Fizzarolli
commited on
Commit
•
fda67b7
1
Parent(s):
4e2b6d8
blehhhhh
Browse files
app.py
CHANGED
@@ -16,9 +16,9 @@ processor = AutoProcessor.from_pretrained(MODEL_ID)
|
|
16 |
|
17 |
@spaces.GPU
|
18 |
def tag_image(image, max_new_tokens=128):
|
19 |
-
inputs = processor(images=image, text="<image>tag en", return_tensors="pt")
|
20 |
streamer = TextIteratorStreamer(tokenizer=processor.tokenizer, skip_prompt=True)
|
21 |
-
generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=max_new_tokens)
|
22 |
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
23 |
|
24 |
text = ""
|
@@ -26,8 +26,6 @@ def tag_image(image, max_new_tokens=128):
|
|
26 |
for new_text in streamer:
|
27 |
text += new_text
|
28 |
yield text
|
29 |
-
|
30 |
-
thread.join()
|
31 |
|
32 |
return text
|
33 |
|
|
|
16 |
|
17 |
@spaces.GPU
|
18 |
def tag_image(image, max_new_tokens=128):
|
19 |
+
inputs = processor(images=image, text="<image>tag en", return_tensors="pt").to("cuda")
|
20 |
streamer = TextIteratorStreamer(tokenizer=processor.tokenizer, skip_prompt=True)
|
21 |
+
generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=max_new_tokens, use_cache=True, cache_implementation="hybrid")
|
22 |
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
23 |
|
24 |
text = ""
|
|
|
26 |
for new_text in streamer:
|
27 |
text += new_text
|
28 |
yield text
|
|
|
|
|
29 |
|
30 |
return text
|
31 |
|