File size: 6,002 Bytes
3193ef1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
from typing import Optional, Union

from google.cloud import aiplatform
from google.cloud.aiplatform import initializer as aiplatform_initializer
from vertexai.language_models import _language_models
from vertexai.language_models import _language_models as tuning


_DISTILLATION_PIPELINE_URI = (
    "https://us-kfp.pkg.dev/ml-pipeline/distillation/distillation/v1.0.0"
)


class DistillationMixin:
    def distill_from(
        self,
        *,
        dataset: str,
        teacher_model: Union[str, _language_models._TextGenerationModel],
        train_steps: Optional[int] = None,
        learning_rate_multiplier: Optional[float] = None,
        evaluation_spec: Optional[tuning.TuningEvaluationSpec] = None,
        accelerator_type: Optional[tuning._ACCELERATOR_TYPE_TYPE] = None,
        model_display_name: Optional[str] = None,
        max_context_length: Optional[str] = None,
    ):
        """Tunes a smaller model with help from another bigger model.

        Args:
            dataset: A URI pointing to data in JSON lines format.
            teacher_model: The teacher model to use for distillation.
            train_steps: Number of training batches to use (batch size is 8 samples).
            learning_rate_multiplier: Learning rate multiplier to use in tuning.
            evaluation_spec: Specification for the model evaluation during tuning.
            accelerator_type: Type of accelerator to use. Can be "TPU" or "GPU".
            model_display_name: Custom display name for the tuned model.
            max_context_length: The max context length used for tuning.
                Can be either '8k' or '32k'

        Returns:
            A tuning job for distillation.

        Raises:
            RuntimeError: If the model does not support distillation.
        """
        if "/models/" not in self._endpoint_name:
            raise RuntimeError(
                f"Model does not support distillation: {self._endpoint_name}"
            )
        student_short_model_id = self._endpoint_name.split("/")[-1]

        if isinstance(teacher_model, str):
            teacher_short_model_id = teacher_model
        elif isinstance(teacher_model, _language_models._LanguageModel):
            if "/models/" not in teacher_model._endpoint_name:
                raise RuntimeError(
                    f"Teacher model does not support distillation: {teacher_model._endpoint_name}"
                )
            teacher_short_model_id = teacher_model._endpoint_name.split("/")[-1]
        else:
            raise RuntimeError(f"Unsupported teacher model type: {teacher_model}")

        pipeline_job = submit_distillation_pipeline_job(
            teacher_model=teacher_short_model_id,
            student_model=student_short_model_id,
            dataset=dataset,
            train_steps=train_steps,
            learning_rate_multiplier=learning_rate_multiplier,
            evaluation_spec=evaluation_spec,
            accelerator_type=accelerator_type,
            model_display_name=model_display_name,
            max_context_length=max_context_length,
        )
        tuning_job = tuning._LanguageModelTuningJob(
            base_model=self,
            job=pipeline_job,
        )
        return tuning_job


def submit_distillation_pipeline_job(
    *,
    teacher_model: str,
    student_model: str,
    dataset: str,
    train_steps: Optional[int] = None,
    learning_rate_multiplier: Optional[float] = None,
    evaluation_spec: Optional[tuning.TuningEvaluationSpec] = None,
    accelerator_type: Optional[tuning._ACCELERATOR_TYPE_TYPE] = None,
    model_display_name: Optional[str] = None,
    max_context_length: Optional[str] = None,
):
    teacher_short_model_id = teacher_model.split("/")[-1]
    student_short_model_id = student_model.split("/")[-1]
    pipeline_arguments = {
        "teacher_model_reference": teacher_model,
        "student_model_reference": student_model,
        "dataset_uri": dataset,
        "project": aiplatform_initializer.global_config.project,
        "location": aiplatform_initializer.global_config.location,
    }
    if train_steps is not None:
        pipeline_arguments["train_steps"] = train_steps
    if learning_rate_multiplier is not None:
        pipeline_arguments["learning_rate_multiplier"] = learning_rate_multiplier
    if evaluation_spec is not None:
        pipeline_arguments["evaluation_data_uri"] = evaluation_spec.evaluation_data
        pipeline_arguments["evaluation_interval"] = evaluation_spec.evaluation_interval
        pipeline_arguments[
            "enable_early_stopping"
        ] = evaluation_spec.enable_early_stopping
        pipeline_arguments[
            "enable_checkpoint_selection"
        ] = evaluation_spec.enable_checkpoint_selection
        pipeline_arguments["tensorboard_resource_id"] = evaluation_spec.tensorboard
        # pipeline_parameter_values["evaluation_output_root_dir"] = ...
    if accelerator_type is not None:
        pipeline_arguments["accelerator_type"] = accelerator_type
    if aiplatform_initializer.global_config.encryption_spec_key_name is not None:
        pipeline_arguments[
            "encryption_spec_key_name"
        ] = aiplatform_initializer.global_config.encryption_spec_key_name
    if max_context_length is not None:
        pipeline_arguments["max_context_length"] = max_context_length
    if model_display_name is None:
        model_display_name = (
            f"{student_short_model_id} distilled from {teacher_short_model_id}"
        )
    pipeline_arguments["model_display_name"] = model_display_name
    # # Not exposing these parameters:
    # temperature: Optional[float] = None,
    # tpu_training_skip_cmek: Optional[bool] = None,
    # api_endpoint: Optional[str] = None,
    # version: Optional[str] = None,
    pipeline_job = aiplatform.PipelineJob(
        template_path=_DISTILLATION_PIPELINE_URI,
        display_name=None,
        parameter_values=pipeline_arguments,
    )
    pipeline_job.submit()
    return pipeline_job