Add all answers in output
Browse files- handler.py +12 -4
handler.py
CHANGED
@@ -13,17 +13,25 @@ class EndpointHandler:
|
|
13 |
self.model = ViltForQuestionAnswering.from_pretrained(path)
|
14 |
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
15 |
|
16 |
-
def __call__(self, data: Dict[str, Any]) -> Dict[str,
|
17 |
# process input
|
18 |
inputs = data.pop("inputs", data)
|
19 |
image = inputs["image"]
|
20 |
image = Image.open(io.BytesIO(eval(image)))
|
21 |
text = inputs["text"]
|
22 |
-
|
23 |
# preprocess
|
24 |
encoding = self.processor(image, text, return_tensors="pt")
|
25 |
outputs = self.model(**encoding)
|
26 |
# postprocess the prediction
|
27 |
logits = outputs.logits
|
28 |
-
|
29 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
self.model = ViltForQuestionAnswering.from_pretrained(path)
|
14 |
self.device = "cuda" if torch.cuda.is_available() else "cpu"
|
15 |
|
16 |
+
def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
|
17 |
# process input
|
18 |
inputs = data.pop("inputs", data)
|
19 |
image = inputs["image"]
|
20 |
image = Image.open(io.BytesIO(eval(image)))
|
21 |
text = inputs["text"]
|
22 |
+
|
23 |
# preprocess
|
24 |
encoding = self.processor(image, text, return_tensors="pt")
|
25 |
outputs = self.model(**encoding)
|
26 |
# postprocess the prediction
|
27 |
logits = outputs.logits
|
28 |
+
best_idx = logits.argmax(-1).item()
|
29 |
+
best_answer = self.model.config.id2label[best_idx]
|
30 |
+
probabilities = torch.softmax(logits, dim=-1)[0]
|
31 |
+
id2label = self.model.config.id2label
|
32 |
+
answers = []
|
33 |
+
for idx, prob in enumerate(probabilities):
|
34 |
+
answer = id2label[idx]
|
35 |
+
answer_score = float(prob)
|
36 |
+
answers.append({"answer": answer, "answer_score": answer_score})
|
37 |
+
return {"best_answer": best_answer, "answers": answers}
|