Update app.py

app.py

@@ -22,6 +22,7 @@ dataset = load_dataset("mskov/miso_test", split="test").cast_column("audio", Aud
 
 print(dataset, "and at 0[audio][array] ", dataset[0]["audio"]["array"], type(dataset[0]["audio"]["array"]), "and at audio : ", dataset[0]["audio"])
 
+
 def transcribe(audio):
     text = pipe(audio)["text"]
     return text
@@ -35,27 +36,73 @@ iface = gr.Interface(
 
 iface.launch()
 
+def evalWhisper(model, dataset):
+        model.eval()
+    print("model.eval ", model.eval())
+    
+    # Define a list to store the print statements
+    log_texts = []
+    
+    with torch.no_grad():
+        outputs = model(**input_data)  # Define input_data appropriately
+        print("outputs ", outputs)
+        log_texts.append(f"outputs: {outputs}")
+    
+    # Convert predicted token IDs back to text
+    predicted_text = tokenizer.batch_decode(outputs.logits.argmax(dim=-1), skip_special_tokens=True)
+    
+    # Get ground truth labels from the dataset
+    labels = dataset["audio"]  # Replace "labels" with the appropriate key in your dataset
+    print("labels are ", labels)
+    log_texts.append(f"labels: {labels}")
+    
+    # Compute WER
+    wer_score = wer(labels, predicted_text)  # Define wer function
+    
+    # Print or return WER score
+    wer_message = f"Word Error Rate (WER): {wer_score}"
+    print(wer_message)
+    log_texts.append(wer_message)
 
-# Evaluate the model
-model.eval()
-print("model.eval ", model.eval())
-with torch.no_grad():
-    outputs = model(input_ids=input_ids, attention_mask=attention_mask)
-    print("outputs ", outputs)
-
-# Convert predicted token IDs back to text
-predicted_text = tokenizer.batch_decode(outputs.logits.argmax(dim=-1), skip_special_tokens=True)
+    print(log_texts)
+    
+    return log_texts
 
-# Get ground truth labels from the dataset
-labels = dataset["audio"]  # Replace "labels" with the appropriate key in your dataset
-print("labels are ", labels)
+# Call evalWhisper and get the log texts
+log_texts = evalWhisper(model, dataset)
 
-# Compute WER
-wer_score = wer(labels, predicted_text)
+# Display the log texts using gr.Interface
+log_text = "\n".join(log_texts)
+log_interface = gr.Interface(
+    fn=lambda: log_text,
+    inputs=None,
+    outputs="text",
+    title="EvalWhisper Log",
+)
+log_interface.launch()
 
-# Print or return WER score
-print(f"Word Error Rate (WER): {wer_score}")
 
+'''
+    # Evaluate the model
+    model.eval()
+    print("model.eval ", model.eval())
+    with torch.no_grad():
+        outputs = model(input_ids=input_ids, attention_mask=attention_mask)
+        print("outputs ", outputs)
+    
+    # Convert predicted token IDs back to text
+    predicted_text = tokenizer.batch_decode(outputs.logits.argmax(dim=-1), skip_special_tokens=True)
+    
+    # Get ground truth labels from the dataset
+    labels = dataset["audio"]  # Replace "labels" with the appropriate key in your dataset
+    print("labels are ", labels)
+    
+    # Compute WER
+    wer_score = wer(labels, predicted_text)
+    
+    # Print or return WER score
+    print(f"Word Error Rate (WER): {wer_score}")
+'''
 '''
 print("check check")
 print(inputs)