Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -14,7 +14,7 @@ LICENSE = """
|
|
14 |
|
15 |
"""
|
16 |
|
17 |
-
DEFAULT_SYSTEM_PROMPT = ""
|
18 |
|
19 |
API_URL = os.environ.get("API_URL")
|
20 |
TOKEN = os.environ.get("TOKEN")
|
@@ -25,19 +25,9 @@ HEADERS = {
|
|
25 |
}
|
26 |
|
27 |
MODEL_NAME="breeze-7b-instruct-v01"
|
28 |
-
TEMPERATURE=1
|
29 |
-
MAX_TOKENS=16
|
30 |
-
TOP_P=0
|
31 |
PRESENCE_PENALTY=0
|
32 |
FREQUENCY_PENALTY=0
|
33 |
|
34 |
-
|
35 |
-
eos_token = "</s>"
|
36 |
-
MAX_MAX_NEW_TOKENS = 4096
|
37 |
-
DEFAULT_MAX_NEW_TOKENS = 1536
|
38 |
-
|
39 |
-
max_prompt_length = 8192 - MAX_MAX_NEW_TOKENS - 10
|
40 |
-
|
41 |
model_name = "MediaTek-Research/Breeze-7B-Instruct-v0.1"
|
42 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
43 |
|
@@ -77,10 +67,10 @@ with gr.Blocks() as demo:
|
|
77 |
)
|
78 |
temperature = gr.Slider(
|
79 |
label='Temperature',
|
80 |
-
minimum=0.
|
81 |
-
maximum=
|
82 |
step=0.1,
|
83 |
-
value=0.
|
84 |
)
|
85 |
top_p = gr.Slider(
|
86 |
label='Top-p (nucleus sampling)',
|
@@ -89,19 +79,13 @@ with gr.Blocks() as demo:
|
|
89 |
step=0.05,
|
90 |
value=0.95,
|
91 |
)
|
92 |
-
|
93 |
-
label='Top-k',
|
94 |
-
minimum=1,
|
95 |
-
maximum=1000,
|
96 |
-
step=1,
|
97 |
-
value=50,
|
98 |
-
)
|
99 |
|
100 |
def user(user_message, history):
|
101 |
return "", history + [[user_message, None]]
|
102 |
|
103 |
|
104 |
-
def bot(history, max_new_tokens, temperature, top_p,
|
105 |
chat_data = []
|
106 |
for user_msg, assistant_msg in history:
|
107 |
if user_msg is not None:
|
@@ -117,7 +101,7 @@ with gr.Blocks() as demo:
|
|
117 |
data = {
|
118 |
"model": MODEL_NAME,
|
119 |
"prompt": str(message),
|
120 |
-
"temperature": float(temperature),
|
121 |
"n": 1,
|
122 |
"max_tokens": int(max_new_tokens),
|
123 |
"stop": "",
|
@@ -126,18 +110,23 @@ with gr.Blocks() as demo:
|
|
126 |
"echo": False,
|
127 |
"presence_penalty": PRESENCE_PENALTY,
|
128 |
"frequency_penalty": FREQUENCY_PENALTY,
|
|
|
129 |
}
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
|
|
|
|
|
|
|
|
141 |
|
142 |
msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
|
143 |
fn=bot,
|
@@ -146,7 +135,6 @@ with gr.Blocks() as demo:
|
|
146 |
max_new_tokens,
|
147 |
temperature,
|
148 |
top_p,
|
149 |
-
top_k,
|
150 |
system_prompt,
|
151 |
],
|
152 |
outputs=chatbot
|
@@ -160,7 +148,6 @@ with gr.Blocks() as demo:
|
|
160 |
max_new_tokens,
|
161 |
temperature,
|
162 |
top_p,
|
163 |
-
top_k,
|
164 |
system_prompt,
|
165 |
],
|
166 |
outputs=chatbot
|
|
|
14 |
|
15 |
"""
|
16 |
|
17 |
+
DEFAULT_SYSTEM_PROMPT = "You are a helpful AI assistant built by MediaTek Research. The user you are helping speaks Traditional Chinese and comes from Taiwan."
|
18 |
|
19 |
API_URL = os.environ.get("API_URL")
|
20 |
TOKEN = os.environ.get("TOKEN")
|
|
|
25 |
}
|
26 |
|
27 |
MODEL_NAME="breeze-7b-instruct-v01"
|
|
|
|
|
|
|
28 |
PRESENCE_PENALTY=0
|
29 |
FREQUENCY_PENALTY=0
|
30 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
model_name = "MediaTek-Research/Breeze-7B-Instruct-v0.1"
|
32 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
33 |
|
|
|
67 |
)
|
68 |
temperature = gr.Slider(
|
69 |
label='Temperature',
|
70 |
+
minimum=0.0,
|
71 |
+
maximum=0.5,
|
72 |
step=0.1,
|
73 |
+
value=0.0,
|
74 |
)
|
75 |
top_p = gr.Slider(
|
76 |
label='Top-p (nucleus sampling)',
|
|
|
79 |
step=0.05,
|
80 |
value=0.95,
|
81 |
)
|
82 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
83 |
|
84 |
def user(user_message, history):
|
85 |
return "", history + [[user_message, None]]
|
86 |
|
87 |
|
88 |
+
def bot(history, max_new_tokens, temperature, top_p, system_prompt):
|
89 |
chat_data = []
|
90 |
for user_msg, assistant_msg in history:
|
91 |
if user_msg is not None:
|
|
|
101 |
data = {
|
102 |
"model": MODEL_NAME,
|
103 |
"prompt": str(message),
|
104 |
+
"temperature": float(temperature) + 0.0001,
|
105 |
"n": 1,
|
106 |
"max_tokens": int(max_new_tokens),
|
107 |
"stop": "",
|
|
|
110 |
"echo": False,
|
111 |
"presence_penalty": PRESENCE_PENALTY,
|
112 |
"frequency_penalty": FREQUENCY_PENALTY,
|
113 |
+
"stream": True,
|
114 |
}
|
115 |
+
|
116 |
+
with requests.post(url, headers=HEADERS, data=json.dumps(data), stream=True) as r:
|
117 |
+
for response in r.iter_lines():
|
118 |
+
if len(response) > 0:
|
119 |
+
text = response.decode()
|
120 |
+
if text != "data: [DONE]":
|
121 |
+
if text.startswith("data: "):
|
122 |
+
text = text[5:]
|
123 |
+
delta = json.loads(text)["choices"][0]["text"]
|
124 |
+
|
125 |
+
if history[-1][1] is None:
|
126 |
+
history[-1][1] = delta
|
127 |
+
else:
|
128 |
+
history[-1][1] += delta
|
129 |
+
yield history
|
130 |
|
131 |
msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
|
132 |
fn=bot,
|
|
|
135 |
max_new_tokens,
|
136 |
temperature,
|
137 |
top_p,
|
|
|
138 |
system_prompt,
|
139 |
],
|
140 |
outputs=chatbot
|
|
|
148 |
max_new_tokens,
|
149 |
temperature,
|
150 |
top_p,
|
|
|
151 |
system_prompt,
|
152 |
],
|
153 |
outputs=chatbot
|