From 1622059179ee2da36b9c46c3aa32752eb45b4214 Mon Sep 17 00:00:00 2001
From: oobabooga <112222186+oobabooga@users.noreply.github.com>
Date: Wed, 15 Feb 2023 00:03:19 -0300
Subject: [PATCH] Move BLIP to the CPU

It's just as fast
---
 modules/bot_picture.py | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/modules/bot_picture.py b/modules/bot_picture.py
index f19e398a..f407c379 100644
--- a/modules/bot_picture.py
+++ b/modules/bot_picture.py
@@ -5,10 +5,9 @@ from transformers import BlipForConditionalGeneration
 from transformers import BlipProcessor
 
 processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
-model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base", torch_dtype=torch.float16).to("cuda")
+model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base", torch_dtype=torch.float32).to("cpu")
 
-# raw_image = Image.open('/tmp/istockphoto-470604022-612x612.jpg').convert('RGB')
 def caption_image(raw_image):
-    inputs = processor(raw_image, return_tensors="pt").to("cuda", torch.float16)
+    inputs = processor(raw_image.convert('RGB'), return_tensors="pt").to("cpu", torch.float32)
     out = model.generate(**inputs, max_new_tokens=100)
     return processor.decode(out[0], skip_special_tokens=True)