ternary_quant/__init__.py

"""Post-training ternary quantization for HuggingFace LLMs."""

from ternary_quant.quantizer import TernaryQuantizer
from ternary_quant.pipeline import quantize_model
from ternary_quant.inference import (
    TernaryLinear,
    clear_quantized_runtime_cache,
    load_ternary_model,
    prepare_quantized_runtime,
)
from ternary_quant.generative_adapters import (
    BroadQuantizationConfig,
    build_calibration_batches,
    generative_model_info_to_dict,
    inspect_generative_model,
    load_generative_model,
    quantize_components_inplace,
)
from ternary_quant.mac_kernels import metal_ternary_kernels_available
from ternary_quant.quantizer_small import (
    GroupwiseAsymmetricTernaryQuantizer,
    build_role_aware_plan,
    quantize_small_model_inplace,
)
from ternary_quant.ptq_families import (
    ProgressiveTritPlaneQuantizer,
    TritPlaneParameter,
    build_family_config,
    get_default_family_candidates,
    quantize_family_inplace,
    summarize_family_quantization,
)

__version__ = "0.1.5"
__all__ = [
    "TernaryQuantizer",
    "quantize_model",
    "GroupwiseAsymmetricTernaryQuantizer",
    "ProgressiveTritPlaneQuantizer",
    "TritPlaneParameter",
    "build_role_aware_plan",
    "quantize_small_model_inplace",
    "build_family_config",
    "get_default_family_candidates",
    "quantize_family_inplace",
    "summarize_family_quantization",
    "BroadQuantizationConfig",
    "inspect_generative_model",
    "load_generative_model",
    "build_calibration_batches",
    "quantize_components_inplace",
    "generative_model_info_to_dict",
    "TernaryLinear",
    "load_ternary_model",
    "prepare_quantized_runtime",
    "clear_quantized_runtime_cache",
    "metal_ternary_kernels_available",
]