.

inference/infer_single.py

@@ -46,7 +46,7 @@ def generate_and_score_essay(topic, essay):
         truncation=True,
         padding_side='left'
     ).to(device)
-    with torch.no_grad(), autocast(device_type='cuda', dtype=torch.float16):
+    with torch.no_grad():
         outputs = QWEN_MODEL.generate(
             **inputs,
             generation_config=gen_config,
@@ -86,7 +86,7 @@ def generate_and_score_essay(topic, essay):
         padding=True
     ).to(device)
     LONGFORMER_MODEL.eval()
-    with torch.no_grad(), autocast(device_type='cuda', dtype=torch.float16):
+    with torch.no_grad():
         outputs = LONGFORMER_MODEL(**score_inputs)  # Get full outputs dictionary
         scores = outputs['logits'].cpu().numpy()
     scores = [round(x) for x in scores[0]]