Add application file

app.py

@@ -1,12 +1,13 @@
 from fastapi import FastAPI
+import onnxruntime as ort
 from transformers import AutoTokenizer
-from optimum.onnxruntime import ORTModelForSequenceClassification
-import torch
 import time
+import torch
 
-# Load the tokenizer and optimized model
+# Load the tokenizer and ONNX model
 tokenizer = AutoTokenizer.from_pretrained("cardiffnlp/twitter-roberta-base-sentiment")
-model = ORTModelForSequenceClassification.from_pretrained("cardiffnlp/twitter-roberta-base-sentiment", from_transformers=True)
+onnx_model_path = "D:/demodeploy/sentiment_model.onnx"
+onnx_session = ort.InferenceSession(onnx_model_path)
 
 app = FastAPI()
 
@@ -34,16 +35,22 @@ def analyze_sentiment(tweet: str):
 
     # Tokenize the input tweet
     inputs = tokenizer(tweet_proc, return_tensors="pt")
+    input_ids = inputs["input_ids"].numpy()
+    attention_mask = inputs["attention_mask"].numpy()
+
+    # Perform the inference using ONNX
+    onnx_inputs = {
+        'input_ids': input_ids,
+        'attention_mask': attention_mask
+    }
+    outputs = onnx_session.run(None, onnx_inputs)
 
-    # Perform the inference
-    with torch.no_grad():
-        outputs = model(**inputs)
-    
     # Calculate the inference time
     inference_time = time.time() - start_time
 
     # Get the probabilities from the logits
-    probabilities = torch.softmax(outputs.logits, dim=1)
+    logits = outputs[0]
+    probabilities = torch.softmax(torch.tensor(logits), dim=1)
 
     # Get the label with the highest probability
     max_prob, max_index = torch.max(probabilities, dim=1)
@@ -66,5 +73,3 @@ def analyze_sentiment(tweet: str):
         "score": highest_score,
         "inference_time": round(inference_time, 4)  # In seconds
     }
-
-

sentiment_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b357bdbb92348c9fed6359f64b0bc6dab75c78bde8c121ea49ac70dbb7eb9b5b
+size 498833392