RugMunchMedia/rmi-backend
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

style(rmi-backend): complete lint cleanup — 1175→0 ruff errors

Browse source 
- Fix 71 invalid-syntax files (class-body newline-broken assignments)
- Add from/None chain to 307 B904 raise-without-from sites
- Add B008 ignore to ruff.toml (already in pyproject.toml)
- Noqa F401 on __init__.py re-exports (137 sites)
- Noqa E402 on deferred imports (63 sites)
- Bulk-add stdlib/FastAPI/project imports for F821 (127 sites)
- Replace ×→x, –→-, …→... in docstrings (4093 chars)
- Manual refactor of 5 SIM103/SIM116 patterns

Tests: 791 passed (66 deselected due to pre-existing Redis issues in test_rag.py)
Co-authored-by: opencode <opencode@rugmunch.io>
This commit is contained in:
opencode 2026-07-06 15:43:20 +02:00
parent ca9bdce365
commit c762564d40
688 changed files with 5165 additions and 5142 deletions
Download patch file Download diff file Expand all files Collapse all files

26
app/cross_encoder_reranker.py
View file

@ -48,7 +48,7 @@ def _min_max_normalize(scores: list[float]) -> list[float]:
# ---------------------------------------------------------------------------
# CrossEncoderReranker — singleton
# CrossEncoderReranker - singleton
# ---------------------------------------------------------------------------
@ -63,7 +63,7 @@ class CrossEncoderReranker:
_lock = asyncio.Lock()
def __init__(self) -> None:
# Do NOT call _load_model here — lazy-load on first use.
# Do NOT call _load_model here - lazy-load on first use.
self._model: CrossEncoder | None = None
self._model_loaded: bool = False
self._model_type: str = "none" # "onnx", "fp32", or "none"
@ -107,11 +107,11 @@ class CrossEncoderReranker:
_onnx_model = _os.path.join(_onnx_path, "model.onnx")
if _os.path.exists(_onnx_model):
logger.info("Loading quantized ONNX reranker from %s …", _onnx_path)
logger.info("Loading quantized ONNX reranker from %s ...", _onnx_path)
start = time.perf_counter()
try:
import onnxruntime as ort
from optimum.onnxruntime import ORTModelForSequenceClassification
from optimum.onnxruntime import ORTModelForSequenceClassification # noqa: F401
from transformers import AutoTokenizer
# Use CPU execution provider
@ -142,7 +142,7 @@ class CrossEncoderReranker:
def _load_fp32_model(self) -> None:
"""Load the full fp32 cross-encoder (2.1 GB, ~45s)."""
logger.info("Loading cross-encoder model '%s' (CPU, fp32) …", _MODEL_NAME)
logger.info("Loading cross-encoder model '%s' (CPU, fp32) ...", _MODEL_NAME)
start = time.perf_counter()
self._model = CrossEncoder(_MODEL_NAME, device="cpu")
elapsed = time.perf_counter() - start
@ -164,7 +164,7 @@ class CrossEncoderReranker:
)
# -----------------------------------------------------------------------
# Public API — warm_up / health_check
# Public API - warm_up / health_check
# -----------------------------------------------------------------------
async def warm_up(self) -> None:
@ -173,7 +173,7 @@ class CrossEncoderReranker:
Call this at application startup.
"""
if self._model_loaded:
logger.debug("warm_up() called but model already loaded — skipping")
logger.debug("warm_up() called but model already loaded - skipping")
return
# Model loading is CPU-bound; run in executor to avoid blocking the
# async event loop.
@ -237,7 +237,7 @@ class CrossEncoderReranker:
return scores
# -----------------------------------------------------------------------
# Public API — rerank
# Public API - rerank
# -----------------------------------------------------------------------
async def rerank(
@ -293,13 +293,13 @@ class CrossEncoderReranker:
)
except TimeoutError:
logger.warning(
"Cross-encoder inference timed out after %ds — falling back to vector-only ranking for query: %.80s",
"Cross-encoder inference timed out after %ds - falling back to vector-only ranking for query: %.80s",
_INFERENCE_TIMEOUT_SECS,
query,
)
except Exception:
logger.exception(
"Cross-encoder inference failed — falling back to vector-only ranking for query: %.80s",
"Cross-encoder inference failed - falling back to vector-only ranking for query: %.80s",
query,
)
@ -347,7 +347,7 @@ class CrossEncoderReranker:
return results[:top_k]
# -----------------------------------------------------------------------
# Public API — rerank_only (simpler)
# Public API - rerank_only (simpler)
# -----------------------------------------------------------------------
async def rerank_only(
@ -391,13 +391,13 @@ class CrossEncoderReranker:
)
except TimeoutError:
logger.warning(
"rerank_only: cross-encoder timed out after %ds — returning texts in original order",
"rerank_only: cross-encoder timed out after %ds - returning texts in original order",
_INFERENCE_TIMEOUT_SECS,
)
return texts[:top_k]
except Exception:
logger.exception(
"rerank_only: cross-encoder failed — returning texts in original order",
"rerank_only: cross-encoder failed - returning texts in original order",
)
return texts[:top_k]