Upload folder using huggingface_hub
Browse files- config.json +12 -1
- spancnn_pipeline.py +24 -0
config.json
CHANGED
@@ -31,5 +31,16 @@
|
|
31 |
"transformers_version": "4.28.1",
|
32 |
"type_vocab_size": 2,
|
33 |
"use_cache": true,
|
34 |
-
"vocab_size": 30522
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
35 |
}
|
|
|
31 |
"transformers_version": "4.28.1",
|
32 |
"type_vocab_size": 2,
|
33 |
"use_cache": true,
|
34 |
+
"vocab_size": 30522,
|
35 |
+
"custom_pipelines": {
|
36 |
+
"spancnn-classification": {
|
37 |
+
"impl": "spancnn_pipeline.SpanClassificationPipeline",
|
38 |
+
"pt": [
|
39 |
+
"AutoModelForSequenceClassification"
|
40 |
+
],
|
41 |
+
"tf": [
|
42 |
+
"TFAutoModelForSequenceClassification"
|
43 |
+
]
|
44 |
+
}
|
45 |
+
}
|
46 |
}
|
spancnn_pipeline.py
ADDED
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import pipeline, Pipeline, AutoTokenizer, AutoConfig, AutoModelForSequenceClassification
|
2 |
+
from transformers.pipelines import PIPELINE_REGISTRY
|
3 |
+
import torch
|
4 |
+
|
5 |
+
class SpanClassificationPipeline(Pipeline):
|
6 |
+
def __init__(self, model, tokenizer, device="cpu", **kwargs):
|
7 |
+
super().__init__(model=model, tokenizer=tokenizer, device=device, **kwargs)
|
8 |
+
self.model.to(self.device)
|
9 |
+
self.model.eval()
|
10 |
+
|
11 |
+
def _sanitize_parameters(self, **kwargs):
|
12 |
+
return {}, kwargs, {}
|
13 |
+
|
14 |
+
def preprocess(self, inputs):
|
15 |
+
return self.tokenizer(inputs, return_tensors="pt").to(self.device)
|
16 |
+
|
17 |
+
def _forward(self, model_inputs):
|
18 |
+
with torch.no_grad():
|
19 |
+
outputs = self.model(**model_inputs)
|
20 |
+
return outputs
|
21 |
+
|
22 |
+
def postprocess(self, model_outputs):
|
23 |
+
logits = model_outputs.logits
|
24 |
+
return int(torch.argmax(logits, dim=1).item())
|