Spaces:
Sleeping
Sleeping
Commit
•
3462a26
1
Parent(s):
4bf5f6e
Upload 2 files (#1)
Browse files- Upload 2 files (6200ce9dc854dd613088977d179f0fd372a3bddd)
Co-authored-by: John Smith <John6666@users.noreply.huggingface.co>
- app.py +182 -182
- requirements.txt +5 -5
app.py
CHANGED
@@ -1,182 +1,182 @@
|
|
1 |
-
import spaces
|
2 |
-
import subprocess
|
3 |
-
from llama_cpp import Llama
|
4 |
-
from llama_cpp_agent import LlamaCppAgent, MessagesFormatterType
|
5 |
-
from llama_cpp_agent.providers import LlamaCppPythonProvider
|
6 |
-
from llama_cpp_agent.chat_history import BasicChatHistory
|
7 |
-
from llama_cpp_agent.chat_history.messages import Roles
|
8 |
-
import gradio as gr
|
9 |
-
from huggingface_hub import hf_hub_download
|
10 |
-
import os
|
11 |
-
import cv2
|
12 |
-
|
13 |
-
huggingface_token = os.environ
|
14 |
-
|
15 |
-
# Download the Meta-Llama-3.1-8B-Instruct model
|
16 |
-
hf_hub_download(
|
17 |
-
repo_id="bartowski/Meta-Llama-3.1-8B-Instruct-GGUF",
|
18 |
-
filename="Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf",
|
19 |
-
local_dir="./models",
|
20 |
-
token=huggingface_token
|
21 |
-
)
|
22 |
-
|
23 |
-
hf_hub_download(
|
24 |
-
repo_id="bartowski/Mistral-Nemo-Instruct-2407-GGUF",
|
25 |
-
filename="Mistral-Nemo-Instruct-2407-Q5_K_M.gguf",
|
26 |
-
local_dir="./models",
|
27 |
-
token=huggingface_token
|
28 |
-
)
|
29 |
-
|
30 |
-
hf_hub_download(
|
31 |
-
repo_id="bartowski/gemma-2-2b-it-GGUF",
|
32 |
-
filename="gemma-2-2b-it-Q6_K_L.gguf",
|
33 |
-
local_dir="./models",
|
34 |
-
token=huggingface_token
|
35 |
-
)
|
36 |
-
|
37 |
-
hf_hub_download(
|
38 |
-
repo_id="bartowski/openchat-3.6-8b-20240522-GGUF",
|
39 |
-
filename="openchat-3.6-8b-20240522-Q6_K.gguf",
|
40 |
-
local_dir="./models",
|
41 |
-
token=huggingface_token
|
42 |
-
)
|
43 |
-
|
44 |
-
hf_hub_download(
|
45 |
-
repo_id="bartowski/Llama-3-Groq-8B-Tool-Use-GGUF",
|
46 |
-
filename="Llama-3-Groq-8B-Tool-Use-Q6_K.gguf",
|
47 |
-
local_dir="./models",
|
48 |
-
token=huggingface_token
|
49 |
-
)
|
50 |
-
|
51 |
-
|
52 |
-
llm = None
|
53 |
-
llm_model = None
|
54 |
-
|
55 |
-
cv2.setNumThreads(1)
|
56 |
-
|
57 |
-
@spaces.GPU()
|
58 |
-
def respond(
|
59 |
-
message,
|
60 |
-
history: list[tuple[str, str]],
|
61 |
-
model,
|
62 |
-
system_message,
|
63 |
-
max_tokens,
|
64 |
-
temperature,
|
65 |
-
top_p,
|
66 |
-
top_k,
|
67 |
-
repeat_penalty,
|
68 |
-
):
|
69 |
-
chat_template = MessagesFormatterType.GEMMA_2
|
70 |
-
|
71 |
-
global llm
|
72 |
-
global llm_model
|
73 |
-
|
74 |
-
# Load model only if it's not already loaded or if a new model is selected
|
75 |
-
if llm is None or llm_model != model:
|
76 |
-
try:
|
77 |
-
llm = Llama(
|
78 |
-
model_path=f"models/{model}",
|
79 |
-
flash_attn=True,
|
80 |
-
n_gpu_layers=81, # Adjust based on available GPU resources
|
81 |
-
n_batch=1024,
|
82 |
-
n_ctx=8192,
|
83 |
-
)
|
84 |
-
llm_model = model
|
85 |
-
except Exception as e:
|
86 |
-
return f"Error loading model: {str(e)}"
|
87 |
-
|
88 |
-
provider = LlamaCppPythonProvider(llm)
|
89 |
-
|
90 |
-
agent = LlamaCppAgent(
|
91 |
-
provider,
|
92 |
-
system_prompt=f"{system_message}",
|
93 |
-
predefined_messages_formatter_type=chat_template,
|
94 |
-
debug_output=True
|
95 |
-
)
|
96 |
-
|
97 |
-
settings = provider.get_provider_default_settings()
|
98 |
-
settings.temperature = temperature
|
99 |
-
settings.top_k = top_k
|
100 |
-
settings.top_p = top_p
|
101 |
-
settings.max_tokens = max_tokens
|
102 |
-
settings.repeat_penalty = repeat_penalty
|
103 |
-
settings.stream = True
|
104 |
-
|
105 |
-
messages = BasicChatHistory()
|
106 |
-
|
107 |
-
# Add user and assistant messages to the history
|
108 |
-
for msn in history:
|
109 |
-
user = {'role': Roles.user, 'content': msn[0]}
|
110 |
-
assistant = {'role': Roles.assistant, 'content': msn[1]}
|
111 |
-
messages.add_message(user)
|
112 |
-
messages.add_message(assistant)
|
113 |
-
|
114 |
-
# Stream the response
|
115 |
-
try:
|
116 |
-
stream = agent.get_chat_response(
|
117 |
-
message,
|
118 |
-
llm_sampling_settings=settings,
|
119 |
-
chat_history=messages,
|
120 |
-
returns_streaming_generator=True,
|
121 |
-
print_output=False
|
122 |
-
)
|
123 |
-
|
124 |
-
outputs = ""
|
125 |
-
for output in stream:
|
126 |
-
outputs += output
|
127 |
-
yield outputs
|
128 |
-
except Exception as e:
|
129 |
-
yield f"Error during response generation: {str(e)}"
|
130 |
-
|
131 |
-
demo = gr.ChatInterface(
|
132 |
-
respond,
|
133 |
-
additional_inputs=[
|
134 |
-
gr.Dropdown([
|
135 |
-
'Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf',
|
136 |
-
'Mistral-Nemo-Instruct-2407-Q5_K_M.gguf',
|
137 |
-
'gemma-2-2b-it-Q6_K_L.gguf',
|
138 |
-
'openchat-3.6-8b-20240522-Q6_K.gguf',
|
139 |
-
'Llama-3-Groq-8B-Tool-Use-Q6_K.gguf'
|
140 |
-
],
|
141 |
-
value="gemma-2-2b-it-Q6_K_L.gguf",
|
142 |
-
label="Model"
|
143 |
-
),
|
144 |
-
gr.Textbox(value="You are a helpful assistant.", label="System message"),
|
145 |
-
gr.Slider(minimum=1, maximum=4096, value=2048, step=1, label="Max tokens"),
|
146 |
-
gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
|
147 |
-
gr.Slider(
|
148 |
-
minimum=0.1,
|
149 |
-
maximum=1.0,
|
150 |
-
value=0.95,
|
151 |
-
step=0.05,
|
152 |
-
label="Top-p",
|
153 |
-
),
|
154 |
-
gr.Slider(
|
155 |
-
minimum=0,
|
156 |
-
maximum=100,
|
157 |
-
value=40,
|
158 |
-
step=1,
|
159 |
-
label="Top-k",
|
160 |
-
),
|
161 |
-
gr.Slider(
|
162 |
-
minimum=0.0,
|
163 |
-
maximum=2.0,
|
164 |
-
value=1.1,
|
165 |
-
step=0.1,
|
166 |
-
label="Repetition penalty",
|
167 |
-
),
|
168 |
-
],
|
169 |
-
retry_btn="Retry",
|
170 |
-
undo_btn="Undo",
|
171 |
-
clear_btn="Clear",
|
172 |
-
submit_btn="Send",
|
173 |
-
title="Chat with lots of Models and LLMs using llama.cpp",
|
174 |
-
chatbot=gr.Chatbot(
|
175 |
-
scale=1,
|
176 |
-
likeable=False,
|
177 |
-
show_copy_button=True
|
178 |
-
)
|
179 |
-
)
|
180 |
-
|
181 |
-
if __name__ == "__main__":
|
182 |
-
demo.launch()
|
|
|
1 |
+
import spaces
|
2 |
+
import subprocess
|
3 |
+
from llama_cpp import Llama
|
4 |
+
from llama_cpp_agent import LlamaCppAgent, MessagesFormatterType
|
5 |
+
from llama_cpp_agent.providers import LlamaCppPythonProvider
|
6 |
+
from llama_cpp_agent.chat_history import BasicChatHistory
|
7 |
+
from llama_cpp_agent.chat_history.messages import Roles
|
8 |
+
import gradio as gr
|
9 |
+
from huggingface_hub import hf_hub_download
|
10 |
+
import os
|
11 |
+
import cv2
|
12 |
+
|
13 |
+
huggingface_token = os.environ.get('HF_TOKEN')
|
14 |
+
|
15 |
+
# Download the Meta-Llama-3.1-8B-Instruct model
|
16 |
+
hf_hub_download(
|
17 |
+
repo_id="bartowski/Meta-Llama-3.1-8B-Instruct-GGUF",
|
18 |
+
filename="Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf",
|
19 |
+
local_dir="./models",
|
20 |
+
token=huggingface_token
|
21 |
+
)
|
22 |
+
|
23 |
+
hf_hub_download(
|
24 |
+
repo_id="bartowski/Mistral-Nemo-Instruct-2407-GGUF",
|
25 |
+
filename="Mistral-Nemo-Instruct-2407-Q5_K_M.gguf",
|
26 |
+
local_dir="./models",
|
27 |
+
token=huggingface_token
|
28 |
+
)
|
29 |
+
|
30 |
+
hf_hub_download(
|
31 |
+
repo_id="bartowski/gemma-2-2b-it-GGUF",
|
32 |
+
filename="gemma-2-2b-it-Q6_K_L.gguf",
|
33 |
+
local_dir="./models",
|
34 |
+
token=huggingface_token
|
35 |
+
)
|
36 |
+
|
37 |
+
hf_hub_download(
|
38 |
+
repo_id="bartowski/openchat-3.6-8b-20240522-GGUF",
|
39 |
+
filename="openchat-3.6-8b-20240522-Q6_K.gguf",
|
40 |
+
local_dir="./models",
|
41 |
+
token=huggingface_token
|
42 |
+
)
|
43 |
+
|
44 |
+
hf_hub_download(
|
45 |
+
repo_id="bartowski/Llama-3-Groq-8B-Tool-Use-GGUF",
|
46 |
+
filename="Llama-3-Groq-8B-Tool-Use-Q6_K.gguf",
|
47 |
+
local_dir="./models",
|
48 |
+
token=huggingface_token
|
49 |
+
)
|
50 |
+
|
51 |
+
|
52 |
+
llm = None
|
53 |
+
llm_model = None
|
54 |
+
|
55 |
+
cv2.setNumThreads(1)
|
56 |
+
|
57 |
+
@spaces.GPU()
|
58 |
+
def respond(
|
59 |
+
message,
|
60 |
+
history: list[tuple[str, str]],
|
61 |
+
model,
|
62 |
+
system_message,
|
63 |
+
max_tokens,
|
64 |
+
temperature,
|
65 |
+
top_p,
|
66 |
+
top_k,
|
67 |
+
repeat_penalty,
|
68 |
+
):
|
69 |
+
chat_template = MessagesFormatterType.GEMMA_2
|
70 |
+
|
71 |
+
global llm
|
72 |
+
global llm_model
|
73 |
+
|
74 |
+
# Load model only if it's not already loaded or if a new model is selected
|
75 |
+
if llm is None or llm_model != model:
|
76 |
+
try:
|
77 |
+
llm = Llama(
|
78 |
+
model_path=f"models/{model}",
|
79 |
+
flash_attn=True,
|
80 |
+
n_gpu_layers=81, # Adjust based on available GPU resources
|
81 |
+
n_batch=1024,
|
82 |
+
n_ctx=8192,
|
83 |
+
)
|
84 |
+
llm_model = model
|
85 |
+
except Exception as e:
|
86 |
+
return f"Error loading model: {str(e)}"
|
87 |
+
|
88 |
+
provider = LlamaCppPythonProvider(llm)
|
89 |
+
|
90 |
+
agent = LlamaCppAgent(
|
91 |
+
provider,
|
92 |
+
system_prompt=f"{system_message}",
|
93 |
+
predefined_messages_formatter_type=chat_template,
|
94 |
+
debug_output=True
|
95 |
+
)
|
96 |
+
|
97 |
+
settings = provider.get_provider_default_settings()
|
98 |
+
settings.temperature = temperature
|
99 |
+
settings.top_k = top_k
|
100 |
+
settings.top_p = top_p
|
101 |
+
settings.max_tokens = max_tokens
|
102 |
+
settings.repeat_penalty = repeat_penalty
|
103 |
+
settings.stream = True
|
104 |
+
|
105 |
+
messages = BasicChatHistory()
|
106 |
+
|
107 |
+
# Add user and assistant messages to the history
|
108 |
+
for msn in history:
|
109 |
+
user = {'role': Roles.user, 'content': msn[0]}
|
110 |
+
assistant = {'role': Roles.assistant, 'content': msn[1]}
|
111 |
+
messages.add_message(user)
|
112 |
+
messages.add_message(assistant)
|
113 |
+
|
114 |
+
# Stream the response
|
115 |
+
try:
|
116 |
+
stream = agent.get_chat_response(
|
117 |
+
message,
|
118 |
+
llm_sampling_settings=settings,
|
119 |
+
chat_history=messages,
|
120 |
+
returns_streaming_generator=True,
|
121 |
+
print_output=False
|
122 |
+
)
|
123 |
+
|
124 |
+
outputs = ""
|
125 |
+
for output in stream:
|
126 |
+
outputs += output
|
127 |
+
yield outputs
|
128 |
+
except Exception as e:
|
129 |
+
yield f"Error during response generation: {str(e)}"
|
130 |
+
|
131 |
+
demo = gr.ChatInterface(
|
132 |
+
fn=respond,
|
133 |
+
additional_inputs=[
|
134 |
+
gr.Dropdown([
|
135 |
+
'Meta-Llama-3.1-8B-Instruct-Q5_K_M.gguf',
|
136 |
+
'Mistral-Nemo-Instruct-2407-Q5_K_M.gguf',
|
137 |
+
'gemma-2-2b-it-Q6_K_L.gguf',
|
138 |
+
'openchat-3.6-8b-20240522-Q6_K.gguf',
|
139 |
+
'Llama-3-Groq-8B-Tool-Use-Q6_K.gguf'
|
140 |
+
],
|
141 |
+
value="gemma-2-2b-it-Q6_K_L.gguf",
|
142 |
+
label="Model"
|
143 |
+
),
|
144 |
+
gr.Textbox(value="You are a helpful assistant.", label="System message"),
|
145 |
+
gr.Slider(minimum=1, maximum=4096, value=2048, step=1, label="Max tokens"),
|
146 |
+
gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
|
147 |
+
gr.Slider(
|
148 |
+
minimum=0.1,
|
149 |
+
maximum=1.0,
|
150 |
+
value=0.95,
|
151 |
+
step=0.05,
|
152 |
+
label="Top-p",
|
153 |
+
),
|
154 |
+
gr.Slider(
|
155 |
+
minimum=0,
|
156 |
+
maximum=100,
|
157 |
+
value=40,
|
158 |
+
step=1,
|
159 |
+
label="Top-k",
|
160 |
+
),
|
161 |
+
gr.Slider(
|
162 |
+
minimum=0.0,
|
163 |
+
maximum=2.0,
|
164 |
+
value=1.1,
|
165 |
+
step=0.1,
|
166 |
+
label="Repetition penalty",
|
167 |
+
),
|
168 |
+
],
|
169 |
+
retry_btn="Retry",
|
170 |
+
undo_btn="Undo",
|
171 |
+
clear_btn="Clear",
|
172 |
+
submit_btn="Send",
|
173 |
+
title="Chat with lots of Models and LLMs using llama.cpp",
|
174 |
+
chatbot=gr.Chatbot(
|
175 |
+
scale=1,
|
176 |
+
likeable=False,
|
177 |
+
show_copy_button=True
|
178 |
+
)
|
179 |
+
)
|
180 |
+
|
181 |
+
if __name__ == "__main__":
|
182 |
+
demo.launch()
|
requirements.txt
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
|
4 |
-
llama-cpp-
|
5 |
-
|
6 |
opencv-python
|
|
|
1 |
+
spaces
|
2 |
+
huggingface_hub
|
3 |
+
scikit-build-core
|
4 |
+
https://github.com/abetlen/llama-cpp-python/releases/download/v0.2.90-cu124/llama_cpp_python-0.2.90-cp310-cp310-linux_x86_64.whl
|
5 |
+
git+https://github.com/Maximilian-Winter/llama-cpp-agent
|
6 |
opencv-python
|