Spaces:

prithivMLmods
/

Multimodal-OCR3

Running on Zero

prithivMLmods commited on Oct 17

Commit

e27521f

verified ·

1 Parent(s): 4d60d58

update app

Files changed (1) hide show

app.py CHANGED Viewed

@@ -129,7 +129,8 @@ processor_v = AutoProcessor.from_pretrained(MODEL_ID_V, trust_remote_code=True)
 model_v = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     MODEL_ID_V,
     trust_remote_code=True,
-    torch_dtype=torch.float16
 ).to(device).eval()
 # Load Dots.OCR

 model_v = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     MODEL_ID_V,
     trust_remote_code=True,
+    torch_dtype=torch.float16,
+    attn_implementation="flash_attention_2"
 ).to(device).eval()
 # Load Dots.OCR