Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,11 +1,12 @@
|
|
|
|
|
|
1 |
from PIL import Image
|
2 |
import gradio as gr
|
3 |
import spaces
|
|
|
4 |
import os
|
5 |
-
|
6 |
-
|
7 |
-
from llama_cpp import Llama
|
8 |
-
from llama_cpp.llama_chat_format import Llava15ChatHandler
|
9 |
|
10 |
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
|
11 |
MODEL_LIST = ["openbmb/MiniCPM-Llama3-V-2_5","openbmb/MiniCPM-Llama3-V-2_5-int4"]
|
@@ -26,108 +27,64 @@ CSS = """
|
|
26 |
}
|
27 |
"""
|
28 |
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
filename="ggml-model-Q5_K_M.gguf",
|
37 |
-
chat_handler=chat_handler,
|
38 |
-
n_ctx=4096,
|
39 |
-
verbose=True
|
40 |
-
)
|
41 |
|
42 |
-
'''
|
43 |
-
filenames = [
|
44 |
-
"*mmproj*",
|
45 |
-
"ggml-model-Q5_K_M.gguf"
|
46 |
-
]
|
47 |
|
48 |
-
|
49 |
-
downloaded_model_path = hf_hub_download(
|
50 |
-
repo_id="openbmb/MiniCPM-Llama3-V-2_5-gguf",
|
51 |
-
filename=filename,
|
52 |
-
local_dir="model"
|
53 |
-
)
|
54 |
-
'''
|
55 |
-
|
56 |
-
|
57 |
-
def image_to_base64_data_uri(file_path):
|
58 |
-
with open(file_path, "rb") as img_file:
|
59 |
-
base64_data = base64.b64encode(img_file.read()).decode('utf-8')
|
60 |
-
return f"data:image/png;base64,{base64_data}"
|
61 |
-
|
62 |
-
@spaces.GPU(queue=False)
|
63 |
def stream_chat(message, history: list, temperature: float, max_new_tokens: int):
|
64 |
print(f'message is - {message}')
|
65 |
print(f'history is - {history}')
|
66 |
-
|
67 |
-
|
68 |
if message["files"]:
|
69 |
-
image = message["files"][-1]
|
70 |
-
|
71 |
-
"role": "user",
|
72 |
-
"content": [
|
73 |
-
{"type": "text", "text": message['text']},
|
74 |
-
{"type": "image_url", "image_url":{"url": image}}
|
75 |
-
]
|
76 |
-
})
|
77 |
else:
|
78 |
if len(history) == 0:
|
79 |
raise gr.Error("Please upload an image first.")
|
80 |
image = None
|
81 |
else:
|
82 |
-
image = history[0][0][0]
|
83 |
for prompt, answer in history:
|
84 |
if answer is None:
|
85 |
-
|
86 |
-
"role": "user",
|
87 |
-
"content": [
|
88 |
-
{"type": "text", "text": prompt},
|
89 |
-
{"type": "image_url", "image_url": {"url": image}}
|
90 |
-
]
|
91 |
-
},{
|
92 |
-
"role": "assistant",
|
93 |
-
"content": ""
|
94 |
-
}])
|
95 |
else:
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
"content": answer
|
105 |
-
}])
|
106 |
-
messages.append({"role": "user", "content": message['text']})
|
107 |
-
print(f"Messages is -\n{messages}")
|
108 |
-
|
109 |
-
|
110 |
-
response = llm.create_chat_completion(
|
111 |
-
messages = messages,
|
112 |
temperature=temperature,
|
113 |
-
|
|
|
114 |
)
|
|
|
|
|
115 |
|
116 |
-
|
|
|
117 |
|
118 |
|
119 |
chatbot = gr.Chatbot(height=450)
|
120 |
chat_input = gr.MultimodalTextbox(
|
121 |
-
interactive=True,
|
122 |
-
file_types=["image"],
|
123 |
-
placeholder="Enter message or upload file...",
|
124 |
show_label=False,
|
125 |
|
126 |
)
|
127 |
EXAMPLES = [
|
128 |
-
[{"text": "
|
129 |
-
[{"text": "
|
130 |
-
[{"text": "
|
131 |
]
|
132 |
|
133 |
with gr.Blocks(css=CSS) as demo:
|
|
|
1 |
+
from threading import Thread
|
2 |
+
import torch
|
3 |
from PIL import Image
|
4 |
import gradio as gr
|
5 |
import spaces
|
6 |
+
from transformers import AutoModel, AutoTokenizer, TextIteratorStreamer
|
7 |
import os
|
8 |
+
|
9 |
+
|
|
|
|
|
10 |
|
11 |
os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"
|
12 |
MODEL_LIST = ["openbmb/MiniCPM-Llama3-V-2_5","openbmb/MiniCPM-Llama3-V-2_5-int4"]
|
|
|
27 |
}
|
28 |
"""
|
29 |
|
30 |
+
model = AutoModel.from_pretrained(
|
31 |
+
MODEL_ID,
|
32 |
+
torch_dtype=torch.float16,
|
33 |
+
trust_remote_code=True
|
34 |
+
).to(0)
|
35 |
+
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
|
36 |
+
model.eval()
|
|
|
|
|
|
|
|
|
|
|
37 |
|
|
|
|
|
|
|
|
|
|
|
38 |
|
39 |
+
@spaces.GPU()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
def stream_chat(message, history: list, temperature: float, max_new_tokens: int):
|
41 |
print(f'message is - {message}')
|
42 |
print(f'history is - {history}')
|
43 |
+
conversation = []
|
|
|
44 |
if message["files"]:
|
45 |
+
image = Image.open(message["files"][-1]).convert('RGB')
|
46 |
+
conversation.append({"role": "user", "content": message['text']})
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
else:
|
48 |
if len(history) == 0:
|
49 |
raise gr.Error("Please upload an image first.")
|
50 |
image = None
|
51 |
else:
|
52 |
+
image = Image.open(history[0][0][0])
|
53 |
for prompt, answer in history:
|
54 |
if answer is None:
|
55 |
+
conversation.extend([{"role": "user", "content": prompt},{"role": "assistant", "content": ""}])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
else:
|
57 |
+
conversation.extend([{"role": "user", "content": prompt}, {"role": "assistant", "content": answer}])
|
58 |
+
conversation.append({"role": "user", "content": message['text']})
|
59 |
+
print(f"Conversation is -\n{conversation}")
|
60 |
+
|
61 |
+
generate_kwargs = dict(
|
62 |
+
image=image,
|
63 |
+
msgs=conversation,
|
64 |
+
max_new_tokens=max_new_tokens,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
65 |
temperature=temperature,
|
66 |
+
sampling=True,
|
67 |
+
tokenizer=tokenizer,
|
68 |
)
|
69 |
+
if temperature == 0:
|
70 |
+
generate_kwargs["sampling"] = False
|
71 |
|
72 |
+
response = model.chat(**generate_kwargs)
|
73 |
+
return response
|
74 |
|
75 |
|
76 |
chatbot = gr.Chatbot(height=450)
|
77 |
chat_input = gr.MultimodalTextbox(
|
78 |
+
interactive=True,
|
79 |
+
file_types=["image"],
|
80 |
+
placeholder="Enter message or upload file...",
|
81 |
show_label=False,
|
82 |
|
83 |
)
|
84 |
EXAMPLES = [
|
85 |
+
[{"text": "Describe it in great detailed.", "files": ["./laptop.jpg"]}],
|
86 |
+
[{"text": "Describe it in great detailed.", "files": ["./hotel.jpg"]}],
|
87 |
+
[{"text": "Describe it in great detailed.", "files": ["./spacecat.png"]}]
|
88 |
]
|
89 |
|
90 |
with gr.Blocks(css=CSS) as demo:
|