Spaces:
Runtime error
Runtime error
Commit
•
fd397ae
1
Parent(s):
4808ae9
Updating to api url
Browse files- app.py +47 -102
- requirements.txt +2 -1
app.py
CHANGED
@@ -5,125 +5,69 @@ from typing import Iterator
|
|
5 |
import gradio as gr
|
6 |
import spaces
|
7 |
import torch
|
8 |
-
from
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
MAX_MAX_NEW_TOKENS = 2048
|
11 |
-
DEFAULT_MAX_NEW_TOKENS =
|
12 |
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
13 |
|
14 |
DESCRIPTION = """
|
15 |
Llama3-TenyxChat-70B is part of the TenyxChat series, models trained to function as useful assistants.
|
16 |
The model is obtained via direct preference tuning using Tenyx's fine-tuning technology. Model details available at our model page.
|
17 |
-
|
18 |
-
**The model is currently loaded in 8-bit**.
|
19 |
"""
|
20 |
|
21 |
|
22 |
LICENSE = """
|
23 |
This demo is governed by the license available [here.](https://huggingface.co/spaces/tenyx/Llama3-TenyxChat-70B/blob/main/LICENSE.txt)"""
|
24 |
|
25 |
-
|
26 |
-
if not torch.cuda.is_available():
|
27 |
-
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
28 |
-
|
29 |
-
|
30 |
-
if torch.cuda.is_available():
|
31 |
-
model_id = "tenyx/Llama3-TenyxChat-70B"
|
32 |
-
# model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", torch_dtype=torch.bfloat16)
|
33 |
-
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_in_4bit=True)
|
34 |
-
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
35 |
-
tokenizer.use_default_system_prompt = False
|
36 |
-
|
37 |
-
|
38 |
@spaces.GPU
|
39 |
def generate(
|
40 |
message: str,
|
41 |
chat_history: list[tuple[str, str]],
|
42 |
-
system_prompt: str,
|
43 |
-
max_new_tokens: int = 1024,
|
44 |
-
temperature: float = 0.6,
|
45 |
-
top_p: float = 0.9,
|
46 |
-
top_k: int = 50,
|
47 |
-
repetition_penalty: float = 1.2,
|
48 |
) -> Iterator[str]:
|
49 |
conversation = [{"role": "system", "content": "You are a helpful assistant developed by Tenyx, a conversational voice AI company."}]
|
50 |
-
|
51 |
-
conversation.append({"role": "system", "content": system_prompt})
|
52 |
for user, assistant in chat_history:
|
53 |
conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
|
54 |
conversation.append({"role": "user", "content": message})
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
#
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
)
|
78 |
-
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
79 |
-
t.start()
|
80 |
-
|
81 |
-
outputs = []
|
82 |
-
for text in streamer:
|
83 |
-
outputs.append(text)
|
84 |
-
yield "".join(outputs)
|
85 |
-
|
86 |
-
|
87 |
-
chat_interface = gr.ChatInterface(
|
88 |
fn=generate,
|
89 |
-
additional_inputs=[
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
label="Temperature",
|
100 |
-
minimum=0.1,
|
101 |
-
maximum=4.0,
|
102 |
-
step=0.1,
|
103 |
-
value=0.1,
|
104 |
-
),
|
105 |
-
gr.Slider(
|
106 |
-
label="Top-p (nucleus sampling)",
|
107 |
-
minimum=0.05,
|
108 |
-
maximum=1.0,
|
109 |
-
step=0.05,
|
110 |
-
value=0.9,
|
111 |
-
),
|
112 |
-
gr.Slider(
|
113 |
-
label="Top-k",
|
114 |
-
minimum=1,
|
115 |
-
maximum=1000,
|
116 |
-
step=1,
|
117 |
-
value=50,
|
118 |
-
),
|
119 |
-
gr.Slider(
|
120 |
-
label="Repetition penalty",
|
121 |
-
minimum=1.0,
|
122 |
-
maximum=2.0,
|
123 |
-
step=0.05,
|
124 |
-
value=1.2,
|
125 |
-
),
|
126 |
-
],
|
127 |
stop_btn=None,
|
128 |
examples=[
|
129 |
["Hello there! How are you doing?"],
|
@@ -134,13 +78,14 @@ chat_interface = gr.ChatInterface(
|
|
134 |
],
|
135 |
)
|
136 |
|
137 |
-
with gr.Blocks(
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
|
|
142 |
|
143 |
if __name__ == "__main__":
|
144 |
-
demo.queue(max_size=4).launch()
|
145 |
|
146 |
|
|
|
5 |
import gradio as gr
|
6 |
import spaces
|
7 |
import torch
|
8 |
+
from openai import OpenAI, APIError
|
9 |
+
|
10 |
+
client = OpenAI(
|
11 |
+
base_url="https://hjopms3xd7gembdu.us-east-1.aws.endpoints.huggingface.cloud/v1/",
|
12 |
+
api_key="hf_XXXXX"
|
13 |
+
)
|
14 |
|
15 |
MAX_MAX_NEW_TOKENS = 2048
|
16 |
+
DEFAULT_MAX_NEW_TOKENS = 512
|
17 |
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
18 |
|
19 |
DESCRIPTION = """
|
20 |
Llama3-TenyxChat-70B is part of the TenyxChat series, models trained to function as useful assistants.
|
21 |
The model is obtained via direct preference tuning using Tenyx's fine-tuning technology. Model details available at our model page.
|
|
|
|
|
22 |
"""
|
23 |
|
24 |
|
25 |
LICENSE = """
|
26 |
This demo is governed by the license available [here.](https://huggingface.co/spaces/tenyx/Llama3-TenyxChat-70B/blob/main/LICENSE.txt)"""
|
27 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
@spaces.GPU
|
29 |
def generate(
|
30 |
message: str,
|
31 |
chat_history: list[tuple[str, str]],
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
) -> Iterator[str]:
|
33 |
conversation = [{"role": "system", "content": "You are a helpful assistant developed by Tenyx, a conversational voice AI company."}]
|
34 |
+
|
|
|
35 |
for user, assistant in chat_history:
|
36 |
conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
|
37 |
conversation.append({"role": "user", "content": message})
|
38 |
+
|
39 |
+
try:
|
40 |
+
response = client.chat.completions.create(
|
41 |
+
model="tgi",
|
42 |
+
messages=conversation,
|
43 |
+
stop=["<|end_of_text|>", "<|eot_id|>"],
|
44 |
+
stream=True,
|
45 |
+
max_tokens=1024,
|
46 |
+
)
|
47 |
+
outputs = []
|
48 |
+
for chunk in response:
|
49 |
+
outputs.append(chunk.choices[0].delta.content)
|
50 |
+
yield "".join(outputs)
|
51 |
+
|
52 |
+
except APIError as e:
|
53 |
+
# Handle API errors or network errors here
|
54 |
+
print(f"Error: {e}")
|
55 |
+
yield "An error occurred. Please try again later."
|
56 |
+
|
57 |
+
|
58 |
+
|
59 |
+
demo = gr.ChatInterface(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
60 |
fn=generate,
|
61 |
+
# additional_inputs=[
|
62 |
+
# gr.Textbox(label="System prompt", lines=6),
|
63 |
+
# gr.Slider(
|
64 |
+
# label="Max new tokens",
|
65 |
+
# minimum=1,
|
66 |
+
# maximum=MAX_MAX_NEW_TOKENS,
|
67 |
+
# step=1,
|
68 |
+
# value=DEFAULT_MAX_NEW_TOKENS,
|
69 |
+
# ),
|
70 |
+
# ],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
71 |
stop_btn=None,
|
72 |
examples=[
|
73 |
["Hello there! How are you doing?"],
|
|
|
78 |
],
|
79 |
)
|
80 |
|
81 |
+
# with gr.Blocks() as demo:
|
82 |
+
# # gr.Markdown(DESCRIPTION)
|
83 |
+
# # gr.Markdown(LICENSE)
|
84 |
+
# # gr.DuplicateButton(value="Duplicate Space for private use", elem_id="duplicate-button")
|
85 |
+
# chat_interface.render()
|
86 |
+
|
87 |
|
88 |
if __name__ == "__main__":
|
89 |
+
demo.queue(max_size=4).launch(share=True)
|
90 |
|
91 |
|
requirements.txt
CHANGED
@@ -6,4 +6,5 @@ spaces==0.26.2
|
|
6 |
scipy
|
7 |
sentencepiece
|
8 |
torch==2.3.0
|
9 |
-
transformers==4.40.1
|
|
|
|
6 |
scipy
|
7 |
sentencepiece
|
8 |
torch==2.3.0
|
9 |
+
transformers==4.40.1
|
10 |
+
openai==1.25.0
|