Spaces:
Sleeping
Sleeping
Create app/models.py
Browse files- app/models.py +96 -0
app/models.py
ADDED
|
@@ -0,0 +1,96 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
import os
|
| 3 |
+
import threading
|
| 4 |
+
import torch
|
| 5 |
+
|
| 6 |
+
from functools import lru_cache
|
| 7 |
+
from transformers import pipeline
|
| 8 |
+
from gliner import GLiNER
|
| 9 |
+
from sentence_transformers import SentenceTransformer
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
| 13 |
+
|
| 14 |
+
DTYPE = (
|
| 15 |
+
torch.float16
|
| 16 |
+
if DEVICE == "cuda"
|
| 17 |
+
else torch.float32
|
| 18 |
+
)
|
| 19 |
+
|
| 20 |
+
NER_MODEL = os.getenv(
|
| 21 |
+
"NER_MODEL",
|
| 22 |
+
"urchade/gliner_multi-v2.1"
|
| 23 |
+
)
|
| 24 |
+
|
| 25 |
+
SENTIMENT_MODEL = os.getenv(
|
| 26 |
+
"SENTIMENT_MODEL",
|
| 27 |
+
"nahiar/sentiment-analysis-v2"
|
| 28 |
+
)
|
| 29 |
+
|
| 30 |
+
SARCASM_MODEL = os.getenv(
|
| 31 |
+
"SARCASM_MODEL",
|
| 32 |
+
"helinivan/multilingual-sarcasm-detector"
|
| 33 |
+
)
|
| 34 |
+
|
| 35 |
+
EMBEDDING_MODEL = os.getenv(
|
| 36 |
+
"EMBEDDING_MODEL",
|
| 37 |
+
"sentence-transformers/paraphrase-multilingual-mpnet-base-v2"
|
| 38 |
+
)
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
class ModelRegistry:
|
| 42 |
+
"""Lazy-loaded, process-local model registry."""
|
| 43 |
+
|
| 44 |
+
def __init__(self):
|
| 45 |
+
self._models = {}
|
| 46 |
+
self._lock = threading.RLock()
|
| 47 |
+
|
| 48 |
+
def get(self, name, loader):
|
| 49 |
+
if name not in self._models:
|
| 50 |
+
with self._lock:
|
| 51 |
+
if name not in self._models:
|
| 52 |
+
self._models[name] = loader()
|
| 53 |
+
|
| 54 |
+
return self._models[name]
|
| 55 |
+
|
| 56 |
+
def ner(self):
|
| 57 |
+
return self.get(
|
| 58 |
+
"ner",
|
| 59 |
+
lambda: GLiNER.from_pretrained(NER_MODEL)
|
| 60 |
+
)
|
| 61 |
+
|
| 62 |
+
def sentiment(self):
|
| 63 |
+
return self.get(
|
| 64 |
+
"sentiment",
|
| 65 |
+
lambda: pipeline(
|
| 66 |
+
"text-classification",
|
| 67 |
+
model=SENTIMENT_MODEL,
|
| 68 |
+
device=0 if DEVICE == "cuda" else -1,
|
| 69 |
+
torch_dtype=DTYPE,
|
| 70 |
+
truncation=True
|
| 71 |
+
)
|
| 72 |
+
)
|
| 73 |
+
|
| 74 |
+
def sarcasm(self):
|
| 75 |
+
return self.get(
|
| 76 |
+
"sarcasm",
|
| 77 |
+
lambda: pipeline(
|
| 78 |
+
"text-classification",
|
| 79 |
+
model=SARCASM_MODEL,
|
| 80 |
+
device=0 if DEVICE == "cuda" else -1,
|
| 81 |
+
torch_dtype=DTYPE,
|
| 82 |
+
truncation=True
|
| 83 |
+
)
|
| 84 |
+
)
|
| 85 |
+
|
| 86 |
+
def embedding(self):
|
| 87 |
+
return self.get(
|
| 88 |
+
"embedding",
|
| 89 |
+
lambda: SentenceTransformer(
|
| 90 |
+
EMBEDDING_MODEL,
|
| 91 |
+
device=DEVICE
|
| 92 |
+
)
|
| 93 |
+
)
|
| 94 |
+
|
| 95 |
+
|
| 96 |
+
models = ModelRegistry()
|