Source code for oumi.core.configs.analyze_config
# Copyright 2025 - Oumi
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from dataclasses import dataclass, field
from enum import Enum
from typing import Any, Optional
from omegaconf import MISSING
from oumi.core.configs.base_config import BaseConfig
from oumi.core.configs.params.base_params import BaseParams
[docs]
class DatasetSource(Enum):
"""Source of the dataset for analysis."""
CONFIG = "config"
"""Load dataset from config parameters (dataset_name, dataset_path, etc.)"""
DIRECT = "direct"
"""Pass dataset directly to DatasetAnalyzer.__init__()"""
[docs]
@dataclass
class SampleAnalyzerParams(BaseParams):
"""Params for a single sample analyzer plugin."""
id: str = MISSING
"""Unique identifier for the analyzer."""
params: dict[str, Any] = field(default_factory=dict)
"""Analyzer-specific parameters passed to the analyzer constructor."""
[docs]
@dataclass
class AnalyzeConfig(BaseConfig):
"""Configuration for dataset analysis and aggregation."""
# Required field - must come first
dataset_source: DatasetSource = MISSING
"""Source of the dataset for analysis. Use CONFIG to load from config parameters
or DIRECT to pass dataset directly to DatasetAnalyzer.__init__().
This field is required and must be explicitly set.
"""
# Simple fields for common use cases
dataset_name: Optional[str] = None
"""Dataset name."""
dataset_path: Optional[str] = None
"""Path to a custom dataset file (JSON or JSONL format).
If provided, this takes precedence over dataset_name for loading custom datasets.
"""
dataset_format: Optional[str] = None
"""Format of the custom dataset. Either 'oumi' (conversation format) or 'alpaca'.
Only used when dataset_path is provided.
"""
split: str = "train"
"""The split of the dataset to load.
This is typically one of "train", "test", or "validation". Defaults to "train".
"""
subset: Optional[str] = None
"""The subset of the dataset to load. If None, uses the base dataset."""
sample_count: Optional[int] = None
"""The number of examples to sample from the dataset.
If None, uses the full dataset. If specified, must be non-negative.
"""
output_path: str = "."
"""Directory path where output files will be saved.
Defaults to current directory ('.').
"""
analyzers: list[SampleAnalyzerParams] = field(default_factory=list)
"""List of analyzer configurations (plugin-style)."""
tokenizer_config: Optional[dict[str, Any]] = None
"""Tokenizer configuration for building a tokenizer.
If None, no tokenizer will be used.
Expected format:
{
"model_name": "gpt2", # Required: model name for tokenizer
"tokenizer_kwargs": {}, # Optional: additional tokenizer parameters
"trust_remote_code": False # Optional: whether to trust remote code
}
"""
# Add processor parameters for vision-language datasets
processor_name: Optional[str] = None
"""Processor name for vision-language datasets."""
processor_kwargs: dict[str, Any] = field(default_factory=dict)
"""Processor-specific parameters."""
trust_remote_code: bool = False
"""Whether to trust remote code for processor loading."""
is_multimodal: Optional[bool] = None
"""If True, treat the dataset as multimodal (vision-language) when using a
custom dataset_path. If False, treat as text-only.
"""
[docs]
def __post_init__(self):
"""Validates the configuration parameters."""
if self.dataset_source == DatasetSource.CONFIG:
# Only require dataset info when loading from config
if not self.dataset_name and not self.dataset_path:
raise ValueError(
"Either 'dataset_name' or 'dataset_path' must be provided when "
"dataset_source=DatasetSource.CONFIG"
)
else:
# When using direct dataset, dataset_name is optional but recommended
if not self.dataset_name:
self.dataset_name = "Custom Dataset"
# Validate dataset_format requirements
if self.dataset_path is not None:
if self.dataset_format is None:
raise ValueError(
"'dataset_format' must be specified when using 'dataset_path'. "
"Use 'oumi' for conversation format or 'alpaca' for instruction "
"format."
)
elif self.dataset_format not in ["oumi", "alpaca"]:
raise ValueError("'dataset_format' must be either 'oumi' or 'alpaca'")
# Require explicit is_multimodal setting for custom datasets
if self.is_multimodal is None:
raise ValueError(
"'is_multimodal' must be specified when using 'dataset_path'. "
"Set to 'True' for vision-language datasets or 'False' for "
"text-only datasets."
)
# Additional validation for multimodal
if self.is_multimodal is True:
# Currently VLJsonlinesDataset expects oumi conversation format
if self.dataset_format != "oumi":
raise ValueError(
"Multimodal datasets require dataset_format='oumi'"
)
if not self.processor_name:
raise ValueError(
"'processor_name' must be specified when 'is_multimodal' "
"is True"
)
# Validate sample_count
if self.sample_count is not None and self.sample_count <= 0:
raise ValueError("`sample_count` must be greater than 0.")
# Validate analyzer configurations
analyzer_ids = set()
for analyzer in self.analyzers:
# Validate analyzer ID
if not analyzer.id:
raise ValueError("Analyzer 'id' must be provided")
if analyzer.id in analyzer_ids:
raise ValueError(f"Duplicate analyzer ID found: '{analyzer.id}'")
analyzer_ids.add(analyzer.id)
