Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -9,12 +9,12 @@ client = InferenceClient(
|
|
9 |
def format_prompt(message, history):
|
10 |
prompt = ""
|
11 |
for user_prompt, bot_response in history:
|
12 |
-
prompt += f"GPT4 Correct User: {user_prompt}<|end_of_turn|>GPT4 Correct Assistant: {
|
13 |
prompt += f"GPT4 Correct User: {message}<|end_of_turn|>GPT4 Correct Assistant:"
|
14 |
return prompt
|
15 |
|
16 |
def generate(
|
17 |
-
prompt, history, temperature=0.9, max_new_tokens=
|
18 |
):
|
19 |
temperature = float(temperature)
|
20 |
if temperature < 1e-2:
|
@@ -35,6 +35,8 @@ def generate(
|
|
35 |
output = ""
|
36 |
|
37 |
for response in stream:
|
|
|
|
|
38 |
output += response.token.text
|
39 |
yield output
|
40 |
return output
|
|
|
9 |
def format_prompt(message, history):
|
10 |
prompt = ""
|
11 |
for user_prompt, bot_response in history:
|
12 |
+
prompt += f"GPT4 Correct User: {user_prompt}<|end_of_turn|>GPT4 Correct Assistant: {bot_response}<|end_of_turn|>"
|
13 |
prompt += f"GPT4 Correct User: {message}<|end_of_turn|>GPT4 Correct Assistant:"
|
14 |
return prompt
|
15 |
|
16 |
def generate(
|
17 |
+
prompt, history, temperature=0.9, max_new_tokens=1024, top_p=0.95, repetition_penalty=1.0,
|
18 |
):
|
19 |
temperature = float(temperature)
|
20 |
if temperature < 1e-2:
|
|
|
35 |
output = ""
|
36 |
|
37 |
for response in stream:
|
38 |
+
if response.token.text=="<|end_of_turn|>":
|
39 |
+
break
|
40 |
output += response.token.text
|
41 |
yield output
|
42 |
return output
|