Spaces:
Runtime error
Runtime error
chat interface
Browse files
app.py
CHANGED
@@ -1,73 +1,50 @@
|
|
1 |
import spaces
|
2 |
import torch
|
3 |
-
import re
|
4 |
import gradio as gr
|
5 |
from threading import Thread
|
6 |
-
from transformers import
|
|
|
|
|
7 |
import subprocess
|
8 |
|
9 |
-
|
10 |
-
|
11 |
|
12 |
-
#
|
13 |
model_id = "vikhyatk/moondream2"
|
14 |
revision = "2024-04-02"
|
15 |
tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
|
16 |
moondream = AutoModelForCausalLM.from_pretrained(
|
17 |
-
model_id,
|
18 |
torch_dtype=torch.bfloat16, device_map={"": "cuda"},
|
19 |
-
attn_implementation="flash_attention_2"
|
|
|
20 |
moondream.eval()
|
21 |
|
22 |
-
|
23 |
@spaces.GPU(duration=10)
|
24 |
-
def
|
|
|
25 |
image_embeds = moondream.encode_image(img)
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
"image_embeds": image_embeds,
|
31 |
-
"question": prompt,
|
32 |
-
"tokenizer": tokenizer,
|
33 |
-
"streamer": streamer,
|
34 |
-
},
|
35 |
-
)
|
36 |
-
thread.start()
|
37 |
-
buffer = ""
|
38 |
-
for new_text in streamer:
|
39 |
-
buffer += new_text
|
40 |
-
yield buffer.strip()
|
41 |
-
|
42 |
-
# Create the Gradio interface
|
43 |
-
with gr.Blocks(theme="Monochrome") as demo:
|
44 |
-
gr.Markdown(
|
45 |
-
"""
|
46 |
-
# AskMoondream: Moondream 2 Demonstration Space
|
47 |
-
Moondream2 is a 1.86B parameter model initialized with weights from SigLIP and Phi 1.5.
|
48 |
-
Modularity AI presents this open source huggingface space for running fast experimental inferences on Moondream2.
|
49 |
-
"""
|
50 |
-
)
|
51 |
|
52 |
-
# Chatbot layout
|
53 |
-
chatbot = gr.Chatbot()
|
54 |
|
55 |
-
|
|
|
|
|
56 |
with gr.Row():
|
57 |
img = gr.Image(type="pil", label="Upload an Image")
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
# Function to send message and get response
|
64 |
-
def send_message(history, prompt):
|
65 |
-
history.append((prompt, None))
|
66 |
-
response = answer_question(img.value, prompt)
|
67 |
-
history.append((None, response))
|
68 |
-
return history, "" # Clear the input box
|
69 |
|
70 |
-
|
71 |
-
|
|
|
72 |
|
73 |
-
demo
|
|
|
|
1 |
import spaces
|
2 |
import torch
|
|
|
3 |
import gradio as gr
|
4 |
from threading import Thread
|
5 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
6 |
+
|
7 |
+
# Install the necessary package for the model
|
8 |
import subprocess
|
9 |
|
10 |
+
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"},
|
11 |
+
shell=True)
|
12 |
|
13 |
+
# Initialize the tokenizer and model
|
14 |
model_id = "vikhyatk/moondream2"
|
15 |
revision = "2024-04-02"
|
16 |
tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
|
17 |
moondream = AutoModelForCausalLM.from_pretrained(
|
18 |
+
model_id, revision=revision, trust_remote_code=True,
|
19 |
torch_dtype=torch.bfloat16, device_map={"": "cuda"},
|
20 |
+
attn_implementation="flash_attention_2"
|
21 |
+
)
|
22 |
moondream.eval()
|
23 |
|
24 |
+
|
25 |
@spaces.GPU(duration=10)
|
26 |
+
def chatbot_response(img, text_input):
|
27 |
+
# Here we assume an encoded image processing if needed
|
28 |
image_embeds = moondream.encode_image(img)
|
29 |
+
inputs = tokenizer.encode(text_input, return_tensors="pt")
|
30 |
+
outputs = moondream.generate(inputs, max_length=200)
|
31 |
+
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
32 |
+
return response
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
|
|
|
|
|
34 |
|
35 |
+
# Setting up Gradio Interface
|
36 |
+
with gr.Blocks(theme="Monochrome") as demo:
|
37 |
+
gr.Markdown("# AskMoondream Chatbot")
|
38 |
with gr.Row():
|
39 |
img = gr.Image(type="pil", label="Upload an Image")
|
40 |
+
text_input = gr.Textbox(label="Ask a question or describe an image", placeholder="Type here...")
|
41 |
+
with gr.Row():
|
42 |
+
submit = gr.Button("Submit")
|
43 |
+
response = gr.TextArea(label="Response", placeholder="Moondream's response will appear here...")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
|
45 |
+
# Define what happens when the user interacts with the interface
|
46 |
+
submit.click(chatbot_response, inputs=[img, text_input], outputs=response)
|
47 |
+
text_input.submit(chatbot_response, inputs=[img, text_input], outputs=response)
|
48 |
|
49 |
+
# Launch the demo
|
50 |
+
demo.queue().launch()
|