Skip to content

generator

Encapsulate a model and an output type into a reusable object.

AsyncBlackBoxGenerator

Asynchronous generator for which we don't control constrained generation.

The output type provided is not compiled into a logits processor, but is instead directly passed on to the model.

Source code in outlines/generator.py 
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
class AsyncBlackBoxGenerator:
    """Asynchronous generator for which we don't control constrained
    generation.

    The output type provided is not compiled into a logits processor, but is
    instead directly passed on to the model.

    """
    output_type: Optional[Any]

    def __init__(self, model: AsyncBlackBoxModel, output_type: Optional[Any]):
        """
        Parameters
        ----------
        model
            An instance of an Outlines model.
        output_type
            The output type that will be used to constrain the generation.

        """
        self.model = model
        self.output_type = output_type

        if isinstance(self.output_type, FSM):
            raise NotImplementedError(
                "FSM generation is not supported for API-based models"
            )

    async def __call__(self, prompt: Any, **inference_kwargs) -> Any:
        """Generate a response from the model.

        Parameters
        ----------
        prompt
            The prompt to use to generate a response.
        **inference_kwargs
            Additional keyword arguments to pass to the model.

        Returns
        -------
        Any
            The response generated by the model.

        """
        return await self.model.generate(
            prompt, self.output_type, **inference_kwargs
        )

    async def batch(self, prompts: List[Any], **inference_kwargs) -> List[Any]:
        """Generate a batch of responses from the model.

        Parameters
        ----------
        prompts
            The list of prompts to use to generate a batch of responses.
        **inference_kwargs
            Additional keyword arguments to pass to the model.

        Returns
        -------
        List[Any]
            The list of responses generated by the model.

        """
        return await self.model.generate_batch(
            prompts, self.output_type, **inference_kwargs
        )

    async def stream(self, prompt: Any, **inference_kwargs) -> AsyncIterator[Any]:
        """Generate a stream of responses from the model.

        Parameters
        ----------
        prompt
            The prompt to use to generate a response.
        **inference_kwargs
            Additional keyword arguments to pass to the model.

        Returns
        -------
        Any
            The response generated by the model.

        """
        async for chunk in self.model.generate_stream(  # pragma: no cover
            prompt, self.output_type, **inference_kwargs
        ):
            yield chunk

__call__(prompt, **inference_kwargs) async

Generate a response from the model.

Parameters:

Name Type Description Default
prompt Any

The prompt to use to generate a response.

 required
**inference_kwargs

Additional keyword arguments to pass to the model.

 {}

Returns:

Type Description
Any

The response generated by the model.
Source code in outlines/generator.py 
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
async def __call__(self, prompt: Any, **inference_kwargs) -> Any:
    """Generate a response from the model.

    Parameters
    ----------
    prompt
        The prompt to use to generate a response.
    **inference_kwargs
        Additional keyword arguments to pass to the model.

    Returns
    -------
    Any
        The response generated by the model.

    """
    return await self.model.generate(
        prompt, self.output_type, **inference_kwargs
    )

__init__(model, output_type)

Parameters:

Name Type Description Default
model AsyncBlackBoxModel

An instance of an Outlines model.

 required
output_type Optional[Any]

The output type that will be used to constrain the generation.

 required
Source code in outlines/generator.py 
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
def __init__(self, model: AsyncBlackBoxModel, output_type: Optional[Any]):
    """
    Parameters
    ----------
    model
        An instance of an Outlines model.
    output_type
        The output type that will be used to constrain the generation.

    """
    self.model = model
    self.output_type = output_type

    if isinstance(self.output_type, FSM):
        raise NotImplementedError(
            "FSM generation is not supported for API-based models"
        )

batch(prompts, **inference_kwargs) async

Generate a batch of responses from the model.

Parameters:

Name Type Description Default
prompts List[Any]

The list of prompts to use to generate a batch of responses.

 required
**inference_kwargs

Additional keyword arguments to pass to the model.

 {}

Returns:

Type Description
List[Any]

The list of responses generated by the model.
Source code in outlines/generator.py 
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
async def batch(self, prompts: List[Any], **inference_kwargs) -> List[Any]:
    """Generate a batch of responses from the model.

    Parameters
    ----------
    prompts
        The list of prompts to use to generate a batch of responses.
    **inference_kwargs
        Additional keyword arguments to pass to the model.

    Returns
    -------
    List[Any]
        The list of responses generated by the model.

    """
    return await self.model.generate_batch(
        prompts, self.output_type, **inference_kwargs
    )

stream(prompt, **inference_kwargs) async

Generate a stream of responses from the model.

Parameters:

Name Type Description Default
prompt Any

The prompt to use to generate a response.

 required
**inference_kwargs

Additional keyword arguments to pass to the model.

 {}

Returns:

Type Description
Any

The response generated by the model.
Source code in outlines/generator.py 
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
async def stream(self, prompt: Any, **inference_kwargs) -> AsyncIterator[Any]:
    """Generate a stream of responses from the model.

    Parameters
    ----------
    prompt
        The prompt to use to generate a response.
    **inference_kwargs
        Additional keyword arguments to pass to the model.

    Returns
    -------
    Any
        The response generated by the model.

    """
    async for chunk in self.model.generate_stream(  # pragma: no cover
        prompt, self.output_type, **inference_kwargs
    ):
        yield chunk

BlackBoxGenerator

Synchronous generator for which we don't control constrained generation.

The output type provided is not compiled into a logits processor, but is instead directly passed on to the model.

Source code in outlines/generator.py 
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
class BlackBoxGenerator:
    """Synchronous generator for which we don't control constrained
    generation.

    The output type provided is not compiled into a logits processor, but is
    instead directly passed on to the model.

    """
    output_type: Optional[Any]

    def __init__(self, model: BlackBoxModel, output_type: Optional[Any]):
        """
        Parameters
        ----------
        model
            An instance of an Outlines model.
        output_type
            The output type that will be used to constrain the generation.

        """
        self.model = model
        self.output_type = output_type

        if isinstance(self.output_type, FSM):
            raise NotImplementedError(
                "FSM generation is not supported for API-based models"
            )

    def __call__(self, prompt: Any, **inference_kwargs) -> Any:
        """Generate a response from the model.

        Parameters
        ----------
        prompt
            The prompt to use to generate a response.
        **inference_kwargs
            Additional keyword arguments to pass to the model.

        Returns
        -------
        Any
            The response generated by the model.

        """
        return self.model.generate(
            prompt, self.output_type, **inference_kwargs
        )

    def batch(self, prompts: List[Any], **inference_kwargs) -> List[Any]:
        """Generate a batch of responses from the model.

        Parameters
        ----------
        prompts
            The list of prompts to use to generate a batch of responses.
        **inference_kwargs
            Additional keyword arguments to pass to the model.

        Returns
        -------
        List[Any]
            The list of responses generated by the model.

        """
        return self.model.generate_batch(
            prompts, self.output_type, **inference_kwargs
        )

    def stream(self, prompt: Any, **inference_kwargs) -> Iterator[Any]:
        """Generate a stream of responses from the model.

        Parameters
        ----------
        prompt
            The prompt to use to generate a response.
        **inference_kwargs
            Additional keyword arguments to pass to the model.

        Returns
        -------
        Any
            The response generated by the model.

        """
        return self.model.generate_stream(
            prompt, self.output_type, **inference_kwargs
        )

__call__(prompt, **inference_kwargs)

Generate a response from the model.

Parameters:

Name Type Description Default
prompt Any

The prompt to use to generate a response.

 required
**inference_kwargs

Additional keyword arguments to pass to the model.

 {}

Returns:

Type Description
Any

The response generated by the model.
Source code in outlines/generator.py 
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
def __call__(self, prompt: Any, **inference_kwargs) -> Any:
    """Generate a response from the model.

    Parameters
    ----------
    prompt
        The prompt to use to generate a response.
    **inference_kwargs
        Additional keyword arguments to pass to the model.

    Returns
    -------
    Any
        The response generated by the model.

    """
    return self.model.generate(
        prompt, self.output_type, **inference_kwargs
    )

__init__(model, output_type)

Parameters:

Name Type Description Default
model BlackBoxModel

An instance of an Outlines model.

 required
output_type Optional[Any]

The output type that will be used to constrain the generation.

 required
Source code in outlines/generator.py 
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
def __init__(self, model: BlackBoxModel, output_type: Optional[Any]):
    """
    Parameters
    ----------
    model
        An instance of an Outlines model.
    output_type
        The output type that will be used to constrain the generation.

    """
    self.model = model
    self.output_type = output_type

    if isinstance(self.output_type, FSM):
        raise NotImplementedError(
            "FSM generation is not supported for API-based models"
        )

batch(prompts, **inference_kwargs)

Generate a batch of responses from the model.

Parameters:

Name Type Description Default
prompts List[Any]

The list of prompts to use to generate a batch of responses.

 required
**inference_kwargs

Additional keyword arguments to pass to the model.

 {}

Returns:

Type Description
List[Any]

The list of responses generated by the model.
Source code in outlines/generator.py 
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
def batch(self, prompts: List[Any], **inference_kwargs) -> List[Any]:
    """Generate a batch of responses from the model.

    Parameters
    ----------
    prompts
        The list of prompts to use to generate a batch of responses.
    **inference_kwargs
        Additional keyword arguments to pass to the model.

    Returns
    -------
    List[Any]
        The list of responses generated by the model.

    """
    return self.model.generate_batch(
        prompts, self.output_type, **inference_kwargs
    )

stream(prompt, **inference_kwargs)

Generate a stream of responses from the model.

Parameters:

Name Type Description Default
prompt Any

The prompt to use to generate a response.

 required
**inference_kwargs

Additional keyword arguments to pass to the model.

 {}

Returns:

Type Description
Any

The response generated by the model.
Source code in outlines/generator.py 
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
def stream(self, prompt: Any, **inference_kwargs) -> Iterator[Any]:
    """Generate a stream of responses from the model.

    Parameters
    ----------
    prompt
        The prompt to use to generate a response.
    **inference_kwargs
        Additional keyword arguments to pass to the model.

    Returns
    -------
    Any
        The response generated by the model.

    """
    return self.model.generate_stream(
        prompt, self.output_type, **inference_kwargs
    )

SteerableGenerator

Represents a generator for which we control constrained generation.

The generator is responsible for building and storing the logits processor (which can be quite expensive to build), and then passing it to the model when the generator is called.

The argument defining constrained generation can be of 2 types associated to different methods to create an instance of the generator: - output_type (through __init__): an output type as defined in the outlines.types module - processor (through from_processor): an already built logits processor as defined in the outlines.processors module

The 2 parameters are mutually exclusive.

Source code in outlines/generator.py 
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
class SteerableGenerator:
    """Represents a generator for which we control constrained generation.

    The generator is responsible for building and storing the logits processor
    (which can be quite expensive to build), and then passing it to the model
    when the generator is called.

    The argument defining constrained generation can be of 2 types associated
    to different methods to create an instance of the generator:
    - `output_type` (through `__init__`): an output type as defined in the
      `outlines.types` module
    - `processor` (through `from_processor`): an already built logits processor
       as defined in the `outlines.processors` module

    The 2 parameters are mutually exclusive.

    """
    logits_processor: Optional[OutlinesLogitsProcessor]

    def __init__(self, model: SteerableModel, output_type: Optional[Any]):
        """
        Parameters
        ----------
        model
            An instance of an Outlines model.
        output_type
            The output type expressed as a Python type

        """
        self.model = model
        if output_type is None:
            self.logits_processor = None
        else:
            term = python_types_to_terms(output_type)
            if isinstance(term, CFG):
                cfg_string = term.definition
                self.logits_processor = CFGLogitsProcessor(
                    cfg_string,
                    self.model.tokenizer,
                    self.model.tensor_library_name,
                )
            elif isinstance(term, FSM):
                guide = RegexGuide.from_interegular_fsm(
                    term.fsm,
                    self.model.tokenizer,
                )
                self.logits_processor = GuideLogitsProcessor(
                    self.model.tokenizer, guide, self.model.tensor_library_name
                )
            else:
                regex_string = to_regex(term)
                self.logits_processor = RegexLogitsProcessor(
                    regex_string,
                    self.model.tokenizer,
                    self.model.tensor_library_name,
                )

    @classmethod
    def from_processor(
        cls, model: SteerableModel, processor: OutlinesLogitsProcessor
    ):
        """Create a generator from a logits processor.

        Parameters
        ----------
        model
            An instance of an Outlines model.
        processor
            An instance of an OutlinesLogitsProcessor.

        """
        if not isinstance(processor, OutlinesLogitsProcessor):
            raise TypeError(
                "The processor argument must be an instance of "
                "OutlinesLogitsProcessor"
            )
        instance = cls.__new__(cls)
        instance.model = model
        instance.logits_processor = processor

        return instance

    def __call__(self, prompt: Any, **inference_kwargs) -> Any:
        """Generate a response from the model.

        Parameters
        ----------
        prompt
            The prompt to use to generate a response.
        **inference_kwargs
            Additional keyword arguments to pass to the model.

        Returns
        -------
        Any
            The response generated by the model.

        """
        return self.model.generate(
            prompt, self.logits_processor, **inference_kwargs
        )

    def batch(self, prompts: List[Any], **inference_kwargs) -> List[Any]:
        """Generate a batch of responses from the model.

        Parameters
        ----------
        prompts
            The list of prompts to use to generate a batch of responses.
        **inference_kwargs
            Additional keyword arguments to pass to the model.

        Returns
        -------
        List[Any]
            The list of responses generated by the model.

        """
        return self.model.generate_batch(
            prompts, self.logits_processor, **inference_kwargs
        )

    def stream(self, prompt: Any, **inference_kwargs) -> Iterator[Any]:
        """Generate a stream of responses from the model.

        Parameters
        ----------
        prompt
            The prompt to use to generate a response.
        **inference_kwargs
            Additional keyword arguments to pass to the model.

        Returns
        -------
        Any
            The response generated by the model.

        """
        return self.model.generate_stream(
            prompt, self.logits_processor, **inference_kwargs
        )

__call__(prompt, **inference_kwargs)

Generate a response from the model.

Parameters:

Name Type Description Default
prompt Any

The prompt to use to generate a response.

 required
**inference_kwargs

Additional keyword arguments to pass to the model.

 {}

Returns:

Type Description
Any

The response generated by the model.
Source code in outlines/generator.py 
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
def __call__(self, prompt: Any, **inference_kwargs) -> Any:
    """Generate a response from the model.

    Parameters
    ----------
    prompt
        The prompt to use to generate a response.
    **inference_kwargs
        Additional keyword arguments to pass to the model.

    Returns
    -------
    Any
        The response generated by the model.

    """
    return self.model.generate(
        prompt, self.logits_processor, **inference_kwargs
    )

__init__(model, output_type)

Parameters:

Name Type Description Default
model SteerableModel

An instance of an Outlines model.

 required
output_type Optional[Any]

The output type expressed as a Python type

 required
Source code in outlines/generator.py 
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
def __init__(self, model: SteerableModel, output_type: Optional[Any]):
    """
    Parameters
    ----------
    model
        An instance of an Outlines model.
    output_type
        The output type expressed as a Python type

    """
    self.model = model
    if output_type is None:
        self.logits_processor = None
    else:
        term = python_types_to_terms(output_type)
        if isinstance(term, CFG):
            cfg_string = term.definition
            self.logits_processor = CFGLogitsProcessor(
                cfg_string,
                self.model.tokenizer,
                self.model.tensor_library_name,
            )
        elif isinstance(term, FSM):
            guide = RegexGuide.from_interegular_fsm(
                term.fsm,
                self.model.tokenizer,
            )
            self.logits_processor = GuideLogitsProcessor(
                self.model.tokenizer, guide, self.model.tensor_library_name
            )
        else:
            regex_string = to_regex(term)
            self.logits_processor = RegexLogitsProcessor(
                regex_string,
                self.model.tokenizer,
                self.model.tensor_library_name,
            )

batch(prompts, **inference_kwargs)

Generate a batch of responses from the model.

Parameters:

Name Type Description Default
prompts List[Any]

The list of prompts to use to generate a batch of responses.

 required
**inference_kwargs

Additional keyword arguments to pass to the model.

 {}

Returns:

Type Description
List[Any]

The list of responses generated by the model.
Source code in outlines/generator.py 
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
def batch(self, prompts: List[Any], **inference_kwargs) -> List[Any]:
    """Generate a batch of responses from the model.

    Parameters
    ----------
    prompts
        The list of prompts to use to generate a batch of responses.
    **inference_kwargs
        Additional keyword arguments to pass to the model.

    Returns
    -------
    List[Any]
        The list of responses generated by the model.

    """
    return self.model.generate_batch(
        prompts, self.logits_processor, **inference_kwargs
    )

from_processor(model, processor) classmethod

Create a generator from a logits processor.

Parameters:

Name Type Description Default
model SteerableModel

An instance of an Outlines model.

 required
processor OutlinesLogitsProcessor

An instance of an OutlinesLogitsProcessor.

 required
Source code in outlines/generator.py 
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
@classmethod
def from_processor(
    cls, model: SteerableModel, processor: OutlinesLogitsProcessor
):
    """Create a generator from a logits processor.

    Parameters
    ----------
    model
        An instance of an Outlines model.
    processor
        An instance of an OutlinesLogitsProcessor.

    """
    if not isinstance(processor, OutlinesLogitsProcessor):
        raise TypeError(
            "The processor argument must be an instance of "
            "OutlinesLogitsProcessor"
        )
    instance = cls.__new__(cls)
    instance.model = model
    instance.logits_processor = processor

    return instance

stream(prompt, **inference_kwargs)

Generate a stream of responses from the model.

Parameters:

Name Type Description Default
prompt Any

The prompt to use to generate a response.

 required
**inference_kwargs

Additional keyword arguments to pass to the model.

 {}

Returns:

Type Description
Any

The response generated by the model.
Source code in outlines/generator.py 
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
def stream(self, prompt: Any, **inference_kwargs) -> Iterator[Any]:
    """Generate a stream of responses from the model.

    Parameters
    ----------
    prompt
        The prompt to use to generate a response.
    **inference_kwargs
        Additional keyword arguments to pass to the model.

    Returns
    -------
    Any
        The response generated by the model.

    """
    return self.model.generate_stream(
        prompt, self.logits_processor, **inference_kwargs
    )

Generator(model, output_type=None, *, processor=None)

Create a generator for the given model and output parameters.

The 2 parameters output_type and processor are mutually exclusive. The parameters processor is only supported for SteerableModel instances (typically local models) and is intended to be only used by advanced users.

Parameters:

Name Type Description Default
model Union[Model, AsyncModel]

An instance of an Outlines model.

 required
output_type Optional[Any]

The output type expressed as a Python type or a type defined in the outlines.types.dsl module.

 None
processor Optional[OutlinesLogitsProcessor]

An instance of an OutlinesLogitsProcessor.

 None

Returns:

Type Description
Union[SteerableGenerator, BlackBoxGenerator, AsyncBlackBoxGenerator]

A generator instance.
Source code in outlines/generator.py 
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
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
def Generator(
    model: Union[Model, AsyncModel],
    output_type: Optional[Any] = None,
    *,
    processor: Optional[OutlinesLogitsProcessor] = None,
) -> Union[SteerableGenerator, BlackBoxGenerator, AsyncBlackBoxGenerator]:
    """Create a generator for the given model and output parameters.

    The 2 parameters output_type and processor are mutually exclusive. The
    parameters processor is only supported for SteerableModel instances
    (typically local models) and is intended to be only used by advanced users.

    Parameters
    ----------
    model
        An instance of an Outlines model.
    output_type
        The output type expressed as a Python type or a type defined in the
        outlines.types.dsl module.
    processor
        An instance of an OutlinesLogitsProcessor.

    Returns
    -------
    Union[SteerableGenerator, BlackBoxGenerator, AsyncBlackBoxGenerator]
        A generator instance.

    """
    provided_output_params = sum(
        param is not None
        for param in [output_type, processor]
    )
    if provided_output_params > 1:
        raise ValueError(
            "At most one of output_type or processor can be provided"
        )

    if isinstance(model, SteerableModel): # type: ignore
        if processor is not None:
            return SteerableGenerator.from_processor(model, processor) # type: ignore
        else:
            return SteerableGenerator(model, output_type) # type: ignore
    else:
        if processor is not None:
            raise NotImplementedError(
                "This model does not support logits processors"
            )
        if isinstance(model, AsyncBlackBoxModel): # type: ignore
            return AsyncBlackBoxGenerator(model, output_type) # type: ignore
        elif isinstance(model, BlackBoxModel): # type: ignore
            return BlackBoxGenerator(model, output_type) # type: ignore
        else:
            raise ValueError(
                "The model argument must be an instance of "
                "SteerableModel, BlackBoxModel or AsyncBlackBoxModel"
            )