Spaces:
Running
on
Zero
Running
on
Zero
gokaygokay
commited on
Commit
•
7c0f531
1
Parent(s):
12ec307
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,200 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#import spaces
|
2 |
+
import json
|
3 |
+
import subprocess
|
4 |
+
import os
|
5 |
+
import sys
|
6 |
+
|
7 |
+
def run_command(command):
|
8 |
+
process = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE, shell=True)
|
9 |
+
output, error = process.communicate()
|
10 |
+
if process.returncode != 0:
|
11 |
+
print(f"Error executing command: {command}")
|
12 |
+
print(error.decode('utf-8'))
|
13 |
+
exit(1)
|
14 |
+
return output.decode('utf-8')
|
15 |
+
|
16 |
+
def install_packages():
|
17 |
+
# Clone the repository with submodules
|
18 |
+
run_command("git clone --recurse-submodules https://github.com/abetlen/llama-cpp-python.git")
|
19 |
+
|
20 |
+
# Change to the cloned directory
|
21 |
+
os.chdir("llama-cpp-python")
|
22 |
+
|
23 |
+
# Checkout the specific commit in the llama.cpp submodule
|
24 |
+
os.chdir("vendor/llama.cpp")
|
25 |
+
run_command("git checkout 50e0535")
|
26 |
+
os.chdir("../..")
|
27 |
+
|
28 |
+
# Upgrade pip
|
29 |
+
run_command("pip install --upgrade pip")
|
30 |
+
|
31 |
+
# Install all optional dependencies
|
32 |
+
run_command("pip install -e .[all]")
|
33 |
+
|
34 |
+
# Clear the local build cache
|
35 |
+
run_command("make clean")
|
36 |
+
|
37 |
+
# Reinstall the package
|
38 |
+
run_command("pip install -e .")
|
39 |
+
|
40 |
+
# Install llama-cpp-agent
|
41 |
+
run_command("pip install llama-cpp-agent")
|
42 |
+
|
43 |
+
print("Installation complete!")
|
44 |
+
|
45 |
+
try:
|
46 |
+
install_packages()
|
47 |
+
|
48 |
+
# If installation is successful, import the libraries
|
49 |
+
from llama_cpp import Llama
|
50 |
+
from llama_cpp_agent import LlamaCppAgent, MessagesFormatterType
|
51 |
+
from llama_cpp_agent.providers import LlamaCppPythonProvider
|
52 |
+
from llama_cpp_agent.chat_history import BasicChatHistory
|
53 |
+
from llama_cpp_agent.chat_history.messages import Roles
|
54 |
+
|
55 |
+
print("Libraries imported successfully!")
|
56 |
+
except Exception as e:
|
57 |
+
print(f"Installation failed or libraries couldn't be imported: {str(e)}")
|
58 |
+
sys.exit(1)
|
59 |
+
|
60 |
+
import gradio as gr
|
61 |
+
from huggingface_hub import hf_hub_download
|
62 |
+
|
63 |
+
hf_hub_download(
|
64 |
+
repo_id="MaziyarPanahi/Mistral-Nemo-Instruct-2407-GGUF",
|
65 |
+
filename="Mistral-Nemo-Instruct-2407.Q5_K_M.gguf",
|
66 |
+
local_dir="./models"
|
67 |
+
)
|
68 |
+
|
69 |
+
llm = None
|
70 |
+
llm_model = None
|
71 |
+
|
72 |
+
#@spaces.GPU(duration=120)
|
73 |
+
def respond(
|
74 |
+
message,
|
75 |
+
history: list[tuple[str, str]],
|
76 |
+
model,
|
77 |
+
system_message,
|
78 |
+
max_tokens,
|
79 |
+
temperature,
|
80 |
+
top_p,
|
81 |
+
top_k,
|
82 |
+
repeat_penalty,
|
83 |
+
):
|
84 |
+
|
85 |
+
|
86 |
+
chat_template = MessagesFormatterType.MISTRAL
|
87 |
+
|
88 |
+
global llm
|
89 |
+
global llm_model
|
90 |
+
|
91 |
+
if llm is None or llm_model != model:
|
92 |
+
llm = Llama(
|
93 |
+
model_path=f"models/{model}",
|
94 |
+
flash_attn=True,
|
95 |
+
n_gpu_layers=81,
|
96 |
+
n_batch=1024,
|
97 |
+
n_ctx=32768,
|
98 |
+
)
|
99 |
+
llm_model = model
|
100 |
+
|
101 |
+
provider = LlamaCppPythonProvider(llm)
|
102 |
+
|
103 |
+
agent = LlamaCppAgent(
|
104 |
+
provider,
|
105 |
+
system_prompt=f"{system_message}",
|
106 |
+
predefined_messages_formatter_type=chat_template,
|
107 |
+
debug_output=True
|
108 |
+
)
|
109 |
+
|
110 |
+
settings = provider.get_provider_default_settings()
|
111 |
+
settings.temperature = temperature
|
112 |
+
settings.top_k = top_k
|
113 |
+
settings.top_p = top_p
|
114 |
+
settings.max_tokens = max_tokens
|
115 |
+
settings.repeat_penalty = repeat_penalty
|
116 |
+
settings.stream = True
|
117 |
+
|
118 |
+
messages = BasicChatHistory()
|
119 |
+
|
120 |
+
for msn in history:
|
121 |
+
user = {
|
122 |
+
'role': Roles.user,
|
123 |
+
'content': msn[0]
|
124 |
+
}
|
125 |
+
assistant = {
|
126 |
+
'role': Roles.assistant,
|
127 |
+
'content': msn[1]
|
128 |
+
}
|
129 |
+
messages.add_message(user)
|
130 |
+
messages.add_message(assistant)
|
131 |
+
|
132 |
+
stream = agent.get_chat_response(
|
133 |
+
message,
|
134 |
+
llm_sampling_settings=settings,
|
135 |
+
chat_history=messages,
|
136 |
+
returns_streaming_generator=True,
|
137 |
+
print_output=False
|
138 |
+
)
|
139 |
+
|
140 |
+
outputs = ""
|
141 |
+
for output in stream:
|
142 |
+
outputs += output
|
143 |
+
yield outputs
|
144 |
+
|
145 |
+
description = """<p><center>
|
146 |
+
<a href="https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407" target="_blank">[Instruct Model]</a>
|
147 |
+
<a href="https://huggingface.co/mistralai/Mistral-Nemo-Base-2407" target="_blank">[Base Model]</a>
|
148 |
+
<a href="https://huggingface.co/second-state/Mistral-Nemo-Instruct-2407-GGUF" target="_blank">[GGUF Version]</a>
|
149 |
+
</center></p>
|
150 |
+
"""
|
151 |
+
|
152 |
+
demo = gr.ChatInterface(
|
153 |
+
respond,
|
154 |
+
additional_inputs=[
|
155 |
+
gr.Dropdown([
|
156 |
+
'Mistral-Nemo-Instruct-2407.Q5_K_M.gguf'
|
157 |
+
],
|
158 |
+
value="Mistral-Nemo-Instruct-2407.Q5_K_M.gguf",
|
159 |
+
label="Model"
|
160 |
+
),
|
161 |
+
gr.Textbox(value="You are a helpful assistant.", label="System message"),
|
162 |
+
gr.Slider(minimum=1, maximum=4096, value=2048, step=1, label="Max tokens"),
|
163 |
+
gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
|
164 |
+
gr.Slider(
|
165 |
+
minimum=0.1,
|
166 |
+
maximum=1.0,
|
167 |
+
value=0.95,
|
168 |
+
step=0.05,
|
169 |
+
label="Top-p",
|
170 |
+
),
|
171 |
+
gr.Slider(
|
172 |
+
minimum=0,
|
173 |
+
maximum=100,
|
174 |
+
value=40,
|
175 |
+
step=1,
|
176 |
+
label="Top-k",
|
177 |
+
),
|
178 |
+
gr.Slider(
|
179 |
+
minimum=0.0,
|
180 |
+
maximum=2.0,
|
181 |
+
value=1.1,
|
182 |
+
step=0.1,
|
183 |
+
label="Repetition penalty",
|
184 |
+
),
|
185 |
+
],
|
186 |
+
retry_btn="Retry",
|
187 |
+
undo_btn="Undo",
|
188 |
+
clear_btn="Clear",
|
189 |
+
submit_btn="Send",
|
190 |
+
title="Chat with Mistral-NeMo using llama.cpp",
|
191 |
+
description=description,
|
192 |
+
chatbot=gr.Chatbot(
|
193 |
+
scale=1,
|
194 |
+
likeable=False,
|
195 |
+
show_copy_button=True
|
196 |
+
)
|
197 |
+
)
|
198 |
+
|
199 |
+
if __name__ == "__main__":
|
200 |
+
demo.launch(debug=True)
|