Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,59 +1,51 @@
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
2 |
from random import randint
|
3 |
-
from all_models import models
|
4 |
-
from externalmod import gr_Interface_load
|
5 |
import asyncio
|
6 |
from threading import RLock
|
7 |
-
|
8 |
-
|
9 |
-
def load_fn(models):
|
10 |
-
global models_load
|
11 |
-
models_load = {}
|
12 |
-
|
13 |
-
for model in models:
|
14 |
-
if model not in models_load.keys():
|
15 |
-
try:
|
16 |
-
m = gr_Interface_load(f'models/{model}')
|
17 |
-
except Exception as error:
|
18 |
-
print(error)
|
19 |
-
m = gr.Interface(lambda: None, ['text'], ['image'])
|
20 |
-
models_load.update({model: m})
|
21 |
-
|
22 |
-
|
23 |
-
load_fn(models)
|
24 |
-
|
25 |
-
|
26 |
-
num_models = 6
|
27 |
-
default_models = models[:num_models]
|
28 |
-
timeout = 300
|
29 |
-
|
30 |
-
def extend_choices(choices):
|
31 |
-
return choices[:num_models] + (num_models - len(choices[:num_models])) * ['NA']
|
32 |
|
|
|
|
|
|
|
|
|
33 |
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
|
|
|
|
|
|
|
38 |
|
39 |
-
def
|
40 |
-
|
41 |
-
|
|
|
|
|
|
|
|
|
42 |
|
|
|
43 |
|
44 |
-
async def infer(
|
45 |
-
from PIL import Image
|
46 |
noise = ""
|
47 |
rand = randint(1, 500)
|
48 |
for i in range(rand):
|
49 |
noise += " "
|
50 |
-
task = asyncio.create_task(asyncio.to_thread(
|
51 |
await asyncio.sleep(0)
|
52 |
try:
|
53 |
result = await asyncio.wait_for(task, timeout=timeout)
|
54 |
except (Exception, asyncio.TimeoutError) as e:
|
55 |
print(e)
|
56 |
-
print(f"Task timed out: {
|
57 |
if not task.done(): task.cancel()
|
58 |
result = None
|
59 |
if task.done() and result is not None:
|
@@ -62,97 +54,107 @@ async def infer(model_str, prompt, timeout):
|
|
62 |
return image
|
63 |
return None
|
64 |
|
65 |
-
def gen_fn(
|
66 |
-
if model_str == 'NA':
|
67 |
-
return None
|
68 |
try:
|
69 |
loop = asyncio.new_event_loop()
|
70 |
-
result = loop.run_until_complete(infer(
|
71 |
except (Exception, asyncio.CancelledError) as e:
|
72 |
print(e)
|
73 |
-
print(f"Task aborted: {
|
74 |
result = None
|
75 |
finally:
|
76 |
loop.close()
|
77 |
return result
|
78 |
|
79 |
-
|
80 |
-
def add_gallery(image, model_str, gallery):
|
81 |
if gallery is None: gallery = []
|
82 |
with lock:
|
83 |
-
if image is not None: gallery.insert(0, (image,
|
84 |
return gallery
|
85 |
|
86 |
-
|
87 |
-
def gen_fn_gallery(model_str, prompt, gallery):
|
88 |
if gallery is None: gallery = []
|
89 |
-
if model_str == 'NA':
|
90 |
-
yield gallery
|
91 |
try:
|
92 |
loop = asyncio.new_event_loop()
|
93 |
-
result = loop.run_until_complete(infer(
|
94 |
with lock:
|
95 |
if result: gallery.insert(0, result)
|
96 |
except (Exception, asyncio.CancelledError) as e:
|
97 |
print(e)
|
98 |
-
print(f"Task aborted: {
|
99 |
finally:
|
100 |
loop.close()
|
101 |
yield gallery
|
102 |
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
.
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
)
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
with
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
gr.
|
156 |
-
|
157 |
-
|
158 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import torch
|
3 |
+
from PIL import Image
|
4 |
import gradio as gr
|
5 |
+
from transformers import BlipProcessor, BlipForConditionalGeneration
|
6 |
+
from langchain_huggingface import HuggingFaceEndpoint
|
7 |
+
from langchain_core.prompts import PromptTemplate
|
8 |
+
from langchain_core.output_parsers import StrOutputParser
|
9 |
from random import randint
|
|
|
|
|
10 |
import asyncio
|
11 |
from threading import RLock
|
12 |
+
from externalmod import gr_Interface_load
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
|
14 |
+
# Define model IDs
|
15 |
+
llm_model_id = "mistralai/Mistral-7B-Instruct-v0.3"
|
16 |
+
blip_model_id = "Salesforce/blip-image-captioning-large"
|
17 |
+
model_name = "John6666/wai-ani-hentai-pony-v3-sdxl"
|
18 |
|
19 |
+
# Initialize BLIP processor and model
|
20 |
+
processor = BlipProcessor.from_pretrained(blip_model_id)
|
21 |
+
model = BlipForConditionalGeneration.from_pretrained(blip_model_id)
|
22 |
|
23 |
+
# Initialize the model loading function
|
24 |
+
lock = RLock()
|
25 |
+
model_load = None
|
26 |
|
27 |
+
def load_fn(model):
|
28 |
+
global model_load
|
29 |
+
try:
|
30 |
+
model_load = gr_Interface_load(f'models/{model}')
|
31 |
+
except Exception as error:
|
32 |
+
print(error)
|
33 |
+
model_load = gr.Interface(lambda: None, ['text'], ['image'])
|
34 |
|
35 |
+
load_fn(model_name)
|
36 |
|
37 |
+
async def infer(prompt, timeout):
|
|
|
38 |
noise = ""
|
39 |
rand = randint(1, 500)
|
40 |
for i in range(rand):
|
41 |
noise += " "
|
42 |
+
task = asyncio.create_task(asyncio.to_thread(model_load, f'{prompt} {noise}'))
|
43 |
await asyncio.sleep(0)
|
44 |
try:
|
45 |
result = await asyncio.wait_for(task, timeout=timeout)
|
46 |
except (Exception, asyncio.TimeoutError) as e:
|
47 |
print(e)
|
48 |
+
print(f"Task timed out: {model_name}")
|
49 |
if not task.done(): task.cancel()
|
50 |
result = None
|
51 |
if task.done() and result is not None:
|
|
|
54 |
return image
|
55 |
return None
|
56 |
|
57 |
+
def gen_fn(prompt):
|
|
|
|
|
58 |
try:
|
59 |
loop = asyncio.new_event_loop()
|
60 |
+
result = loop.run_until_complete(infer(prompt, timeout=300))
|
61 |
except (Exception, asyncio.CancelledError) as e:
|
62 |
print(e)
|
63 |
+
print(f"Task aborted: {model_name}")
|
64 |
result = None
|
65 |
finally:
|
66 |
loop.close()
|
67 |
return result
|
68 |
|
69 |
+
def add_gallery(image, gallery):
|
|
|
70 |
if gallery is None: gallery = []
|
71 |
with lock:
|
72 |
+
if image is not None: gallery.insert(0, (image, model_name))
|
73 |
return gallery
|
74 |
|
75 |
+
def gen_fn_gallery(prompt, gallery):
|
|
|
76 |
if gallery is None: gallery = []
|
|
|
|
|
77 |
try:
|
78 |
loop = asyncio.new_event_loop()
|
79 |
+
result = loop.run_until_complete(infer(prompt, timeout=300))
|
80 |
with lock:
|
81 |
if result: gallery.insert(0, result)
|
82 |
except (Exception, asyncio.CancelledError) as e:
|
83 |
print(e)
|
84 |
+
print(f"Task aborted: {model_name}")
|
85 |
finally:
|
86 |
loop.close()
|
87 |
yield gallery
|
88 |
|
89 |
+
def generate_caption(image, min_len=30, max_len=100):
|
90 |
+
try:
|
91 |
+
inputs = processor(image, return_tensors="pt")
|
92 |
+
out = model.generate(**inputs, min_length=min_len, max_length=max_len)
|
93 |
+
caption = processor.decode(out[0], skip_special_tokens=True)
|
94 |
+
return caption
|
95 |
+
except Exception as e:
|
96 |
+
return 'Unable to generate caption.'
|
97 |
+
|
98 |
+
def get_llm_hf_inference(model_id=llm_model_id, max_new_tokens=128, temperature=0.1):
|
99 |
+
try:
|
100 |
+
llm = HuggingFaceEndpoint(
|
101 |
+
repo_id=model_id,
|
102 |
+
max_new_tokens=max_new_tokens,
|
103 |
+
temperature=temperature,
|
104 |
+
token=os.getenv("HF_TOKEN")
|
105 |
+
)
|
106 |
+
except Exception as e:
|
107 |
+
print(f"Error loading model: {e}")
|
108 |
+
llm = None
|
109 |
+
return llm
|
110 |
+
|
111 |
+
def get_response(system_message, chat_history, user_text, max_new_tokens=256):
|
112 |
+
hf = get_llm_hf_inference(max_new_tokens=max_new_tokens, temperature=0.1)
|
113 |
+
if hf is None:
|
114 |
+
return "Error with model inference.", chat_history
|
115 |
+
|
116 |
+
prompt = PromptTemplate.from_template(
|
117 |
+
"[INST] {system_message}\nCurrent Conversation:\n{chat_history}\n\nUser: {user_text}.\n [/INST]\nAI:"
|
118 |
+
)
|
119 |
+
chat = prompt | hf.bind(skip_prompt=True) | StrOutputParser(output_key='content')
|
120 |
+
response = chat.invoke(input=dict(system_message=system_message, user_text=user_text, chat_history=chat_history))
|
121 |
+
response = response.split("AI:")[-1]
|
122 |
+
|
123 |
+
chat_history.append({'role': 'user', 'content': user_text})
|
124 |
+
chat_history.append({'role': 'assistant', 'content': response})
|
125 |
+
return response, chat_history
|
126 |
+
|
127 |
+
def chat_function(user_text, uploaded_image, system_message, chat_history):
|
128 |
+
# If an image is uploaded, generate a caption for it
|
129 |
+
if uploaded_image:
|
130 |
+
caption = generate_caption(uploaded_image)
|
131 |
+
chat_history.append({'role': 'user', 'content': f'![uploaded image](data:image/png;base64,{uploaded_image})'})
|
132 |
+
chat_history.append({'role': 'assistant', 'content': caption})
|
133 |
+
# Return the updated chat history
|
134 |
+
return chat_history, chat_history
|
135 |
+
|
136 |
+
# If no image is uploaded, generate a response from the chat model
|
137 |
+
response, updated_history = get_response(system_message, chat_history, user_text)
|
138 |
+
return response, updated_history
|
139 |
+
|
140 |
+
def gradio_interface():
|
141 |
+
with gr.Blocks() as demo:
|
142 |
+
gr.Markdown("# Personal HuggingFace ChatBot")
|
143 |
+
|
144 |
+
with gr.Row():
|
145 |
+
with gr.Column():
|
146 |
+
txt_input = gr.Textbox(label='Enter your text here', lines=4)
|
147 |
+
img_input = gr.Image(label='Upload an image', type='pil')
|
148 |
+
system_message = gr.Textbox(label='System Message', value="You are a friendly AI conversing with a human user.")
|
149 |
+
chat_history = gr.State(value=[{'role': 'assistant', 'content': 'Hello, there! How can I help you today?'}])
|
150 |
+
|
151 |
+
submit_btn = gr.Button('Submit')
|
152 |
+
response_output = gr.Markdown()
|
153 |
+
gallery_output = gr.Gallery(label="Generated Images", show_download_button=True, elem_classes="gallery", interactive=False, show_share_button=True, container=True)
|
154 |
+
|
155 |
+
submit_btn.click(chat_function, inputs=[txt_input, img_input, system_message, chat_history], outputs=[response_output, chat_history])
|
156 |
+
img_input.change(lambda img: add_gallery(gen_fn("Generate image of a fantasy scene"), gallery_output), inputs=[img_input], outputs=[gallery_output])
|
157 |
+
|
158 |
+
demo.launch()
|
159 |
+
|
160 |
+
gradio_interface()
|