68 lines
1.9 KiB
Python
68 lines
1.9 KiB
Python
from __future__ import annotations
|
|
|
|
from dataclasses import dataclass
|
|
from typing import Optional
|
|
|
|
|
|
_SEPARATION_ESTIMATES_MB: dict[str, dict[str, int]] = {
|
|
"melband_karaoke": {"eco": 2200, "standard": 2800, "performance": 4000},
|
|
"bs_roformer": {"eco": 1900, "standard": 2400, "performance": 3500},
|
|
"mdxnet": {"eco": 300, "standard": 300, "performance": 300},
|
|
"htdemucs": {"eco": 600, "standard": 1000, "performance": 1800},
|
|
}
|
|
|
|
_WHISPER_ESTIMATES_MB: dict[str, int] = {
|
|
"tiny": 300,
|
|
"tiny.en": 300,
|
|
"small": 600,
|
|
"small.en": 600,
|
|
"medium": 1700,
|
|
"medium.en": 1700,
|
|
"large-v2": 3200,
|
|
"large-v3": 3200,
|
|
"large-v3-turbo": 3200,
|
|
"turbo": 3200,
|
|
}
|
|
|
|
|
|
@dataclass
|
|
class VramBudgetExceededError(RuntimeError):
|
|
required_mb: int
|
|
free_mb: int
|
|
reserve_mb: int
|
|
|
|
def __str__(self) -> str:
|
|
return (
|
|
f"Not enough VRAM: job requires ~{self.required_mb} MB, "
|
|
f"only {self.free_mb} MB free ({self.reserve_mb} MB reserved)."
|
|
)
|
|
|
|
|
|
def estimate_separation_vram_mb(model_id: str, preset: str) -> Optional[int]:
|
|
model = (model_id or "").strip().lower()
|
|
tier = (preset or "standard").strip().lower() or "standard"
|
|
|
|
family = "htdemucs"
|
|
if model.startswith("mel_band"):
|
|
family = "melband_karaoke"
|
|
elif model.startswith("bs_"):
|
|
family = "bs_roformer"
|
|
elif model.startswith("mdxnet"):
|
|
family = "mdxnet"
|
|
elif model.startswith("htdemucs") or model.startswith("mdx"):
|
|
family = "htdemucs"
|
|
|
|
family_table = _SEPARATION_ESTIMATES_MB.get(family)
|
|
if not family_table:
|
|
return None
|
|
if tier not in family_table:
|
|
tier = "standard"
|
|
return family_table.get(tier)
|
|
|
|
|
|
def estimate_whisper_vram_mb(model_name: str) -> Optional[int]:
|
|
name = (model_name or "").strip().lower()
|
|
if not name:
|
|
return None
|
|
return _WHISPER_ESTIMATES_MB.get(name)
|