Update app.py
Browse files
app.py
CHANGED
@@ -32,6 +32,7 @@ async def respond(
|
|
32 |
messages.append({"role": "user", "content": message})
|
33 |
|
34 |
response = ""
|
|
|
35 |
stream = client.chat.completions.create(
|
36 |
model="NousResearch/Hermes-3-Llama-3.1-8B",
|
37 |
max_tokens=max_tokens,
|
@@ -40,11 +41,8 @@ async def respond(
|
|
40 |
top_p=top_p,
|
41 |
messages=messages,
|
42 |
)
|
43 |
-
|
44 |
-
|
45 |
-
if message is None:
|
46 |
-
break
|
47 |
-
token = message.choices[0].text
|
48 |
response += token
|
49 |
return response
|
50 |
except APIError as e:
|
@@ -96,4 +94,4 @@ def launch_app():
|
|
96 |
print("Please try again.")
|
97 |
|
98 |
if __name__ == "__main__":
|
99 |
-
launch_app()
|
|
|
32 |
messages.append({"role": "user", "content": message})
|
33 |
|
34 |
response = ""
|
35 |
+
# Use client to stream chat completions properly
|
36 |
stream = client.chat.completions.create(
|
37 |
model="NousResearch/Hermes-3-Llama-3.1-8B",
|
38 |
max_tokens=max_tokens,
|
|
|
41 |
top_p=top_p,
|
42 |
messages=messages,
|
43 |
)
|
44 |
+
for message in stream: # Iterate over the streamed response
|
45 |
+
token = message['choices'][0]['delta'].get('content', '')
|
|
|
|
|
|
|
46 |
response += token
|
47 |
return response
|
48 |
except APIError as e:
|
|
|
94 |
print("Please try again.")
|
95 |
|
96 |
if __name__ == "__main__":
|
97 |
+
launch_app()
|