lib.rs 43.1 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

Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
7
use serde::{Deserialize, Serialize};
Nicolas Patry's avatar
Nicolas Patry committed
8
use tracing::warn;
9
use utoipa::ToSchema;
Olivier Dehaene's avatar
Olivier Dehaene committed
10
use validation::Validation;
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
11

drbh's avatar
drbh committed
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
#[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>,
}

31
32
/// Hub type
#[derive(Clone, Debug, Deserialize)]
33
pub struct HubModelInfo {
34
35
36
37
38
39
    #[serde(rename(deserialize = "id"))]
    pub model_id: String,
    pub sha: Option<String>,
    pub pipeline_tag: Option<String>,
}

40
41
42
43
44
45
46
47
48
49
50
51
52
53
#[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>),
}

#[derive(Debug, Clone, Deserialize, Default)]
54
pub struct HubTokenizerConfig {
55
    pub chat_template: Option<ChatTemplateVersions>,
56
    pub completion_template: Option<String>,
57
    #[serde(deserialize_with = "token_serde::deserialize")]
58
    pub bos_token: Option<String>,
59
    #[serde(deserialize_with = "token_serde::deserialize")]
60
    pub eos_token: Option<String>,
61
62
63
}

impl HubTokenizerConfig {
64
65
66
    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()
67
68
69
    }
}

drbh's avatar
drbh committed
70
71
72
73
74
75
76
77
78
79
80
81
82
83
#[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 {
    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()
    }
}

84
#[derive(Clone, Debug, Deserialize, ToSchema, Serialize)]
drbh's avatar
drbh committed
85
86
#[serde(tag = "type", content = "value")]
pub(crate) enum GrammarType {
87
88
89
90
91
    /// 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
92
    #[serde(alias = "json_object")]
93
94
    #[schema(example = json ! ({"properties": {"location":{"type": "string"}}}))]
    Json(serde_json::Value),
drbh's avatar
drbh committed
95
96
97
98
    #[serde(rename = "regex")]
    Regex(String),
}

99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
mod token_serde {
    use super::*;
    use serde::de;
    use serde::Deserializer;
    use serde_json::Value;

    pub fn deserialize<'de, D>(deserializer: D) -> Result<Option<String>, D::Error>
    where
        D: Deserializer<'de>,
    {
        let value = Value::deserialize(deserializer)?;

        match value {
            Value::String(s) => Ok(Some(s)),
            Value::Object(map) => {
                if let Some(content) = map.get("content").and_then(|v| v.as_str()) {
                    Ok(Some(content.to_string()))
                } else {
                    Err(de::Error::custom(
                        "content key not found in structured token",
                    ))
                }
            }
122
            Value::Null => Ok(None),
123
124
125
126
127
            _ => Err(de::Error::custom("invalid token format")),
        }
    }
}

128
129
#[derive(Clone, Debug, Serialize, ToSchema)]
pub struct Info {
130
    /// Model info
131
132
133
134
    #[schema(example = "bigscience/blomm-560m")]
    pub model_id: String,
    #[schema(nullable = true, example = "e985a63cdc139290c5f700ff1929f0b5942cced2")]
    pub model_sha: Option<String>,
135
136
137
138
    #[schema(example = "torch.float16")]
    pub model_dtype: String,
    #[schema(example = "cuda")]
    pub model_device_type: String,
139
140
    #[schema(nullable = true, example = "text-generation")]
    pub model_pipeline_tag: Option<String>,
141
142
143
144
145
146
147
148
    /// 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
149
    pub max_input_tokens: usize,
150
151
152
153
154
155
156
157
    #[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,
158
159
    #[schema(nullable = true, example = "null")]
    pub max_batch_size: Option<usize>,
160
161
    #[schema(example = "2")]
    pub validation_workers: usize,
162
163
    #[schema(example = "32")]
    pub max_client_batch_size: usize,
164
    /// Router Info
165
166
    #[schema(example = "text-generation-router")]
    pub router: &'static str,
167
168
169
170
    #[schema(example = "0.5.0")]
    pub version: &'static str,
    #[schema(nullable = true, example = "null")]
    pub sha: Option<&'static str>,
171
172
    #[schema(nullable = true, example = "null")]
    pub docker_label: Option<&'static str>,
173
174
}

drbh's avatar
drbh committed
175
#[derive(Clone, Debug, Deserialize, ToSchema, Default)]
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
176
pub(crate) struct GenerateParameters {
177
    /// Generate best_of sequences and return the one if the highest token logprobs.
178
179
180
    #[serde(default)]
    #[schema(exclusive_minimum = 0, nullable = true, default = "null", example = 1)]
    pub best_of: Option<usize>,
181
182

    /// The value used to module the logits distribution.
183
184
185
186
187
188
189
190
    #[serde(default)]
    #[schema(
        exclusive_minimum = 0.0,
        nullable = true,
        default = "null",
        example = 0.5
    )]
    pub temperature: Option<f32>,
191
192
193

    /// The parameter for repetition penalty. 1.0 means no penalty.
    /// See [this paper](https://arxiv.org/pdf/1909.05858.pdf) for more details.
194
195
196
197
198
199
200
201
    #[serde(default)]
    #[schema(
        exclusive_minimum = 0.0,
        nullable = true,
        default = "null",
        example = 1.03
    )]
    pub repetition_penalty: Option<f32>,
202
203
204
205

    /// 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.
206
    #[serde(default)]
207
208
209
210
211
212
213
    #[schema(
        exclusive_minimum = -2.0,
        nullable = true,
        default = "null",
        example = 0.1
    )]
    pub frequency_penalty: Option<f32>,
214
215

    /// The number of highest probability vocabulary tokens to keep for top-k-filtering.
216
    #[serde(default)]
217
218
    #[schema(exclusive_minimum = 0, nullable = true, default = "null", example = 10)]
    pub top_k: Option<i32>,
219
220

    /// Top-p value for nucleus sampling.
221
222
223
224
225
226
227
228
229
    #[serde(default)]
    #[schema(
        exclusive_minimum = 0.0,
        maximum = 1.0,
        nullable = true,
        default = "null",
        example = 0.95
    )]
    pub top_p: Option<f32>,
230
231
232

    /// Typical Decoding mass
    /// See [Typical Decoding for Natural Language Generation](https://arxiv.org/abs/2202.00666) for more information.
233
    #[serde(default)]
234
235
236
237
238
239
240
241
    #[schema(
        exclusive_minimum = 0.0,
        maximum = 1.0,
        nullable = true,
        default = "null",
        example = 0.95
    )]
    pub typical_p: Option<f32>,
242
243

    /// Activate logits sampling.
244
    #[serde(default)]
245
    #[schema(default = "false", example = true)]
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
246
    pub do_sample: bool,
247
248

    /// Maximum number of tokens to generate.
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
249
    #[serde(default = "default_max_new_tokens")]
250
    #[schema(nullable = true, default = "100", example = "20")]
251
    pub max_new_tokens: Option<u32>,
252
253

    /// Whether to prepend the prompt to the generated text
OlivierDehaene's avatar
OlivierDehaene committed
254
    #[serde(default)]
255
    #[schema(nullable = true, default = "null", example = false)]
256
    pub return_full_text: Option<bool>,
257
258

    /// Stop generating tokens if a member of `stop` is generated.
259
    #[serde(default)]
260
    #[schema(inline, max_items = 4, example = json ! (["photographer"]))]
261
    pub stop: Vec<String>,
262
263

    /// Truncate inputs tokens to the given size.
OlivierDehaene's avatar
OlivierDehaene committed
264
    #[serde(default)]
265
    #[schema(nullable = true, default = "null", example = "null")]
266
    pub truncate: Option<usize>,
267
268

    /// Watermarking with [A Watermark for Large Language Models](https://arxiv.org/abs/2301.10226).
269
    #[serde(default)]
270
271
    #[schema(default = "false", example = true)]
    pub watermark: bool,
272
273

    /// Whether to return generation details.
274
    #[serde(default)]
275
    #[schema(default = "true")]
OlivierDehaene's avatar
OlivierDehaene committed
276
    pub details: bool,
277
278

    /// Whether to return decoder input token logprobs and ids.
279
    #[serde(default)]
280
    #[schema(default = "false")]
281
    pub decoder_input_details: bool,
282
283

    /// Random sampling seed.
284
    #[serde(default)]
285
286
287
288
289
290
    #[schema(
        exclusive_minimum = 0,
        nullable = true,
        default = "null",
        example = "null"
    )]
291
    pub seed: Option<u64>,
292
293

    /// The number of highest probability vocabulary tokens to keep for top-n-filtering.
Nicolas Patry's avatar
Nicolas Patry committed
294
295
296
    #[serde(default)]
    #[schema(exclusive_minimum = 0, nullable = true, default = "null", example = 5)]
    pub top_n_tokens: Option<u32>,
297
298

    /// Grammar constraints for the generation.
drbh's avatar
drbh committed
299
    #[serde(default)]
300
    #[schema(nullable = true, default = "null", example = "null")]
drbh's avatar
drbh committed
301
    pub grammar: Option<GrammarType>,
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
302
303
}

304
fn default_max_new_tokens() -> Option<u32> {
305
    Some(100)
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
306
307
308
309
}

fn default_parameters() -> GenerateParameters {
    GenerateParameters {
310
        best_of: None,
311
312
        temperature: None,
        repetition_penalty: None,
313
        frequency_penalty: None,
314
315
        top_k: None,
        top_p: None,
316
        typical_p: None,
317
        do_sample: true,
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
318
        max_new_tokens: default_max_new_tokens(),
319
        return_full_text: None,
320
        stop: Vec::new(),
321
        truncate: None,
322
        watermark: false,
OlivierDehaene's avatar
OlivierDehaene committed
323
        details: false,
324
        decoder_input_details: false,
325
        seed: None,
Nicolas Patry's avatar
Nicolas Patry committed
326
        top_n_tokens: None,
drbh's avatar
drbh committed
327
        grammar: None,
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
328
329
330
    }
}

331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
mod prompt_serde {
    use serde::{self, Deserialize, Deserializer};
    use serde_json::Value;

    pub fn deserialize<'de, D>(deserializer: D) -> Result<Vec<String>, D::Error>
    where
        D: Deserializer<'de>,
    {
        let value = Value::deserialize(deserializer)?;
        match value {
            Value::String(s) => Ok(vec![s]),
            Value::Array(arr) if arr.is_empty() => Err(serde::de::Error::custom(
                "Empty array detected. Do not use an empty array for the prompt.",
            )),
            Value::Array(arr) => arr
                .iter()
                .map(|v| match v {
                    Value::String(s) => Ok(s.to_owned()),
                    _ => Err(serde::de::Error::custom("Expected a string")),
                })
                .collect(),
            _ => Err(serde::de::Error::custom(
                "Expected a string or an array of strings",
            )),
        }
    }
}

359
360
361
362
363
364
365
366
367
#[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?")]
368
369
    #[serde(deserialize_with = "prompt_serde::deserialize")]
    pub prompt: Vec<String>,
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406

    /// 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>,
407
408
409
410
411

    /// Up to 4 sequences where the API will stop generating further tokens.
    #[serde(default)]
    #[schema(nullable = true, example = "null")]
    pub stop: Option<Vec<String>>,
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
}

#[derive(Clone, Deserialize, Serialize, ToSchema, Default)]
pub(crate) struct Completion {
    pub id: String,
    pub object: 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,
}

435
#[derive(Clone, Deserialize, Serialize, ToSchema)]
436
437
438
pub(crate) struct ChatCompletion {
    pub id: String,
    pub object: String,
439
    #[schema(example = "1706270835")]
440
    pub created: u64,
441
    #[schema(example = "mistralai/Mistral-7B-Instruct-v0.2")]
442
443
444
445
446
447
    pub model: String,
    pub system_fingerprint: String,
    pub choices: Vec<ChatCompletionComplete>,
    pub usage: Usage,
}

448
#[derive(Clone, Deserialize, Serialize, ToSchema)]
449
450
pub(crate) struct ChatCompletionComplete {
    pub index: u32,
Nicolas Patry's avatar
Nicolas Patry committed
451
    pub message: OutputMessage,
452
    pub logprobs: Option<ChatCompletionLogprobs>,
453
454
455
    pub finish_reason: String,
}

456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
#[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;
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498

        // 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
499
500
501
502
503
504
                        .into_iter()
                        .map(|t| ChatCompletionTopLogprob {
                            token: t.text,
                            logprob: t.logprob,
                        })
                        .collect(),
505
506
507
508
509
510
                    None => vec![], // Handle the case where there are no top tokens
                },
            })
            .collect();

        Self { content }
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
    }
}

#[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,
}

527
#[derive(Clone, Deserialize, Serialize, ToSchema, Default)]
528
529
530
531
532
533
534
535
536
537
pub(crate) struct Usage {
    pub prompt_tokens: u32,
    pub completion_tokens: u32,
    pub total_tokens: u32,
}

impl ChatCompletion {
    pub(crate) fn new(
        model: String,
        system_fingerprint: String,
drbh's avatar
drbh committed
538
        output: Option<String>,
539
540
541
        created: u64,
        details: Details,
        return_logprobs: bool,
542
        tool_calls: Option<Vec<ToolCall>>,
543
    ) -> Self {
Nicolas Patry's avatar
Nicolas Patry committed
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
        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(),
                })
            }
        };
568
569
570
571
572
573
574
575
        Self {
            id: String::new(),
            object: "text_completion".into(),
            created,
            model,
            system_fingerprint,
            choices: vec![ChatCompletionComplete {
                index: 0,
Nicolas Patry's avatar
Nicolas Patry committed
576
                message,
577
                logprobs: return_logprobs
578
                    .then(|| ChatCompletionLogprobs::from((details.tokens, details.top_tokens))),
579
580
581
582
583
584
585
586
587
588
                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,
            },
        }
    }
}
589
590
591
592
593
594
595
596
597
#[derive(Clone, Deserialize, Serialize, ToSchema)]
pub(crate) struct CompletionCompleteChunk {
    pub id: String,
    pub object: String,
    pub created: u64,
    pub choices: Vec<CompletionComplete>,
    pub model: String,
    pub system_fingerprint: String,
}
Nicolas Patry's avatar
Nicolas Patry committed
598

599
#[derive(Clone, Serialize, ToSchema)]
600
601
602
pub(crate) struct ChatCompletionChunk {
    pub id: String,
    pub object: String,
603
    #[schema(example = "1706270978")]
604
    pub created: u64,
605
    #[schema(example = "mistralai/Mistral-7B-Instruct-v0.2")]
606
607
608
609
610
    pub model: String,
    pub system_fingerprint: String,
    pub choices: Vec<ChatCompletionChoice>,
}

611
#[derive(Clone, Serialize, ToSchema)]
612
613
614
pub(crate) struct ChatCompletionChoice {
    pub index: u32,
    pub delta: ChatCompletionDelta,
615
    pub logprobs: Option<ChatCompletionLogprobs>,
616
617
618
    pub finish_reason: Option<String>,
}

Nicolas Patry's avatar
Nicolas Patry committed
619
620
621
622
623
624
625
#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
pub struct ToolCallDelta {
    #[schema(example = "assistant")]
    role: String,
    tool_calls: DeltaToolCall,
}

626
627
#[derive(Clone, Debug, Serialize, ToSchema)]
#[serde(untagged)]
Nicolas Patry's avatar
Nicolas Patry committed
628
629
630
enum ChatCompletionDelta {
    Chat(TextMessage),
    Tool(ToolCallDelta),
drbh's avatar
drbh committed
631
632
}

Nicolas Patry's avatar
Nicolas Patry committed
633
#[derive(Clone, Deserialize, Serialize, ToSchema, Debug, PartialEq)]
drbh's avatar
drbh committed
634
635
636
637
638
639
640
pub(crate) struct DeltaToolCall {
    pub index: u32,
    pub id: String,
    pub r#type: String,
    pub function: Function,
}

Nicolas Patry's avatar
Nicolas Patry committed
641
#[derive(Clone, Deserialize, Serialize, ToSchema, Debug, PartialEq)]
drbh's avatar
drbh committed
642
643
644
pub(crate) struct Function {
    pub name: Option<String>,
    pub arguments: String,
645
646
}

drbh's avatar
drbh committed
647
#[allow(clippy::too_many_arguments)]
648
649
650
651
impl ChatCompletionChunk {
    pub(crate) fn new(
        model: String,
        system_fingerprint: String,
drbh's avatar
drbh committed
652
653
        delta: Option<String>,
        tool_calls: Option<Vec<String>>,
654
        created: u64,
655
        logprobs: Option<ChatCompletionLogprobs>,
656
657
        finish_reason: Option<String>,
    ) -> Self {
658
        let delta = match (delta, tool_calls) {
Nicolas Patry's avatar
Nicolas Patry committed
659
660
661
662
663
664
665
            (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 {
666
667
668
669
670
671
672
                    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
673
674
675
676
677
678
                },
            }),
            (None, None) => ChatCompletionDelta::Chat(TextMessage {
                role: "assistant".to_string(),
                content: "".to_string(),
            }),
679
        };
680
681
682
683
684
685
686
        Self {
            id: String::new(),
            object: "text_completion".to_string(),
            created,
            model,
            system_fingerprint,
            choices: vec![ChatCompletionChoice {
687
                index: 0,
688
                delta,
689
690
691
692
693
694
695
696
697
                logprobs,
                finish_reason,
            }],
        }
    }
}

#[derive(Clone, Deserialize, ToSchema, Serialize)]
pub(crate) struct ChatRequest {
698
    #[schema(example = "mistralai/Mistral-7B-Instruct-v0.2")]
drbh's avatar
drbh committed
699
    /// [UNUSED] ID of the model to use. See the model endpoint compatibility table for details on which models work with the Chat API.
700
    pub model: String,
drbh's avatar
drbh committed
701

702
    /// A list of messages comprising the conversation so far.
drbh's avatar
drbh committed
703
    #[schema(example = "[{\"role\": \"user\", \"content\": \"What is Deep Learning?\"}]")]
704
705
706
707
708
    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)]
709
    #[schema(example = "1.0")]
710
711
712
713
714
715
716
717
718
719
720
721
722
723
    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)]
724
    #[schema(example = "false")]
725
726
727
728
729
    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)]
730
    #[schema(example = "5")]
731
732
733
734
    pub top_logprobs: Option<u32>,

    /// The maximum number of tokens that can be generated in the chat completion.
    #[serde(default)]
735
    #[schema(example = "32")]
736
737
738
739
740
741
    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)]
742
    #[schema(nullable = true, example = "2")]
743
744
745
746
747
    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)]
748
    #[schema(nullable = true, example = 0.1)]
749
750
    pub presence_penalty: Option<f32>,

751
752
753
754
755
    /// Up to 4 sequences where the API will stop generating further tokens.
    #[serde(default)]
    #[schema(nullable = true, example = "null")]
    pub stop: Option<Vec<String>>,

756
757
758
759
760
    #[serde(default = "bool::default")]
    pub stream: bool,

    #[schema(nullable = true, example = 42)]
    pub seed: Option<u64>,
761
762
763
764
765
766

    /// 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)]
767
    #[schema(nullable = true, example = 1.0)]
768
769
770
771
772
    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)]
773
    #[schema(nullable = true, example = 0.95)]
774
    pub top_p: Option<f32>,
drbh's avatar
drbh committed
775
776
777
778
779
780
781
782
783
784
785

    /// 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,
786
        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
787
788
789
790
791
792
793
794
    )]
    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")]
    #[serde(deserialize_with = "deserialize_tool_choice::deserialize")]
    pub tool_choice: Option<ToolType>,
drbh's avatar
drbh committed
795
796
797
798
799
800
801

    /// 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
802
803
804
805
}

fn default_tool_prompt() -> Option<String> {
    Some(
806
        "\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
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
    )
}
#[derive(Clone, Deserialize, ToSchema, Serialize)]
enum ToolType {
    FunctionName(String),
    OneOf,
}

/// Deserialize the tool choice from the JSON input or from the function name ("none" is allowed but mapped to None)
mod deserialize_tool_choice {
    use super::*;
    use serde::de;
    use serde::Deserializer;
    use serde_json::Value;

    pub fn deserialize<'de, D>(deserializer: D) -> Result<Option<ToolType>, D::Error>
    where
        D: Deserializer<'de>,
    {
        let value = Value::deserialize(deserializer)?;

        match value {
            Value::String(s) => match s.as_str() {
                "none" => Ok(None),
                "auto" => Ok(Some(ToolType::OneOf)),
                _ => Ok(Some(ToolType::FunctionName(s))),
            },
            Value::Object(map) => {
                if let Some(content) = map
                    .get("function")
                    .and_then(|v| v.get("name"))
                    .and_then(|v| v.as_str())
                {
                    Ok(Some(ToolType::FunctionName(content.to_string())))
                } else {
                    Err(de::Error::custom("function key not found in tool choice"))
                }
            }
            Value::Null => Ok(Some(ToolType::OneOf)),
            _ => Err(de::Error::custom("invalid token format")),
        }
    }
}

851
#[derive(Debug, Deserialize, Serialize, ToSchema, PartialEq)]
drbh's avatar
drbh committed
852
853
854
855
856
857
pub struct Tools {
    #[serde(flatten)]
    functions_map: FunctionsMap,
    properties: Properties,
}

858
#[derive(Debug, Serialize, Deserialize, PartialEq)]
drbh's avatar
drbh committed
859
860
861
862
863
struct FunctionsMap {
    #[serde(rename = "$functions")]
    functions: std::collections::HashMap<String, serde_json::Value>,
}

864
#[derive(Debug, Serialize, Deserialize, PartialEq)]
drbh's avatar
drbh committed
865
866
867
868
869
struct FunctionRef {
    #[serde(rename = "$ref")]
    ref_path: String,
}

870
#[derive(Debug, Serialize, Deserialize, PartialEq)]
drbh's avatar
drbh committed
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
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
886
#[derive(Clone, Debug, Deserialize, Serialize, ToSchema, Default, PartialEq)]
drbh's avatar
drbh committed
887
888
889
890
pub(crate) struct FunctionDefinition {
    #[serde(default)]
    pub description: Option<String>,
    pub name: String,
891
892
    #[serde(alias = "parameters")]
    pub arguments: serde_json::Value,
drbh's avatar
drbh committed
893
894
895
896
897
898
899
900
901
}

#[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,
902
903
}

904
#[derive(Clone, Serialize, Deserialize, Default)]
905
pub(crate) struct ChatTemplateInputs<'a> {
Nicolas Patry's avatar
Nicolas Patry committed
906
    messages: Vec<TextMessage>,
907
908
    bos_token: Option<&'a str>,
    eos_token: Option<&'a str>,
909
    add_generation_prompt: bool,
910
911
    tools: Option<&'a str>,
    tools_prompt: Option<&'a str>,
912
913
}

Nicolas Patry's avatar
Nicolas Patry committed
914
#[derive(Clone, Deserialize, Serialize, ToSchema, Default, Debug, PartialEq)]
drbh's avatar
drbh committed
915
pub(crate) struct ToolCall {
916
    pub id: String,
drbh's avatar
drbh committed
917
918
919
920
    pub r#type: String,
    pub function: FunctionDefinition,
}

Nicolas Patry's avatar
Nicolas Patry committed
921
922
923
#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
struct Url {
    url: String,
drbh's avatar
drbh committed
924
925
}

Nicolas Patry's avatar
Nicolas Patry committed
926
927
928
#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
struct ImageUrl {
    image_url: Url,
drbh's avatar
drbh committed
929
930
}

Nicolas Patry's avatar
Nicolas Patry committed
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
struct Text {
    text: String,
}

#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
#[serde(tag = "type")]
#[serde(rename_all = "snake_case")]
enum MessageChunk {
    Text(Text),
    ImageUrl(ImageUrl),
}

#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
pub struct Message {
    #[schema(example = "user")]
    role: String,
    #[schema(example = "My name is David and I")]
    #[serde(deserialize_with = "message_content_serde::deserialize")]
    content: Vec<MessageChunk>,
drbh's avatar
drbh committed
951
    #[serde(default, skip_serializing_if = "Option::is_none")]
Nicolas Patry's avatar
Nicolas Patry committed
952
953
    #[schema(example = "\"David\"")]
    name: Option<String>,
drbh's avatar
drbh committed
954
955
956
957
}

mod message_content_serde {
    use super::*;
Nicolas Patry's avatar
Nicolas Patry committed
958
    use serde::{Deserialize, Deserializer};
drbh's avatar
drbh committed
959

Nicolas Patry's avatar
Nicolas Patry committed
960
    pub fn deserialize<'de, D>(deserializer: D) -> Result<Vec<MessageChunk>, D::Error>
drbh's avatar
drbh committed
961
962
963
    where
        D: Deserializer<'de>,
    {
Nicolas Patry's avatar
Nicolas Patry committed
964
965
966
967
968
        #[derive(Deserialize)]
        #[serde(untagged)]
        enum Message {
            Text(String),
            Chunks(Vec<MessageChunk>),
drbh's avatar
drbh committed
969
        }
Nicolas Patry's avatar
Nicolas Patry committed
970
971
972
973
974
975
976
977
        let message: Message = Deserialize::deserialize(deserializer)?;
        let chunks = match message {
            Message::Text(text) => {
                vec![MessageChunk::Text(Text { text })]
            }
            Message::Chunks(s) => s,
        };
        Ok(chunks)
drbh's avatar
drbh committed
978
979
980
    }
}

Nicolas Patry's avatar
Nicolas Patry committed
981
982
#[derive(Clone, Deserialize, ToSchema, Serialize, Debug, PartialEq)]
pub struct TextMessage {
983
984
985
    #[schema(example = "user")]
    pub role: String,
    #[schema(example = "My name is David and I")]
Nicolas Patry's avatar
Nicolas Patry committed
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
    pub content: String,
}

impl From<Message> for TextMessage {
    fn from(value: Message) -> Self {
        TextMessage {
            role: value.role,
            content: value
                .content
                .into_iter()
                .map(|c| match c {
                    MessageChunk::Text(Text { text }) => text,
                    MessageChunk::ImageUrl(image) => {
                        let url = image.image_url.url;
                        format!("![]({url})")
                    }
                })
                .collect::<Vec<_>>()
                .join(""),
        }
    }
}

#[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),
1021
1022
}

1023
#[derive(Clone, Debug, Deserialize, ToSchema)]
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
1024
pub(crate) struct GenerateRequest {
1025
    #[schema(example = "My name is Olivier and I")]
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
1026
1027
1028
1029
1030
    pub inputs: String,
    #[serde(default = "default_parameters")]
    pub parameters: GenerateParameters,
}

1031
1032
1033
1034
1035
1036
1037
#[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
1038
    #[schema(default = "false")]
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
    pub stream: bool,
}

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

1051
1052
1053
1054
1055
1056
#[derive(Debug, Serialize, ToSchema)]
pub struct PrefillToken {
    #[schema(example = 0)]
    id: u32,
    #[schema(example = "test")]
    text: String,
1057
    #[schema(nullable = true, example = - 0.34)]
1058
1059
1060
    logprob: f32,
}

1061
#[derive(Debug, Serialize, ToSchema, Clone)]
1062
1063
1064
1065
1066
pub struct Token {
    #[schema(example = 0)]
    id: u32,
    #[schema(example = "test")]
    text: String,
1067
    #[schema(nullable = true, example = - 0.34)]
1068
    logprob: f32,
1069
1070
    #[schema(example = "false")]
    special: bool,
1071
1072
}

1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
#[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
1085
#[derive(Debug, Serialize, ToSchema)]
1086
#[serde(rename_all(serialize = "snake_case"))]
1087
#[schema(example = "Length")]
1088
1089
1090
1091
1092
1093
1094
1095
1096
pub(crate) enum FinishReason {
    #[schema(rename = "length")]
    Length,
    #[serde(rename = "eos_token")]
    #[schema(rename = "eos_token")]
    EndOfSequenceToken,
    #[schema(rename = "stop_sequence")]
    StopSequence,
}
1097

1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
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"),
        }
    }
}

1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
#[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
1120
1121
    #[serde(skip_serializing_if = "Vec::is_empty")]
    pub top_tokens: Vec<Vec<Token>>,
1122
1123
}

1124
#[derive(Serialize, ToSchema)]
OlivierDehaene's avatar
OlivierDehaene committed
1125
pub(crate) struct Details {
1126
1127
1128
    #[schema(example = "length")]
    pub finish_reason: FinishReason,
    #[schema(example = 1)]
OlivierDehaene's avatar
OlivierDehaene committed
1129
    pub generated_tokens: u32,
1130
    #[schema(nullable = true, example = 42)]
1131
    pub seed: Option<u64>,
1132
1133
    pub prefill: Vec<PrefillToken>,
    pub tokens: Vec<Token>,
1134
1135
    #[serde(skip_serializing_if = "Option::is_none")]
    pub best_of_sequences: Option<Vec<BestOfSequence>>,
Nicolas Patry's avatar
Nicolas Patry committed
1136
1137
    #[serde(skip_serializing_if = "Vec::is_empty")]
    pub top_tokens: Vec<Vec<Token>>,
OlivierDehaene's avatar
OlivierDehaene committed
1138
1139
}

1140
#[derive(Serialize, ToSchema)]
1141
pub(crate) struct GenerateResponse {
1142
    #[schema(example = "test")]
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
1143
    pub generated_text: String,
OlivierDehaene's avatar
OlivierDehaene committed
1144
1145
    #[serde(skip_serializing_if = "Option::is_none")]
    pub details: Option<Details>,
Olivier Dehaene's avatar
v0.1.0  
Olivier Dehaene committed
1146
}
1147

1148
1149
1150
1151
#[derive(Serialize, ToSchema)]
#[serde(transparent)]
pub(crate) struct TokenizeResponse(Vec<SimpleToken>);

1152
1153
1154
1155
1156
1157
#[derive(Serialize, ToSchema)]
pub(crate) struct StreamDetails {
    #[schema(example = "length")]
    pub finish_reason: FinishReason,
    #[schema(example = 1)]
    pub generated_tokens: u32,
1158
    #[schema(nullable = true, example = 42)]
1159
1160
1161
1162
    pub seed: Option<u64>,
}

#[derive(Serialize, ToSchema)]
1163
pub(crate) struct StreamResponse {
1164
    pub index: u32,
1165
    pub token: Token,
Nicolas Patry's avatar
Nicolas Patry committed
1166
1167
    #[serde(skip_serializing_if = "Vec::is_empty")]
    pub top_tokens: Vec<Token>,
1168
    #[schema(nullable = true, default = "null", example = "test")]
1169
    pub generated_text: Option<String>,
1170
1171
    #[schema(nullable = true, default = "null")]
    pub details: Option<StreamDetails>,
1172
1173
}

1174
#[derive(Serialize, ToSchema)]
1175
1176
pub(crate) struct ErrorResponse {
    pub error: String,
1177
    pub error_type: String,
1178
}
1179
1180

#[cfg(test)]
1181
mod tests {
1182
    use super::*;
Nicolas Patry's avatar
Nicolas Patry committed
1183
    use serde_json::json;
1184
1185
    use tokenizers::Tokenizer;

1186
    pub(crate) async fn get_tokenizer() -> Tokenizer {
1187
1188
1189
1190
        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()
1191
    }
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205

    #[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
1206
1207
1208
1209
        assert_eq!(
            config.chat_template,
            Some(ChatTemplateVersions::Single("test".to_string()))
        );
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
        assert_eq!(
            config.bos_token,
            Some("<|begin▁of▁sentence|>".to_string())
        );
        assert_eq!(config.eos_token, Some("<|end▁of▁sentence|>".to_string()));

        // 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
1241
1242
1243
1244
        assert_eq!(
            config.chat_template,
            Some(ChatTemplateVersions::Single("test".to_string()))
        );
1245
1246
1247
1248
1249
1250
        assert_eq!(
            config.bos_token,
            Some("<|begin▁of▁sentence|>".to_string())
        );
        assert_eq!(config.eos_token, Some("<|end▁of▁sentence|>".to_string()));
    }
Nicolas Patry's avatar
Nicolas Patry committed
1251
1252
1253

    #[test]
    fn test_chat_simple_string() {
Nicolas Patry's avatar
Nicolas Patry committed
1254
        let json = json!({
Nicolas Patry's avatar
Nicolas Patry committed
1255
            "model": "",
Nicolas Patry's avatar
Nicolas Patry committed
1256
1257
            "messages": [{
                "role": "user",
Nicolas Patry's avatar
Nicolas Patry committed
1258
                "content": "What is Deep Learning?"
Nicolas Patry's avatar
Nicolas Patry committed
1259
            }]
Nicolas Patry's avatar
Nicolas Patry committed
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
        });
        let request: ChatRequest = serde_json::from_str(json.to_string().as_str()).unwrap();

        assert_eq!(
            request.messages[0],
            Message {
                role: "user".to_string(),
                content: vec![MessageChunk::Text(Text {
                    text: "What is Deep Learning?".to_string()
                }),],
                name: None
            }
        );
    }

    #[test]
    fn test_chat_request() {
Nicolas Patry's avatar
Nicolas Patry committed
1277
        let json = json!({
Nicolas Patry's avatar
Nicolas Patry committed
1278
            "model": "",
Nicolas Patry's avatar
Nicolas Patry committed
1279
1280
            "messages": [{
                "role": "user",
Nicolas Patry's avatar
Nicolas Patry committed
1281
1282
                "content": [
                    {"type": "text", "text": "Whats in this image?"},
Nicolas Patry's avatar
Nicolas Patry committed
1283
                    {"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
1284
                ]
Nicolas Patry's avatar
Nicolas Patry committed
1285
            }]
Nicolas Patry's avatar
Nicolas Patry committed
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
        });
        let request: ChatRequest = serde_json::from_str(json.to_string().as_str()).unwrap();

        assert_eq!(
            request.messages[0],
            Message{
                role: "user".to_string(),
                content: vec![
                    MessageChunk::Text(Text { text: "Whats in this image?".to_string() }),
                    MessageChunk::ImageUrl(ImageUrl { image_url: Url { url: "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rabbit.png".to_string() } })
                ],
                name: None
            }
        );
    }
Nicolas Patry's avatar
Nicolas Patry committed
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346

    #[test]
    fn text_message_convert() {
        let message = Message{
                role: "user".to_string(),
                content: vec![
                    MessageChunk::Text(Text { text: "Whats in this image?".to_string() }),
                    MessageChunk::ImageUrl(ImageUrl { image_url: Url { url: "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rabbit.png".to_string() } })
                ],
                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"}}}]}"#
        );
    }
1347
}