File size: 6,456 Bytes
8773ff3
4b83e74
8773ff3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4b83e74
 
 
8773ff3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4b83e74
 
 
8773ff3
 
 
 
 
4b83e74
8773ff3
 
 
 
 
 
 
 
4b83e74
 
 
 
 
 
 
 
 
 
 
8773ff3
 
 
 
 
4b83e74
 
 
8773ff3
 
 
 
 
 
4b83e74
8773ff3
 
4b83e74
 
8773ff3
 
 
 
4b83e74
 
 
 
8773ff3
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
import subprocess
import sys
import time
from typing import List

from distilabel.steps.generators.data import LoadDataFromDicts
from distilabel.steps.expand import ExpandColumns
from distilabel.steps.keep import KeepColumns
from distilabel.steps.tasks.self_instruct import SelfInstruct
from distilabel.steps.tasks.evol_instruct.base import EvolInstruct
from distilabel.llms.huggingface import InferenceEndpointsLLM
from distilabel.pipeline import Pipeline
from distilabel.steps import TextGenerationToArgilla
from dotenv import load_dotenv

from domain import (
    DomainExpert,
    CleanNumberedList,
    create_topics,
    create_examples_template,
    APPLICATION_DESCRIPTION,
)

load_dotenv()


def define_pipeline(
    argilla_api_key: str,
    argilla_api_url: str,
    argilla_dataset_name: str,
    topics: List[str],
    perspectives: List[str],
    domain_expert_prompt: str,
    examples: List[dict],
    hub_token: str,
    endpoint_base_url: str,
):
    """Define the pipeline for the specific domain."""

    terms = create_topics(topics, perspectives)
    template = create_examples_template(examples)
    with Pipeline("farming") as pipeline:
        load_data = LoadDataFromDicts(
            name="load_data",
            data=[{"input": term} for term in terms],
            batch_size=64,
        )
        llm = InferenceEndpointsLLM(
            base_url=endpoint_base_url,
            api_key=hub_token,
        )
        self_instruct = SelfInstruct(
            name="self-instruct",
            application_description=APPLICATION_DESCRIPTION,
            num_instructions=5,
            input_batch_size=8,
            llm=llm,
        )

        evol_instruction_complexity = EvolInstruct(
            name="evol_instruction_complexity",
            llm=llm,
            num_evolutions=2,
            store_evolutions=True,
            input_batch_size=8,
            include_original_instruction=True,
            input_mappings={"instruction": "question"},
        )

        expand_instructions = ExpandColumns(
            name="expand_columns", columns={"instructions": "question"}
        )
        cleaner = CleanNumberedList(name="clean_numbered_list")
        expand_evolutions = ExpandColumns(
            name="expand_columns_evolved",
            columns={"evolved_instructions": "evolved_questions"},
        )

        domain_expert = DomainExpert(
            name="domain_expert",
            llm=llm,
            input_batch_size=8,
            input_mappings={"instruction": "evolved_questions"},
            output_mappings={"generation": "domain_expert_answer"},
        )

        domain_expert._system_prompt = domain_expert_prompt
        domain_expert._template = template

        keep_columns = KeepColumns(
            name="keep_columns",
            columns=["model_name", "evolved_questions", "domain_expert_answer"],
        )

        to_argilla = TextGenerationToArgilla(
            name="text_generation_to_argilla",
            dataset_name=argilla_dataset_name,
            dataset_workspace="admin",
            api_url=argilla_api_url,
            api_key=argilla_api_key,
            input_mappings={
                "instruction": "evolved_questions",
                "generation": "domain_expert_answer",
            },
        )

        load_data.connect(self_instruct)
        self_instruct.connect(expand_instructions)
        expand_instructions.connect(cleaner)
        cleaner.connect(evol_instruction_complexity)
        evol_instruction_complexity.connect(expand_evolutions)
        expand_evolutions.connect(domain_expert)
        domain_expert.connect(keep_columns)
        keep_columns.connect(to_argilla)
    return pipeline


def serialize_pipeline(
    argilla_api_key: str,
    argilla_api_url: str,
    argilla_dataset_name: str,
    topics: List[str],
    perspectives: List[str],
    domain_expert_prompt: str,
    hub_token: str,
    endpoint_base_url: str,
    pipeline_config_path: str = "pipeline.yaml",
    examples: List[dict] = [],
):
    """Serialize the pipeline to a yaml file."""
    pipeline = define_pipeline(
        argilla_api_key=argilla_api_key,
        argilla_api_url=argilla_api_url,
        argilla_dataset_name=argilla_dataset_name,
        topics=topics,
        perspectives=perspectives,
        domain_expert_prompt=domain_expert_prompt,
        hub_token=hub_token,
        endpoint_base_url=endpoint_base_url,
        examples=examples,
    )
    pipeline.save(path=pipeline_config_path, overwrite=True, format="yaml")


def create_pipelines_run_command(
    hub_token: str,
    argilla_api_key: str,
    argilla_api_url: str,
    pipeline_config_path: str = "pipeline.yaml",
    argilla_dataset_name: str = "domain_specific_datasets",
):
    """Create the command to run the pipeline."""
    command_to_run = [
        sys.executable,
        "-m",
        "distilabel",
        "pipeline",
        "run",
        "--config",
        pipeline_config_path,
        "--param",
        f"text_generation_to_argilla.dataset_name={argilla_dataset_name}",
        "--param",
        f"text_generation_to_argilla.api_key={argilla_api_key}",
        "--param",
        f"text_generation_to_argilla.api_url={argilla_api_url}",
        "--param",
        f"self-instruct.llm.api_key={hub_token}",
        "--param",
        f"evol_instruction_complexity.llm.api_key={hub_token}",
        "--param",
        f"domain_expert.llm.api_key={hub_token}",
        "--ignore-cache",
    ]
    return command_to_run


def run_pipeline(
    hub_token: str,
    argilla_api_key: str,
    argilla_api_url: str,
    pipeline_config_path: str = "pipeline.yaml",
    argilla_dataset_name: str = "domain_specific_datasets",
):
    """Run the pipeline and yield the output as a generator of logs."""

    command_to_run = create_pipelines_run_command(
        hub_token=hub_token,
        pipeline_config_path=pipeline_config_path,
        argilla_dataset_name=argilla_dataset_name,
        argilla_api_key=argilla_api_key,
        argilla_api_url=argilla_api_url,
    )

    # Run the script file
    process = subprocess.Popen(
        args=command_to_run,
        stdout=subprocess.PIPE,
        stderr=subprocess.PIPE,
        env={"HF_TOKEN": hub_token},
    )

    while process.stdout and process.stdout.readable():
        time.sleep(0.2)
        line = process.stdout.readline()
        if not line:
            break
        yield line.decode("utf-8")