Seed Datasets

1
import data_designer.config as dd
2
from data_designer.interface import DataDesigner
3
4
# Define your model configuration
5
model_configs = [
6
    dd.ModelConfig(
7
        alias="my-model",
8
        model="nvidia/nemotron-3-nano-30b-a3b",
9
        provider="nvidia",
10
    )
11
]
12
13
config_builder = dd.DataDesignerConfigBuilder(model_configs=model_configs)
14
15
# 1. Attach a seed dataset
16
seed_source = dd.LocalFileSeedSource(path="products.csv")
17
config_builder.with_seed_dataset(seed_source)
18
19
# 2. Reference seed columns in your prompts
20
config_builder.add_column(
21
    dd.LLMTextColumnConfig(
22
        name="review",
23
        model_alias="my-model",
24
        prompt="""\
25
Write a customer review for {{ product_name }}.
26
Category: {{ category }}
27
Price: ${{ price }}
28
""",
29
    )
30
)

1
# Single file
2
seed_source = dd.LocalFileSeedSource(path="data/products.csv")
3
4
# Parquet files with wildcard
5
seed_source = dd.LocalFileSeedSource(path="data/products/*.parquet")

1
seed_source = dd.HuggingFaceSeedSource(
2
    path="datasets/gretelai/symptom_to_diagnosis/data/train.parquet",
3
    token="hf_...",  # Optional, for private datasets
4
)

1
import pandas as pd
2
3
df = pd.read_csv("raw_data.csv")
4
df = df[df["quality_score"] > 0.8]  # Filter to high-quality rows
5
6
seed_source = dd.DataFrameSeedSource(df=df)

1
seed_source = dd.DirectorySeedSource(
2
    path="docs/",
3
    file_pattern="*.md",
4
    recursive=True,
5
)
6
7
config_builder.with_seed_dataset(seed_source)
8
config_builder.add_column(
9
    dd.ExpressionColumnConfig(
10
        name="doc_label",
11
        expr="{{ source_kind }}::{{ relative_path }}",
12
    )
13
)

1
seed_source = dd.FileContentsSeedSource(
2
    path="docs/",
3
    file_pattern="*.md",
4
    encoding="utf-8",
5
)
6
7
config_builder.with_seed_dataset(seed_source)
8
config_builder.add_column(
9
    dd.LLMTextColumnConfig(
10
        name="summary",
11
        model_alias="my-model",
12
        prompt="""\
13
Summarize the following document.
14
15
File: {{ file_name }}
16
Path: {{ relative_path }}
17
18
{{ content }}
19
""",
20
    )
21
)

1
seed_source = dd.AgentRolloutSeedSource(
2
    format=dd.AgentRolloutFormat.CLAUDE_CODE,
3
)
4
5
config_builder.with_seed_dataset(seed_source)

1
config_builder.with_seed_dataset(
2
    seed_source,
3
    sampling_strategy=dd.SamplingStrategy.ORDERED,
4
)

1
config_builder.with_seed_dataset(
2
    seed_source,
3
    sampling_strategy=dd.SamplingStrategy.SHUFFLE,
4
)

1
# Use only rows 100-199 (100 rows total)
2
config_builder.with_seed_dataset(
3
    seed_source,
4
    selection_strategy=dd.IndexRange(start=100, end=199),
5
)

1
# Split into 5 partitions, use the 3rd one (index=2, zero-based)
2
config_builder.with_seed_dataset(
3
    seed_source,
4
    selection_strategy=dd.PartitionBlock(index=2, num_partitions=5),
5
)

1
config_builder.with_seed_dataset(
2
    seed_source,
3
    sampling_strategy=dd.SamplingStrategy.SHUFFLE,
4
    selection_strategy=dd.PartitionBlock(index=0, num_partitions=10),
5
)

1
import data_designer.config as dd
2
from data_designer.interface import DataDesigner
3
4
data_designer = DataDesigner()
5
6
model_configs = [
7
    dd.ModelConfig(
8
        alias="medical-notes",
9
        model="nvidia/nemotron-3-nano-30b-a3b",
10
        provider="nvidia",
11
    )
12
]
13
14
config_builder = dd.DataDesignerConfigBuilder(model_configs=model_configs)
15
16
# Attach seed dataset (has 'diagnosis' and 'symptoms' columns)
17
seed_source = dd.LocalFileSeedSource(path="symptom_to_diagnosis.csv")
18
config_builder.with_seed_dataset(seed_source)
19
20
# Generate patient info
21
config_builder.add_column(
22
    dd.SamplerColumnConfig(
23
        name="patient",
24
        sampler_type=dd.SamplerType.PERSON_FROM_FAKER,
25
        params=dd.PersonFromFakerSamplerParams(),
26
    )
27
)
28
29
config_builder.add_column(
30
    dd.ExpressionColumnConfig(
31
        name="patient_name",
32
        expr="{{ patient.first_name }} {{ patient.last_name }}",
33
    )
34
)
35
36
# Generate notes grounded in seed data
37
config_builder.add_column(
38
    dd.LLMTextColumnConfig(
39
        name="physician_notes",
40
        model_alias="medical-notes",
41
        prompt="""\
42
You are a physician writing notes after a patient visit.
43
44
Patient: {{ patient_name }}
45
Diagnosis: {{ diagnosis }}
46
Reported Symptoms: {{ symptoms }}
47
48
Write detailed clinical notes for this visit.
49
""",
50
    )
51
)
52
53
# Preview
54
preview = data_designer.preview(config_builder, num_records=5)
55
preview.display_sample_record()