Update app.py

app.py

@@ -51,19 +51,19 @@ print("Using device:", device)
 MODEL_ID_M = "LiquidAI/LFM2-VL-450M"
 processor_m = AutoProcessor.from_pretrained(MODEL_ID_M, trust_remote_code=True)
 model_m = AutoModelForImageTextToText.from_pretrained(
-    MODEL_ID_M, trust_remote_code=True, torch_dtype=torch.float16
+    MODEL_ID_M, trust_remote_code=True, torch_dtype=torch.bfloat16
 ).to(device).eval()
 
 MODEL_ID_T = "LiquidAI/LFM2-VL-1.6B"
 processor_t = AutoProcessor.from_pretrained(MODEL_ID_T, trust_remote_code=True)
 model_t = AutoModelForImageTextToText.from_pretrained(
-    MODEL_ID_T, trust_remote_code=True, torch_dtype=torch.float16
+    MODEL_ID_T, trust_remote_code=True, torch_dtype=torch.bfloat16
 ).to(device).eval()
 
 MODEL_ID_C = "HuggingFaceTB/SmolVLM-Instruct-250M"
 processor_c = AutoProcessor.from_pretrained(MODEL_ID_C, trust_remote_code=True)
 model_c = AutoModelForVision2Seq.from_pretrained(
-    MODEL_ID_C, trust_remote_code=True, torch_dtype=torch.float16, _attn_implementation="flash_attention_2"
+    MODEL_ID_C, trust_remote_code=True, torch_dtype=torch.bfloat16, _attn_implementation="flash_attention_2"
 ).to(device).eval()
 
 MODEL_ID_G = "echo840/MonkeyOCR"
@@ -72,13 +72,13 @@ processor_g = AutoProcessor.from_pretrained(
     MODEL_ID_G, trust_remote_code=True, subfolder=SUBFOLDER
 )
 model_g = Qwen2_5_VLForConditionalGeneration.from_pretrained(
-    MODEL_ID_G, trust_remote_code=True, subfolder=SUBFOLDER, torch_dtype=torch.float16
+    MODEL_ID_G, trust_remote_code=True, subfolder=SUBFOLDER, torch_dtype=torch.bfloat16
 ).to(device).eval()
 
 MODEL_ID_I = "HuggingFaceTB/SmolVLM2-2.2B-Instruct"
 processor_i = AutoProcessor.from_pretrained(MODEL_ID_I, trust_remote_code=True)
 model_i = AutoModelForImageTextToText.from_pretrained(
-    MODEL_ID_I, trust_remote_code=True, torch_dtype=torch.float16, _attn_implementation="flash_attention_2"
+    MODEL_ID_I, trust_remote_code=True, torch_dtype=torch.bfloat16, _attn_implementation="flash_attention_2"
 ).to(device).eval()