Maximum number of tokens to generate.
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
263
    #[serde(default = "default_max_new_tokens")]
264
    #[schema(nullable = true, default = "100", example = "20")]
265
    pub max_new_tokens: Option<u32>,
266
267

    /// Whether to prepend the prompt to the generated text
OlivierDehaene's avatar
OlivierDehaene committed
268
    #[serde(default)]
269
    #[schema(nullable = true, default = "null", example = false)]
270
    pub return_full_text: Option<bool>,
271
272

    /// Stop generating tokens if a member of `stop` is generated.
273
    #[serde(default)]
274
    #[schema(inline, max_items = 4, example = json ! (["photographer"]))]
275
    pub stop: Vec<String>,
276
277

    /// Truncate inputs tokens to the given size.
OlivierDehaene's avatar
OlivierDehaene committed
278
    #[serde(default)]
279
    #[schema(nullable = true, default = "null", example = "null")]
280
    pub truncate: Option<usize>,
281
282

    /// Watermarking with [A Watermark for Large Language Models](https://arxiv.org/abs/2301.10226).
283
    #[serde(default)]
284
285
    #[schema(default = "false", example = true)]
    pub watermark: bool,
286
287

    /// Whether to return generation details.
288
    #[serde(default)]
289
    #[schema(default = "true")]
OlivierDehaene's avatar
OlivierDehaene committed
290
    pub details: bool,
291
292

    /// Whether to return decoder input token logprobs and ids.
293
    #[serde(default)]
294
    #[schema(default = "false")]
295
    pub decoder_input_details: bool,
296
297

    /// Random sampling seed.
298
    #[serde(default)]
299
300
301
302
303
304
    #[schema(
        exclusive_minimum = 0,
        nullable = true,
        default = "null",
        example = "null"
    )]
305
    pub seed: Option<u64>,
306
307

    /// The number of highest probability vocabulary tokens to keep for top-n-filtering.
Nicolas Patry's avatar
Nicolas Patry committed
308
309
310
    #[serde(default)]
    #[schema(exclusive_minimum = 0, nullable = true, default = "null", example = 5)]
    pub top_n_tokens: Option<u32>,
311
312

    /// Grammar constraints for the generation.
drbh's avatar
drbh committed
313
    #[serde(default)]
314
    #[schema(nullable = true, default = "null", example = "null")]
drbh's avatar
drbh committed
315
    pub grammar: Option<GrammarType>,
drbh's avatar
drbh committed
316
317
318
319
320

    /// Lora adapter id
    #[serde(default)]
    #[schema(nullable = true, default = "null", example = "null")]
    pub adapter_id: Option<String>,
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
321
322
}

323
fn default_max_new_tokens() -> Option<u32> {
324
    Some(100)
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
325
326
327
328
}

fn default_parameters() -> GenerateParameters {
    GenerateParameters {
329
        best_of: None,
330
331
        temperature: None,
        repetition_penalty: None,
332
        frequency_penalty: None,
333
334
        top_k: None,
        top_p: None,
335
        typical_p: None,
336
        do_sample: true,
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
337
        max_new_tokens: default_max_new_tokens(),
338
        return_full_text: None,
339
        stop: Vec::new(),
340
        truncate: None,
341
        watermark: false,
OlivierDehaene's avatar
OlivierDehaene committed
342
        details: false,
343
        decoder_input_details: false,
344
        seed: None,
Nicolas Patry's avatar
Nicolas Patry committed
345
        top_n_tokens: None,
drbh's avatar
drbh committed
346
        grammar: None,
drbh's avatar
drbh committed
347
        adapter_id: None,
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
348
349
350
    }
}

351
352
353
354
355
356
357
358
359
360
361
362
363
#[derive(Clone, Deserialize, Serialize, ToSchema, Debug)]
#[serde(try_from = "PromptDeserializer")]
pub struct Prompt(pub Vec<String>);

#[derive(Deserialize)]
#[serde(untagged)]
enum PromptDeserializer {
    Single(String),
    Multiple(Vec<String>),
}

impl TryFrom<PromptDeserializer> for Prompt {
    type Error = String;
364

365
    fn try_from(value: PromptDeserializer) -> Result<Self, Self::Error> {
366
        match value {
367
368
369
370
371
372
373
374
375
376
377
            PromptDeserializer::Single(s) => Ok(Prompt(vec![s])),
            PromptDeserializer::Multiple(v) => {
                if v.is_empty() {
                    Err(
                        "Empty array detected. Do not use an empty array for the prompt."
                            .to_string(),
                    )
                } else {
                    Ok(Prompt(v))
                }
            }
378
379
380
381
        }
    }
}

382
383
384
385
386
387
388
389
390
#[derive(Clone, Deserialize, Serialize, ToSchema, Debug)]
pub struct CompletionRequest {
    /// UNUSED
    #[schema(example = "mistralai/Mistral-7B-Instruct-v0.2")]
    /// ID of the model to use. See the model endpoint compatibility table for details on which models work with the Chat API.
    pub model: String,

    /// The prompt to generate completions for.
    #[schema(example = "What is Deep Learning?")]
391
    pub prompt: Prompt,
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428

    /// The maximum number of tokens that can be generated in the chat completion.
    #[serde(default)]
    #[schema(default = "32")]
    pub max_tokens: Option<u32>,

    /// What sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the output more random, while
    /// lower values like 0.2 will make it more focused and deterministic. We generally recommend altering this or `top_p` but not both.
    #[serde(default)]
    #[schema(nullable = true, example = 1.0)]
    pub temperature: Option<f32>,

    /// An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the
    /// tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered.
    #[serde(default)]
    #[schema(nullable = true, example = 0.95)]
    pub top_p: Option<f32>,

    #[serde(default = "bool::default")]
    pub stream: bool,

    #[schema(nullable = true, example = 42)]
    pub seed: Option<u64>,

    /// The text to append to the prompt. This is useful for completing sentences or generating a paragraph of text.
    /// please see the completion_template field in the model's tokenizer_config.json file for completion template.
    #[serde(default)]
    pub suffix: Option<String>,

    #[serde(default)]
    pub repetition_penalty: Option<f32>,

    /// Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far,
    /// decreasing the model's likelihood to repeat the same line verbatim.
    #[serde(default)]
    #[schema(example = "1.0")]
    pub frequency_penalty: Option<f32>,
429
430
431
432
433

    /// Up to 4 sequences where the API will stop generating further tokens.
    #[serde(default)]
    #[schema(nullable = true, example = "null")]
    pub stop: Option<Vec<String>>,
434
435
}

436
437
438
439
440
441
442
443
444
#[derive(Clone, Serialize, ToSchema)]
#[serde(tag = "object")]
enum Completion {
    #[serde(rename = "text_completion")]
    Chunk(Chunk),
    #[serde(rename = "text_completion")]
    Final(CompletionFinal),
}

445
#[derive(Clone, Deserialize, Serialize, ToSchema, Default)]
446
pub(crate) struct CompletionFinal {
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
    pub id: String,
    #[schema(example = "1706270835")]
    pub created: u64,
    #[schema(example = "mistralai/Mistral-7B-Instruct-v0.2")]
    pub model: String,
    pub system_fingerprint: String,
    pub choices: Vec<CompletionComplete>,
    pub usage: Usage,
}

#[derive(Clone, Deserialize, Serialize, ToSchema)]
pub(crate) struct CompletionComplete {
    pub index: u32,
    pub text: String,
    pub logprobs: Option<Vec<f32>>,
    pub finish_reason: String,
}

465
466
467
468
469
470
471
472
473
#[derive(Clone, Deserialize, Serialize, ToSchema)]
pub(crate) struct Chunk {
    pub id: String,
    pub created: u64,
    pub choices: Vec<CompletionComplete>,
    pub model: String,
    pub system_fingerprint: String,
}

474
#[derive(Clone, Deserialize, Serialize, ToSchema)]
475
476
pub(crate) struct ChatCompletion {
    pub id: String,
477
    #[schema(example = "1706270835")]
478
    pub created: u64,
479
    #[schema(example = "mistralai/Mistral-7B-Instruct-v0.2")]
480
481
482
483
484
485
    pub model: String,
    pub system_fingerprint: String,
    pub choices: Vec<ChatCompletionComplete>,
    pub usage: Usage,
}

486
#[derive(Clone, Deserialize, Serialize, ToSchema)]
487
488
pub(crate) struct ChatCompletionComplete {
    pub index: u32,
Nicolas Patry's avatar
Nicolas Patry committed
489
    pub message: OutputMessage,
490
    pub logprobs: Option<ChatCompletionLogprobs>,
491
492
493
    pub finish_reason: String,
}

494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
#[derive(Clone, Deserialize, Serialize, ToSchema)]
pub(crate) struct ChatCompletionLogprobs {
    content: Vec<ChatCompletionLogprob>,
}

impl From<(Token, Vec<Token>)> for ChatCompletionLogprobs {
    fn from(value: (Token, Vec<Token>)) -> Self {
        let (token, top_tokens) = value;

        Self {
            content: vec![ChatCompletionLogprob {
                token: token.text,
                logprob: token.logprob,
                top_logprobs: top_tokens
                    .into_iter()
                    .map(|t| ChatCompletionTopLogprob {
                        token: t.text,
                        logprob: t.logprob,
                    })
                    .collect(),
            }],
        }
    }
}

impl From<(Vec<Token>, Vec<Vec<Token>>)> for ChatCompletionLogprobs {
    fn from(value: (Vec<Token>, Vec<Vec<Token>>)) -> Self {
        let (tokens, top_tokens) = value;
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536

        // Create an iterator that produces None for top_tokens once it's exhausted
        let top_tokens_iter = top_tokens
            .into_iter()
            .map(Some)
            .chain(std::iter::repeat(None));

        let content = tokens
            .into_iter()
            .zip(top_tokens_iter)
            .map(|(t, top_t_option)| ChatCompletionLogprob {
                token: t.text,
                logprob: t.logprob,
                top_logprobs: match top_t_option {
                    Some(top_t) => top_t
537
538
539
540
541
542
                        .into_iter()
                        .map(|t| ChatCompletionTopLogprob {
                            token: t.text,
                            logprob: t.logprob,
                        })
                        .collect(),
543
544
545
546
547
548
                    None => vec![], // Handle the case where there are no top tokens
                },
            })
            .collect();

        Self { content }
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
    }
}

#[derive(Clone, Deserialize, Serialize, ToSchema)]
pub(crate) struct ChatCompletionLogprob {
    token: String,
    logprob: f32,
    top_logprobs: Vec<ChatCompletionTopLogprob>,
}

#[derive(Clone, Deserialize, Serialize, ToSchema)]
pub(crate) struct ChatCompletionTopLogprob {
    token: String,
    logprob: f32,
}

565
#[derive(Clone, Deserialize, Serialize, ToSchema, Default)]
566
567
568
569
570
571
pub(crate) struct Usage {
    pub prompt_tokens: u32,
    pub completion_tokens: u32,
    pub total_tokens: u32,
}

572
573
574
575
576
577
578
579
580
#[derive(Clone, Serialize, ToSchema)]
#[serde(tag = "object")]
enum CompletionType {
    #[serde(rename = "chat.completion.chunk")]
    ChatCompletionChunk(ChatCompletionChunk),
    #[serde(rename = "chat.completion")]
    ChatCompletion(ChatCompletion),
}

581
582
583
584
impl ChatCompletion {
    pub(crate) fn new(
        model: String,
        system_fingerprint: String,
drbh's avatar
drbh committed
585
        output: Option<String>,
586
587
588
        created: u64,
        details: Details,
        return_logprobs: bool,
589
        tool_calls: Option<Vec<ToolCall>>,
590
    ) -> Self {
Nicolas Patry's avatar
Nicolas Patry committed
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
        let message = match (output, tool_calls) {
            (Some(content), None) => OutputMessage::ChatMessage(TextMessage {
                role: "assistant".into(),
                content,
            }),
            (None, Some(tool_calls)) => OutputMessage::ToolCall(ToolCallMessage {
                role: "assistant".to_string(),
                tool_calls,
            }),
            (Some(output), Some(_)) => {
                warn!("Received both chat and tool call");
                OutputMessage::ChatMessage(TextMessage {
                    role: "assistant".into(),
                    content: output,
                })
            }
            (None, None) => {
                warn!("Didn't receive an answer");
                OutputMessage::ChatMessage(TextMessage {
                    role: "assistant".into(),
                    content: "".to_string(),
                })
            }
        };
615
616
617
618
619
620
621
        Self {
            id: String::new(),
            created,
            model,
            system_fingerprint,
            choices: vec![ChatCompletionComplete {
                index: 0,
Nicolas Patry's avatar
Nicolas Patry committed
622
                message,
623
                logprobs: return_logprobs
624
                    .then(|| ChatCompletionLogprobs::from((details.tokens, details.top_tokens))),
625
626
627
628
629
630
631
632
633
634
                finish_reason: details.finish_reason.to_string(),
            }],
            usage: Usage {
                prompt_tokens: details.prefill.len() as u32,
                completion_tokens: details.generated_tokens,
                total_tokens: details.prefill.len() as u32 + details.generated_tokens,
            },
        }
    }
}
635
#[derive(Clone, Serialize, ToSchema)]
636
637
pub(crate) struct ChatCompletionChunk {
    pub id: String,
638
    #[schema(example = "1706270978")]
639
    pub created: u64,
640
    #[schema(example = "mistralai/Mistral-7B-Instruct-v0.2")]
641
642
643
644
645
    pub model: String,
    pub system_fingerprint: String,
    pub choices: Vec<ChatCompletionChoice>,
}

646
#[derive(Clone, Serialize, ToSchema)]
647
648
649
pub(crate) struct ChatCompletionChoice {
    pub index: u32,
    pub delta: ChatCompletionDelta,
650
    pub logprobs: Option<ChatCompletionLogprobs>,
651
652
653
    pub finish_reason: Option<String>,
}

Nicolas Patry's avatar
Nicolas Patry committed
654
655
656
657
658
659
660
#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
pub struct ToolCallDelta {
    #[schema(example = "assistant")]
    role: String,
    tool_calls: DeltaToolCall,
}

661
662
#[derive(Clone, Debug, Serialize, ToSchema)]
#[serde(untagged)]
Nicolas Patry's avatar
Nicolas Patry committed
663
664
665
enum ChatCompletionDelta {
    Chat(TextMessage),
    Tool(ToolCallDelta),
drbh's avatar
drbh committed
666
667
}

Nicolas Patry's avatar
Nicolas Patry committed
668
#[derive(Clone, Deserialize, Serialize, ToSchema, Debug, PartialEq)]
drbh's avatar
drbh committed
669
670
671
672
673
674
675
pub(crate) struct DeltaToolCall {
    pub index: u32,
    pub id: String,
    pub r#type: String,
    pub function: Function,
}

Nicolas Patry's avatar
Nicolas Patry committed
676
#[derive(Clone, Deserialize, Serialize, ToSchema, Debug, PartialEq)]
drbh's avatar
drbh committed
677
678
679
pub(crate) struct Function {
    pub name: Option<String>,
    pub arguments: String,
680
681
}

drbh's avatar
drbh committed
682
#[allow(clippy::too_many_arguments)]
683
684
685
686
impl ChatCompletionChunk {
    pub(crate) fn new(
        model: String,
        system_fingerprint: String,
drbh's avatar
drbh committed
687
688
        delta: Option<String>,
        tool_calls: Option<Vec<String>>,
689
        created: u64,
690
        logprobs: Option<ChatCompletionLogprobs>,
691
692
        finish_reason: Option<String>,
    ) -> Self {
693
        let delta = match (delta, tool_calls) {
Nicolas Patry's avatar
Nicolas Patry committed
694
695
696
697
698
699
700
            (Some(delta), _) => ChatCompletionDelta::Chat(TextMessage {
                role: "assistant".to_string(),
                content: delta,
            }),
            (None, Some(tool_calls)) => ChatCompletionDelta::Tool(ToolCallDelta {
                role: "assistant".to_string(),
                tool_calls: DeltaToolCall {
701
702
703
704
705
706
707
                    index: 0,
                    id: String::new(),
                    r#type: "function".to_string(),
                    function: Function {
                        name: None,
                        arguments: tool_calls[0].to_string(),
                    },
Nicolas Patry's avatar
Nicolas Patry committed
708
709
710
711
712
713
                },
            }),
            (None, None) => ChatCompletionDelta::Chat(TextMessage {
                role: "assistant".to_string(),
                content: "".to_string(),
            }),
714
        };
715
716
717
718
719
720
        Self {
            id: String::new(),
            created,
            model,
            system_fingerprint,
            choices: vec![ChatCompletionChoice {
721
                index: 0,
722
                delta,
723
724
725
726
727
728
729
730
731
                logprobs,
                finish_reason,
            }],
        }
    }
}

#[derive(Clone, Deserialize, ToSchema, Serialize)]
pub(crate) struct ChatRequest {
732
    #[schema(example = "mistralai/Mistral-7B-Instruct-v0.2")]
drbh's avatar
drbh committed
733
    /// [UNUSED] ID of the model to use. See the model endpoint compatibility table for details on which models work with the Chat API.
734
    pub model: String,
drbh's avatar
drbh committed
735

736
    /// A list of messages comprising the conversation so far.
drbh's avatar
drbh committed
737
    #[schema(example = "[{\"role\": \"user\", \"content\": \"What is Deep Learning?\"}]")]
738
739
740
741
742
    pub messages: Vec<Message>,

    /// Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far,
    /// decreasing the model's likelihood to repeat the same line verbatim.
    #[serde(default)]
743
    #[schema(example = "1.0")]
744
745
746
747
748
749
750
751
752
753
754
755
756
757
    pub frequency_penalty: Option<f32>,

    /// UNUSED
    /// Modify the likelihood of specified tokens appearing in the completion. Accepts a JSON object that maps tokens
    /// (specified by their token ID in the tokenizer) to an associated bias value from -100 to 100. Mathematically,
    /// the bias is added to the logits generated by the model prior to sampling. The exact effect will vary per model,
    /// but values between -1 and 1 should decrease or increase likelihood of selection; values like -100 or 100 should
    /// result in a ban or exclusive selection of the relevant token.
    #[serde(default)]
    pub logit_bias: Option<Vec<f32>>,

    /// Whether to return log probabilities of the output tokens or not. If true, returns the log probabilities of each
    /// output token returned in the content of message.
    #[serde(default)]
758
    #[schema(example = "false")]
759
760
761
762
763
    pub logprobs: Option<bool>,

    /// An integer between 0 and 5 specifying the number of most likely tokens to return at each token position, each with
    /// an associated log probability. logprobs must be set to true if this parameter is used.
    #[serde(default)]
764
    #[schema(example = "5")]
765
766
767
768
    pub top_logprobs: Option<u32>,

    /// The maximum number of tokens that can be generated in the chat completion.
    #[serde(default)]
769
    #[schema(example = "32")]
770
771
772
773
774
775
    pub max_tokens: Option<u32>,

    /// UNUSED
    /// How many chat completion choices to generate for each input message. Note that you will be charged based on the
    /// number of generated tokens across all of the choices. Keep n as 1 to minimize costs.
    #[serde(default)]
776
    #[schema(nullable = true, example = "2")]
777
778
779
780
781
    pub n: Option<u32>,

    /// Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far,
    /// increasing the model's likelihood to talk about new topics
    #[serde(default)]
782
    #[schema(nullable = true, example = 0.1)]
783
784
    pub presence_penalty: Option<f32>,

785
786
787
788
789
    /// Up to 4 sequences where the API will stop generating further tokens.
    #[serde(default)]
    #[schema(nullable = true, example = "null")]
    pub stop: Option<Vec<String>>,

790
791
792
793
794
    #[serde(default = "bool::default")]
    pub stream: bool,

    #[schema(nullable = true, example = 42)]
    pub seed: Option<u64>,
795
796
797
798
799
800

    /// What sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the output more random, while
    /// lower values like 0.2 will make it more focused and deterministic.
    ///
    /// We generally recommend altering this or `top_p` but not both.
    #[serde(default)]
801
    #[schema(nullable = true, example = 1.0)]
802
803
804
805
806
    pub temperature: Option<f32>,

    /// An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the
    /// tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered.
    #[serde(default)]
807
    #[schema(nullable = true, example = 0.95)]
808
    pub top_p: Option<f32>,
drbh's avatar
drbh committed
809
810
811
812
813
814
815
816
817
818
819

    /// A list of tools the model may call. Currently, only functions are supported as a tool. Use this to provide a list of
    /// functions the model may generate JSON inputs for.
    #[serde(default)]
    #[schema(nullable = true, example = "null")]
    pub tools: Option<Vec<Tool>>,

    /// A prompt to be appended before the tools
    #[serde(default = "default_tool_prompt")]
    #[schema(
        nullable = true,
820
        example = "\"You will be presented with a JSON schema representing a set of tools.\nIf the user request lacks of sufficient information to make a precise tool selection: Do not invent any tool's properties, instead notify with an error message.\n\nJSON Schema:\n\""
drbh's avatar
drbh committed
821
822
823
824
825
826
827
    )]
    pub tool_prompt: Option<String>,

    /// A specific tool to use. If not provided, the model will default to use any of the tools provided in the tools parameter.
    #[serde(default)]
    #[schema(nullable = true, example = "null")]
    pub tool_choice: Option<ToolType>,
drbh's avatar
drbh committed
828
829
830
831
832
833
834

    /// Response format constraints for the generation.
    ///
    /// NOTE: A request can use `response_format` OR `tools` but not both.
    #[serde(default)]
    #[schema(nullable = true, default = "null", example = "null")]
    pub response_format: Option<GrammarType>,
drbh's avatar
drbh committed
835
836
837
838
}

fn default_tool_prompt() -> Option<String> {
    Some(
839
        "\nYou will be presented with a JSON schema representing a set of tools.\nIf the user request lacks of sufficient information to make a precise tool selection: Do not invent any tool's properties, instead notify with an error message.\n\nJSON Schema:\n".to_string(),
drbh's avatar
drbh committed
840
841
    )
}
842
843
844
845

#[derive(Clone, Debug, Deserialize, PartialEq, Serialize, ToSchema)]
#[serde(untagged)]
pub enum ToolType {
drbh's avatar
drbh committed
846
    OneOf,
847
848
    FunctionName(String),
    Function { function: FunctionName },
drbh's avatar
drbh committed
849
850
}

851
852
853
854
855
856
857
858
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
pub struct FunctionName {
    pub name: String,
}

#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
#[serde(from = "ToolTypeDeserializer")]
pub struct ToolChoice(pub Option<ToolType>);
drbh's avatar
drbh committed
859

860
861
862
863
864
865
#[derive(Deserialize)]
#[serde(untagged)]
enum ToolTypeDeserializer {
    None(Option<String>),
    Some(ToolType),
}
drbh's avatar
drbh committed
866

867
868
impl From<ToolTypeDeserializer> for ToolChoice {
    fn from(value: ToolTypeDeserializer) -> Self {
drbh's avatar
drbh committed
869
        match value {
870
871
872
873
874
            ToolTypeDeserializer::None(opt) => match opt.as_deref() {
                Some("none") => ToolChoice(None),
                Some("auto") => ToolChoice(Some(ToolType::OneOf)),
                Some(s) => ToolChoice(Some(ToolType::FunctionName(s.to_string()))),
                None => ToolChoice(Some(ToolType::OneOf)),
drbh's avatar
drbh committed
875
            },
876
            ToolTypeDeserializer::Some(tool_type) => ToolChoice(Some(tool_type)),
drbh's avatar
drbh committed
877
878
879
880
        }
    }
}

881
#[derive(Debug, Deserialize, Serialize, ToSchema, PartialEq)]
drbh's avatar
drbh committed
882
883
884
885
886
887
pub struct Tools {
    #[serde(flatten)]
    functions_map: FunctionsMap,
    properties: Properties,
}

888
#[derive(Debug, Serialize, Deserialize, PartialEq)]
drbh's avatar
drbh committed
889
890
891
892
893
struct FunctionsMap {
    #[serde(rename = "$functions")]
    functions: std::collections::HashMap<String, serde_json::Value>,
}

894
#[derive(Debug, Serialize, Deserialize, PartialEq)]
drbh's avatar
drbh committed
895
896
897
898
899
struct FunctionRef {
    #[serde(rename = "$ref")]
    ref_path: String,
}

900
#[derive(Debug, Serialize, Deserialize, PartialEq)]
drbh's avatar
drbh committed
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
struct Properties {
    #[serde(serialize_with = "serialize_function")]
    function: Vec<FunctionRef>,
}

fn serialize_function<S>(functions: &Vec<FunctionRef>, serializer: S) -> Result<S::Ok, S::Error>
where
    S: serde::Serializer,
{
    use serde::ser::SerializeStruct;
    let mut state = serializer.serialize_struct("Function", 1)?;
    state.serialize_field("anyOf", functions)?;
    state.end()
}

Nicolas Patry's avatar
Nicolas Patry committed
916
#[derive(Clone, Debug, Deserialize, Serialize, ToSchema, Default, PartialEq)]
drbh's avatar
drbh committed
917
918
919
920
pub(crate) struct FunctionDefinition {
    #[serde(default)]
    pub description: Option<String>,
    pub name: String,
921
922
    #[serde(alias = "parameters")]
    pub arguments: serde_json::Value,
drbh's avatar
drbh committed
923
924
925
926
927
928
929
930
931
}

#[derive(Clone, Debug, Deserialize, Serialize, ToSchema)]
pub(crate) struct Tool {
    // The type of the tool. Currently, only 'function' is supported.
    #[schema(example = "function")]
    pub r#type: String,
    // Grab the tool as generic JSON for debugging purposes.
    pub function: FunctionDefinition,
932
933
}

934
#[derive(Clone, Serialize, Deserialize, Default)]
935
pub(crate) struct ChatTemplateInputs<'a> {
Nicolas Patry's avatar
Nicolas Patry committed
936
    messages: Vec<TextMessage>,
937
938
    bos_token: Option<&'a str>,
    eos_token: Option<&'a str>,
939
    add_generation_prompt: bool,
940
941
    tools: Option<&'a str>,
    tools_prompt: Option<&'a str>,
942
943
}

Nicolas Patry's avatar
Nicolas Patry committed
944
#[derive(Clone, Deserialize, Serialize, ToSchema, Default, Debug, PartialEq)]
drbh's avatar
drbh committed
945
pub(crate) struct ToolCall {
946
    pub id: String,
drbh's avatar
drbh committed
947
948
949
950
    pub r#type: String,
    pub function: FunctionDefinition,
}

Nicolas Patry's avatar
Nicolas Patry committed
951
#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
952
pub struct Url {
Nicolas Patry's avatar
Nicolas Patry committed
953
    url: String,
drbh's avatar
drbh committed
954
955
}

Nicolas Patry's avatar
Nicolas Patry committed
956
957
958
#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
#[serde(tag = "type")]
#[serde(rename_all = "snake_case")]
959
960
961
pub enum MessageChunk {
    Text { text: String },
    ImageUrl { image_url: Url },
Nicolas Patry's avatar
Nicolas Patry committed
962
963
964
965
966
967
968
}

#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
pub struct Message {
    #[schema(example = "user")]
    role: String,
    #[schema(example = "My name is David and I")]
969
    pub content: MessageContent,
drbh's avatar
drbh committed
970
    #[serde(default, skip_serializing_if = "Option::is_none")]
Nicolas Patry's avatar
Nicolas Patry committed
971
972
    #[schema(example = "\"David\"")]
    name: Option<String>,
drbh's avatar
drbh committed
973
974
}

975
976
977
978
979
980
981
982
983
984
985
986
987
988
#[derive(Clone, Deserialize, Serialize, ToSchema, Debug, PartialEq)]
#[serde(untagged)]
pub enum MessageContent {
    SingleText(String),
    MultipleChunks(Vec<MessageChunk>),
}

// Pushing a chunk to a single text message will convert it to a multiple chunks message
impl MessageContent {
    pub fn push(&mut self, chunk: MessageChunk) {
        match self {
            MessageContent::SingleText(text) => {
                *self =
                    MessageContent::MultipleChunks(vec![MessageChunk::Text { text: text.clone() }]);
Nicolas Patry's avatar
Nicolas Patry committed
989
            }
990
991
992
993
            MessageContent::MultipleChunks(chunks) => {
                chunks.push(chunk);
            }
        }
drbh's avatar
drbh committed
994
995
996
    }
}

Nicolas Patry's avatar
Nicolas Patry committed
997
998
#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
pub struct TextMessage {
999
1000
1001
    #[schema(example = "user")]
    pub role: String,
    #[schema(example = "My name is David and I")]
Nicolas Patry's avatar
Nicolas Patry committed
1002
1003
1004
1005
1006
1007
1008
    pub content: String,
}

impl From<Message> for TextMessage {
    fn from(value: Message) -> Self {
        TextMessage {
            role: value.role,
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
            content: match value.content {
                MessageContent::SingleText(text) => text,
                MessageContent::MultipleChunks(chunks) => chunks
                    .into_iter()
                    .map(|chunk| match chunk {
                        MessageChunk::Text { text } => text,
                        MessageChunk::ImageUrl { image_url } => format!("![]({})", image_url.url),
                    })
                    .collect::<Vec<_>>()
                    .join(""),
            },
Nicolas Patry's avatar
Nicolas Patry committed
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
        }
    }
}

#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
pub struct ToolCallMessage {
    #[schema(example = "assistant")]
    role: String,
    tool_calls: Vec<ToolCall>,
}

#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
#[serde(untagged)]
pub(crate) enum OutputMessage {
    ChatMessage(TextMessage),
    ToolCall(ToolCallMessage),
1036
1037
}

1038
#[derive(Clone, Debug, Deserialize, ToSchema)]
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
1039
pub(crate) struct GenerateRequest {
1040
    #[schema(example = "My name is Olivier and I")]
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
1041
1042
1043
1044
1045
    pub inputs: String,
    #[serde(default = "default_parameters")]
    pub parameters: GenerateParameters,
}

1046
1047
1048
1049
1050
1051
1052
#[derive(Clone, Debug, Deserialize, ToSchema)]
pub(crate) struct CompatGenerateRequest {
    #[schema(example = "My name is Olivier and I")]
    pub inputs: String,
    #[serde(default = "default_parameters")]
    pub parameters: GenerateParameters,
    #[serde(default)]
OlivierDehaene's avatar
OlivierDehaene committed
1053
    #[schema(default = "false")]
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
    pub stream: bool,
}

impl From<CompatGenerateRequest> for GenerateRequest {
    fn from(req: CompatGenerateRequest) -> Self {
        Self {
            inputs: req.inputs,
            parameters: req.parameters,
        }
    }
}

1066
1067
1068
1069
1070
1071
#[derive(Debug, Serialize, ToSchema)]
pub struct PrefillToken {
    #[schema(example = 0)]
    id: u32,
    #[schema(example = "test")]
    text: String,
1072
    #[schema(nullable = true, example = - 0.34)]
1073
1074
1075
    logprob: f32,
}

1076
#[derive(Debug, Serialize, ToSchema, Clone)]
1077
1078
1079
1080
1081
pub struct Token {
    #[schema(example = 0)]
    id: u32,
    #[schema(example = "test")]
    text: String,
1082
    #[schema(nullable = true, example = - 0.34)]
1083
    logprob: f32,
1084
1085
    #[schema(example = "false")]
    special: bool,
1086
1087
}

1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
#[derive(Debug, Serialize, ToSchema)]
pub struct SimpleToken {
    #[schema(example = 0)]
    id: u32,
    #[schema(example = "test")]
    text: String,
    #[schema(example = 0)]
    start: usize,
    #[schema(example = 2)]
    stop: usize,
}

OlivierDehaene's avatar
OlivierDehaene committed
1100
#[derive(Debug, Serialize, ToSchema)]
1101
#[serde(rename_all(serialize = "snake_case"))]
1102
#[schema(example = "Length")]
1103
1104
1105
1106
1107
1108
1109
1110
1111
pub(crate) enum FinishReason {
    #[schema(rename = "length")]
    Length,
    #[serde(rename = "eos_token")]
    #[schema(rename = "eos_token")]
    EndOfSequenceToken,
    #[schema(rename = "stop_sequence")]
    StopSequence,
}
1112

1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
impl std::fmt::Display for FinishReason {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self {
            FinishReason::Length => write!(f, "length"),
            FinishReason::EndOfSequenceToken => write!(f, "eos_token"),
            FinishReason::StopSequence => write!(f, "stop_sequence"),
        }
    }
}

1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
#[derive(Serialize, ToSchema)]
pub(crate) struct BestOfSequence {
    #[schema(example = "test")]
    pub generated_text: String,
    #[schema(example = "length")]
    pub finish_reason: FinishReason,
    #[schema(example = 1)]
    pub generated_tokens: u32,
    #[schema(nullable = true, example = 42)]
    pub seed: Option<u64>,
    pub prefill: Vec<PrefillToken>,
    pub tokens: Vec<Token>,
Nicolas Patry's avatar
Nicolas Patry committed
1135
1136
    #[serde(skip_serializing_if = "Vec::is_empty")]
    pub top_tokens: Vec<Vec<Token>>,
1137
1138
}

1139
#[derive(Serialize, ToSchema)]
OlivierDehaene's avatar
OlivierDehaene committed
1140
pub(crate) struct Details {
1141
1142
1143
    #[schema(example = "length")]
    pub finish_reason: FinishReason,
    #[schema(example = 1)]
OlivierDehaene's avatar
OlivierDehaene committed
1144
    pub generated_tokens: u32,
1145
    #[schema(nullable = true, example = 42)]
1146
    pub seed: Option<u64>,
1147
1148
    pub prefill: Vec<PrefillToken>,
    pub tokens: Vec<Token>,
1149
1150
    #[serde(skip_serializing_if = "Option::is_none")]
    pub best_of_sequences: Option<Vec<BestOfSequence>>,
Nicolas Patry's avatar
Nicolas Patry committed
1151
1152
    #[serde(skip_serializing_if = "Vec::is_empty")]
    pub top_tokens: Vec<Vec<Token>>,
OlivierDehaene's avatar
OlivierDehaene committed
1153
1154
}

1155
#[derive(Serialize, ToSchema)]
1156
pub(crate) struct GenerateResponse {
1157
    #[schema(example = "test")]
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
1158
    pub generated_text: String,
OlivierDehaene's avatar
OlivierDehaene committed
1159
1160
    #[serde(skip_serializing_if = "Option::is_none")]
    pub details: Option<Details>,
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
1161
}
1162

1163
1164
1165
1166
#[derive(Serialize, ToSchema)]
#[serde(transparent)]
pub(crate) struct TokenizeResponse(Vec<SimpleToken>);

1167
1168
1169
1170
1171
1172
#[derive(Serialize, ToSchema)]
pub(crate) struct StreamDetails {
    #[schema(example = "length")]
    pub finish_reason: FinishReason,
    #[schema(example = 1)]
    pub generated_tokens: u32,
1173
    #[schema(nullable = true, example = 42)]
1174
1175
1176
1177
    pub seed: Option<u64>,
}

#[derive(Serialize, ToSchema)]
1178
pub(crate) struct StreamResponse {
1179
    pub index: u32,
1180
    pub token: Token,
Nicolas Patry's avatar
Nicolas Patry committed
1181
1182
    #[serde(skip_serializing_if = "Vec::is_empty")]
    pub top_tokens: Vec<Token>,
1183
    #[schema(nullable = true, default = "null", example = "test")]
1184
    pub generated_text: Option<String>,
1185
1186
    #[schema(nullable = true, default = "null")]
    pub details: Option<StreamDetails>,
1187
1188
}

1189
#[derive(Serialize, ToSchema)]
1190
1191
pub(crate) struct ErrorResponse {
    pub error: String,
1192
    pub error_type: String,
1193
}
1194
1195

#[cfg(test)]
1196
mod tests {
1197
    use super::*;
Nicolas Patry's avatar
Nicolas Patry committed
1198
    use serde_json::json;
1199
1200
    use tokenizers::Tokenizer;

1201
    pub(crate) async fn get_tokenizer() -> Tokenizer {
1202
1203
1204
1205
        let api = hf_hub::api::sync::Api::new().unwrap();
        let repo = api.model("gpt2".to_string());
        let filename = repo.get("tokenizer.json").unwrap();
        Tokenizer::from_file(filename).unwrap()
1206
    }
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220

    #[test]
    fn test_hub_nested_tokens_tokenizer_config() {
        // this is a subset of the tokenizer.json file
        // in this case we expect the tokens to be encoded as simple strings
        let json_content = r#"{
            "chat_template": "test",
            "bos_token": "<|begin▁of▁sentence|>",
            "eos_token": "<|end▁of▁sentence|>"
        }"#;

        let config: HubTokenizerConfig = serde_json::from_str(json_content).unwrap();

        // check that we successfully parsed the tokens
1221
1222
1223
1224
        assert_eq!(
            config.chat_template,
            Some(ChatTemplateVersions::Single("test".to_string()))
        );
1225
1226
        assert_eq!(
            config.bos_token,
1227
1228
1229
1230
1231
1232
1233
1234
1235
            Some(TokenizerConfigToken::String(
                "<|begin▁of▁sentence|>".to_string()
            ))
        );
        assert_eq!(
            config.eos_token,
            Some(TokenizerConfigToken::String(
                "<|end▁of▁sentence|>".to_string()
            ))
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
        );

        // in this case we expect the tokens to be encoded as structured tokens
        // we want the content of the structured token
        let json_content = r#"{
            "chat_template": "test",
            "bos_token": {
              "__type": "AddedToken",
              "content": "<|begin▁of▁sentence|>",
              "lstrip": false,
              "normalized": true,
              "rstrip": false,
              "single_word": false
            },
            "eos_token": {
              "__type": "AddedToken",
              "content": "<|end▁of▁sentence|>",
              "lstrip": false,
              "normalized": true,
              "rstrip": false,
              "single_word": false
            }
        }"#;

        let config: HubTokenizerConfig = serde_json::from_str(json_content).unwrap();

        // check that we successfully parsed the tokens
1263
1264
1265
1266
        assert_eq!(
            config.chat_template,
            Some(ChatTemplateVersions::Single("test".to_string()))
        );
1267
1268
        assert_eq!(
            config.bos_token,
1269
1270
1271
1272
1273
1274
1275
1276
1277
            Some(TokenizerConfigToken::Object {
                content: "<|begin▁of▁sentence|>".to_string()
            })
        );
        assert_eq!(
            config.eos_token,
            Some(TokenizerConfigToken::Object {
                content: "<|end▁of▁sentence|>".to_string()
            })
1278
1279
        );
    }
Nicolas Patry's avatar
Nicolas Patry committed
1280
1281
1282

    #[test]
    fn test_chat_simple_string() {
Nicolas Patry's avatar
Nicolas Patry committed
1283
        let json = json!({
Nicolas Patry's avatar
Nicolas Patry committed
1284
            "model": "",
Nicolas Patry's avatar
Nicolas Patry committed
1285
1286
            "messages": [{
                "role": "user",
Nicolas Patry's avatar
Nicolas Patry committed
1287
                "content": "What is Deep Learning?"
Nicolas Patry's avatar
Nicolas Patry committed
1288
            }]
Nicolas Patry's avatar
Nicolas Patry committed
1289
1290
1291
1292
1293
1294
1295
        });
        let request: ChatRequest = serde_json::from_str(json.to_string().as_str()).unwrap();

        assert_eq!(
            request.messages[0],
            Message {
                role: "user".to_string(),
1296
                content: MessageContent::SingleText("What is Deep Learning?".to_string()),
Nicolas Patry's avatar
Nicolas Patry committed
1297
1298
1299
1300
1301
1302
1303
                name: None
            }
        );
    }

    #[test]
    fn test_chat_request() {
Nicolas Patry's avatar
Nicolas Patry committed
1304
        let json = json!({
Nicolas Patry's avatar
Nicolas Patry committed
1305
            "model": "",
Nicolas Patry's avatar
Nicolas Patry committed
1306
1307
            "messages": [{
                "role": "user",
Nicolas Patry's avatar
Nicolas Patry committed
1308
1309
                "content": [
                    {"type": "text", "text": "Whats in this image?"},
Nicolas Patry's avatar
Nicolas Patry committed
1310
                    {"type": "image_url", "image_url": {"url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rabbit.png"}},
Nicolas Patry's avatar
Nicolas Patry committed
1311
                ]
Nicolas Patry's avatar
Nicolas Patry committed
1312
            }]
Nicolas Patry's avatar
Nicolas Patry committed
1313
1314
1315
1316
1317
1318
1319
        });
        let request: ChatRequest = serde_json::from_str(json.to_string().as_str()).unwrap();

        assert_eq!(
            request.messages[0],
            Message{
                role: "user".to_string(),
1320
1321
1322
1323
                content: MessageContent::MultipleChunks(vec![
                    MessageChunk::Text { text: "Whats in this image?".to_string() },
                    MessageChunk::ImageUrl { image_url: Url { url: "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rabbit.png".to_string() }},
                ]),
Nicolas Patry's avatar
Nicolas Patry committed
1324
1325
1326
1327
                name: None
            }
        );
    }
Nicolas Patry's avatar
Nicolas Patry committed
1328
1329
1330
1331
1332

    #[test]
    fn text_message_convert() {
        let message = Message{
                role: "user".to_string(),
1333
1334
1335
1336
                content: MessageContent::MultipleChunks(vec![
                    MessageChunk::Text { text: "Whats in this image?".to_string() },
                    MessageChunk::ImageUrl { image_url: Url { url: "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rabbit.png".to_string() } }
                ]),
Nicolas Patry's avatar
Nicolas Patry committed
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
                name: None
            };
        let textmsg: TextMessage = message.into();
        assert_eq!(textmsg.content, "Whats in this image?![](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rabbit.png)");
    }
    #[test]
    fn openai_output() {
        let message = OutputMessage::ChatMessage(TextMessage {
            role: "assistant".to_string(),
            content: "This is the answer".to_string(),
        });
        let serialized = serde_json::to_string(&message).unwrap();
        assert_eq!(
            serialized,
            r#"{"role":"assistant","content":"This is the answer"}"#
        );

        let message = OutputMessage::ToolCall(ToolCallMessage {
            role: "assistant".to_string(),
            tool_calls: vec![ToolCall {
                id: "0".to_string(),
                r#type: "function".to_string(),
                function: FunctionDefinition {
                    description: None,
                    name: "myfn".to_string(),
                    arguments: json!({
                        "format": "csv"
                    }),
                },
            }],
        });
        let serialized = serde_json::to_string(&message).unwrap();
        assert_eq!(
            serialized,
            r#"{"role":"assistant","tool_calls":[{"id":"0","type":"function","function":{"description":null,"name":"myfn","arguments":{"format":"csv"}}}]}"#
        );
    }
1374
}