"""llama.cpp VRAM estimation helpers.

Pure calculation logic, importable without Gradio so it can be unit-tested.
"""

from .quant import QUANT_BPW, weight_bytes, quant_from_filename
from .kv import kv_cache_bytes, cache_dtype_bytes, compute_scratch_bytes
from .yarn import yarn_effective_context, yarn_warnings
from .gpu import GpuBudget, gpu_split, fit_gpus, GpuSplitResult
from .gguf import (
    GGUFMetadata,
    parse_header_bytes,
    parse_header_with_tensors,
    parse_local_file,
    parse_hf_range,
    metadata_to_arch,
)
from .report import (
    ModelArch,
    Inputs,
    Breakdown,
    estimate,
    command_preview,
    format_bytes,
)

__all__ = [
    "QUANT_BPW",
    "weight_bytes",
    "quant_from_filename",
    "kv_cache_bytes",
    "cache_dtype_bytes",
    "compute_scratch_bytes",
    "yarn_effective_context",
    "yarn_warnings",
    "GpuBudget",
    "GpuSplitResult",
    "gpu_split",
    "fit_gpus",
    "GGUFMetadata",
    "parse_header_bytes",
    "parse_header_with_tensors",
    "parse_local_file",
    "parse_hf_range",
    "metadata_to_arch",
    "ModelArch",
    "Inputs",
    "Breakdown",
    "estimate",
    "command_preview",
    "format_bytes",
]