IRIS / iris /config.py
CompleteKris's picture
Bundle IRIS core for Space build
b2300e2
Raw History Blame Contribute Delete
4.7 kB
from __future__ import annotations
import os
from dataclasses import dataclass
from pathlib import Path
def _bool_env(name: str, default: bool) -> bool:
value = os.getenv(name)
if value is None:
return default
return value.strip().lower() in {"1", "true", "yes", "on"}
def _float_env(name: str, default: float) -> float:
value = os.getenv(name)
return default if value is None else float(value)
def _int_env(name: str, default: int) -> int:
value = os.getenv(name)
return default if value is None else int(value)
@dataclass(frozen=True)
class Settings:
data_dir: Path
database_path: Path
asr_model: str
asr_timestamps: str
embedding_model: str
visual_model: str
device: str
torch_dtype: str
chunk_length_seconds: float
chunk_overlap_seconds: float
frame_sample_interval_seconds: float
max_audio_seconds: float | None
visual_search_enabled: bool
keyword_boost: float
semantic_weight: float
keyword_weight: float
min_speech_score: float
min_visual_score: float
@classmethod
def from_env(cls) -> "Settings":
data_dir = Path(os.getenv("IRIS_DATA_DIR", ".iris-data")).resolve()
return cls(
data_dir=data_dir,
database_path=Path(os.getenv("IRIS_DATABASE_PATH", data_dir / "iris.sqlite3")).resolve(),
asr_model=os.getenv("IRIS_ASR_MODEL", "openai/whisper-large-v3-turbo"),
asr_timestamps=os.getenv("IRIS_ASR_TIMESTAMPS", "word"),
embedding_model=os.getenv("IRIS_EMBEDDING_MODEL", "sentence-transformers/all-MiniLM-L6-v2"),
visual_model=os.getenv("IRIS_VISUAL_MODEL", "openai/clip-vit-large-patch14"),
device=os.getenv("IRIS_DEVICE", "auto"),
torch_dtype=os.getenv("IRIS_TORCH_DTYPE", "auto"),
chunk_length_seconds=_float_env("IRIS_CHUNK_LENGTH_SECONDS", 24.0),
chunk_overlap_seconds=_float_env("IRIS_CHUNK_OVERLAP_SECONDS", 4.0),
frame_sample_interval_seconds=_float_env("IRIS_FRAME_SAMPLE_INTERVAL_SECONDS", 2.0),
max_audio_seconds=_optional_float_env("IRIS_MAX_AUDIO_SECONDS"),
visual_search_enabled=_bool_env("IRIS_VISUAL_SEARCH_ENABLED", True),
keyword_boost=_float_env("IRIS_KEYWORD_BOOST", 0.15),
semantic_weight=_float_env("IRIS_SEMANTIC_WEIGHT", 0.78),
keyword_weight=_float_env("IRIS_KEYWORD_WEIGHT", 0.22),
min_speech_score=_float_env("IRIS_MIN_SPEECH_SCORE", 0.55),
min_visual_score=_float_env("IRIS_MIN_VISUAL_SCORE", 0.45),
)
def ensure_directories(self) -> None:
self.data_dir.mkdir(parents=True, exist_ok=True)
self.database_path.parent.mkdir(parents=True, exist_ok=True)
def resolve_device(device: str) -> str:
if device != "auto":
return device
try:
import torch
return "cuda" if torch.cuda.is_available() else "cpu"
except Exception:
return "cpu"
def transformers_pipeline_device(device: str) -> int:
resolved = resolve_device(device)
return 0 if resolved == "cuda" else -1
def resolve_torch_dtype(device: str, dtype: str):
normalized = dtype.strip().lower()
resolved_device = resolve_device(device)
if normalized == "auto":
normalized = "float16" if resolved_device == "cuda" else "float32"
try:
import torch
if normalized in {"float16", "fp16", "half"}:
return torch.float16
if normalized in {"bfloat16", "bf16"}:
return torch.bfloat16
if normalized in {"float32", "fp32", "full"}:
return torch.float32
except Exception:
pass
return "auto"
def device_info(device: str, dtype: str) -> dict:
resolved = resolve_device(device)
info = {
"requested_device": device,
"resolved_device": resolved,
"requested_dtype": dtype,
"cuda_available": False,
"cuda_version": None,
"gpu_count": 0,
"gpu_name": None,
}
try:
import torch
info.update(
{
"torch_version": torch.__version__,
"cuda_available": torch.cuda.is_available(),
"cuda_version": torch.version.cuda,
"gpu_count": torch.cuda.device_count(),
"gpu_name": torch.cuda.get_device_name(0) if torch.cuda.is_available() else None,
}
)
except Exception as exc:
info["torch_error"] = str(exc)
return info
def _optional_float_env(name: str) -> float | None:
value = os.getenv(name)
if value is None or value.strip() == "":
return None
return float(value)