Build Converter Strategy

Build config dataclasses and strategy helpers for modality converter construction.

KwargsBuildConfig(kwargs=dict()) dataclass

Config for building a converter using raw keyword arguments.

build(cls)

Build a converter using raw keyword arguments.

Parameters:

Name	Type	Description	Default
cls	type[BaseModalityConverter]	The converter class to build.	required

Returns:

Name	Type	Description
BaseModalityConverter	BaseModalityConverter	The built converter.

LMRPBuildConfig(lmrp_config, extra_kwargs=dict()) dataclass

Config for building a converter using an LMRP configuration.

build(cls)

Build a converter using an LMRP configuration.

Parameters:

Name	Type	Description	Default
cls	type[BaseModalityConverter]	The converter class to build.	required

Returns:

Name	Type	Description
BaseModalityConverter	BaseModalityConverter	The built converter.

PresetBuildConfig(preset, extra_kwargs=dict()) dataclass

Config for building a converter using a named preset.

build(cls)

Build a converter using a named preset.

Parameters:

Name	Type	Description	Default
cls	type[BaseModalityConverter]	The converter class to build.	required

Returns:

Name	Type	Description
BaseModalityConverter	BaseModalityConverter	The built converter.

Raises:

Type	Description
ValueError	If the class does not implement from_preset.

build_config_from_params(strategy, preset, kwargs)

Construct the typed build config for the given strategy.

Parameters:

Name	Type	Description	Default
strategy	ModalityConverterBuildStrategy	The build strategy.	required
preset	str | None	The preset name, required for PRESET strategy.	required
kwargs	dict[str, Any]	Remaining keyword arguments from the caller.	required

Returns:

Name	Type	Description
ConverterBuildConfig	ConverterBuildConfig	Config instance carrying the build logic for the strategy.

Raises:

Type	Description
ValueError	If the strategy is unsupported or required params are missing.

determine_build_strategy(preset, kwargs)

Determine the build strategy from the provided configuration.

Priority by order: 1. LMRP config -> LMRP strategy 2. Preset -> PRESET strategy 3. Fallback -> KWARGS strategy

Parameters:

Name	Type	Description	Default
preset	str | None	The preset to use.	required
kwargs	dict[str, Any]	Additional keyword arguments.	required

Returns:

Name	Type	Description
ModalityConverterBuildStrategy	ModalityConverterBuildStrategy	The determined build strategy.

determine_transcriber_approach(model_id)

Determine the modality converter approach based on the model ID's provider and name.

This helper is used to automatically select the most appropriate transcription approach (e.g., Gemini, Google Cloud, Whisper, Prosa) when building an audio-to-text converter.

Parameters:

Name	Type	Description	Default
model_id	ModelId	The model ID to inspect, containing provider and name information.	required

Returns:

Name	Type	Description
ModalityConverterApproach	'ModalityConverterApproach'	The determined approach for transcription.

Raises:

Type	Description
ValueError	If the modality converter approach cannot be determined for the given model ID.

Example

Determining approach from model ID

from gllm_inference.schema.model_id import ModelId

# Gemini model -> LM_BASED approach
model_id = ModelId.from_string("google/gemini-1.5-pro")
approach = determine_transcriber_approach(model_id)
print(approach)  # ModalityConverterApproach.GEMINI

# Prosa model -> ASR approach
model_id = ModelId.from_string("prosa/v2")
approach = determine_transcriber_approach(model_id)
print(approach)  # ModalityConverterApproach.PROSA

determine_transcript_approach(model_id)

Determine the canonical transcription approach based on model metadata.

Parameters:

Name	Type	Description	Default
model_id	ModelId	The model ID to inspect.	required

Returns:

Name	Type	Description
ModalityConverterApproach	'ModalityConverterApproach'	Canonical approach type (lm_based, asr, or transcript_fetch).

Raises:

Type	Description
ValueError	If the approach cannot be determined for the given model ID.

determine_transcript_provider(model_id)

Determine the transcript provider implied by a model ID.

Parameters:

Name	Type	Description	Default
model_id	ModelId	The model ID to inspect.	required

Returns:

Type	Description
str | None	str | None: Provider name when inferable (e.g. whisper, gemini), otherwise None.