Update app.py
Browse files
app.py
CHANGED
@@ -28,70 +28,56 @@ processor = AutoProcessor.from_pretrained(model_id)
|
|
28 |
# Confirming and setting the eos_token_id (if necessary)
|
29 |
model.generation_config.eos_token_id = processor.tokenizer.eos_token_id
|
30 |
|
31 |
-
@spaces.GPU
|
32 |
-
def
|
33 |
-
print(
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
print(f"Image path: {image_path}")
|
41 |
else:
|
|
|
|
|
42 |
for hist in history:
|
43 |
-
if
|
44 |
-
|
45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
46 |
|
47 |
-
|
48 |
-
|
49 |
-
|
|
|
50 |
|
51 |
-
|
52 |
-
|
53 |
-
image.show() # Show the image to confirm it's loaded
|
54 |
-
print(f"Image open: {image}")
|
55 |
-
except Exception as e:
|
56 |
-
print(f"Error opening image: {e}")
|
57 |
-
gr.Error("Failed to open the image.")
|
58 |
-
return
|
59 |
-
|
60 |
-
# Adding more context to the prompt with a placeholder for the image
|
61 |
-
prompt = f"user: Here is an image and a question about it.\n<image>{input['text']}\nassistant: "
|
62 |
-
print("Made the prompt")
|
63 |
|
64 |
-
|
65 |
-
inputs = processor(text=prompt, images=image, return_tensors='pt').to('cuda', torch.float16)
|
66 |
-
print(f"Processed inputs: {inputs}")
|
67 |
-
except Exception as e:
|
68 |
-
print(f"Error processing inputs: {e}")
|
69 |
-
gr.Error("Failed to process the inputs.")
|
70 |
-
return
|
71 |
-
|
72 |
-
# Streamer
|
73 |
-
print('About to init streamer')
|
74 |
-
streamer = TextIteratorStreamer(processor.tokenizer, skip_special_tokens=False, skip_prompt=True)
|
75 |
-
|
76 |
-
# Generation kwargs
|
77 |
-
generation_kwargs = dict(
|
78 |
-
inputs=inputs['input_ids'],
|
79 |
-
attention_mask=inputs['attention_mask'],
|
80 |
-
streamer=streamer,
|
81 |
-
max_new_tokens=1024,
|
82 |
-
do_sample=False
|
83 |
-
)
|
84 |
-
|
85 |
-
thread = threading.Thread(target=model.generate, kwargs=generation_kwargs)
|
86 |
-
print('Thread about to start')
|
87 |
thread.start()
|
88 |
-
|
|
|
|
|
|
|
89 |
buffer = ""
|
90 |
-
|
91 |
for new_text in streamer:
|
|
|
|
|
|
|
92 |
buffer += new_text
|
|
|
|
|
93 |
generated_text_without_prompt = buffer
|
94 |
-
#
|
|
|
|
|
95 |
yield generated_text_without_prompt
|
96 |
|
97 |
|
|
|
28 |
# Confirming and setting the eos_token_id (if necessary)
|
29 |
model.generation_config.eos_token_id = processor.tokenizer.eos_token_id
|
30 |
|
31 |
+
@spaces.GPU
|
32 |
+
def bot_streaming(message, history):
|
33 |
+
print(message)
|
34 |
+
if message["files"]:
|
35 |
+
# message["files"][-1] is a Dict or just a string
|
36 |
+
if type(message["files"][-1]) == dict:
|
37 |
+
image = message["files"][-1]["path"]
|
38 |
+
else:
|
39 |
+
image = message["files"][-1]
|
|
|
40 |
else:
|
41 |
+
# if there's no image uploaded for this turn, look for images in the past turns
|
42 |
+
# kept inside tuples, take the last one
|
43 |
for hist in history:
|
44 |
+
if type(hist[0]) == tuple:
|
45 |
+
image = hist[0][0]
|
46 |
+
try:
|
47 |
+
if image is None:
|
48 |
+
# Handle the case where image is None
|
49 |
+
gr.Error("You need to upload an image for LLaVA to work.")
|
50 |
+
except NameError:
|
51 |
+
# Handle the case where 'image' is not defined at all
|
52 |
+
gr.Error("You need to upload an image for LLaVA to work.")
|
53 |
|
54 |
+
prompt = f"<|start_header_id|>user<|end_header_id|>\n\n<image>\n{message['text']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
|
55 |
+
# print(f"prompt: {prompt}")
|
56 |
+
image = Image.open(image)
|
57 |
+
inputs = processor(prompt, image, return_tensors='pt').to(0, torch.float16)
|
58 |
|
59 |
+
streamer = TextIteratorStreamer(processor, **{"skip_special_tokens": False, "skip_prompt": True})
|
60 |
+
generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024, do_sample=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
61 |
|
62 |
+
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
63 |
thread.start()
|
64 |
+
|
65 |
+
text_prompt = f"<|start_header_id|>user<|end_header_id|>\n\n{message['text']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
|
66 |
+
# print(f"text_prompt: {text_prompt}")
|
67 |
+
|
68 |
buffer = ""
|
69 |
+
time.sleep(0.5)
|
70 |
for new_text in streamer:
|
71 |
+
# find <|eot_id|> and remove it from the new_text
|
72 |
+
if "<|eot_id|>" in new_text:
|
73 |
+
new_text = new_text.split("<|eot_id|>")[0]
|
74 |
buffer += new_text
|
75 |
+
|
76 |
+
# generated_text_without_prompt = buffer[len(text_prompt):]
|
77 |
generated_text_without_prompt = buffer
|
78 |
+
# print(generated_text_without_prompt)
|
79 |
+
time.sleep(0.06)
|
80 |
+
# print(f"new_text: {generated_text_without_prompt}")
|
81 |
yield generated_text_without_prompt
|
82 |
|
83 |
|