Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -18,9 +18,67 @@ hf_hub_download(
|
|
18 |
|
19 |
|
20 |
@spaces.GPU(duration=120) #Is this setting the timeout?
|
21 |
-
def respond(
|
22 |
-
|
23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
|
25 |
def create_interface(model_name):
|
26 |
return gr.ChatInterface(
|
|
|
18 |
|
19 |
|
20 |
@spaces.GPU(duration=120) #Is this setting the timeout?
|
21 |
+
def respond(
|
22 |
+
message,
|
23 |
+
history: list[tuple[str, str]],
|
24 |
+
model,
|
25 |
+
system_message,
|
26 |
+
max_tokens,
|
27 |
+
temperature,
|
28 |
+
top_p,
|
29 |
+
top_k,
|
30 |
+
repeat_penalty,
|
31 |
+
):
|
32 |
+
chat_template = MessagesFormatterType.GEMMA_2
|
33 |
+
|
34 |
+
llm = Llama(
|
35 |
+
model_path=f"models/{model}",
|
36 |
+
flash_attn=True,
|
37 |
+
n_gpu_layers=81,
|
38 |
+
n_batch=1024,
|
39 |
+
n_ctx=8192,
|
40 |
+
)
|
41 |
+
|
42 |
+
provider = LlamaCppPythonProvider(llm)
|
43 |
+
|
44 |
+
# Configure sampling settings
|
45 |
+
settings = provider.get_provider_default_settings()
|
46 |
+
settings.temperature = temperature
|
47 |
+
settings.top_k = top_k
|
48 |
+
settings.top_p = top_p
|
49 |
+
settings.max_tokens = max_tokens
|
50 |
+
settings.repeat_penalty = repeat_penalty
|
51 |
+
settings.stream = True
|
52 |
+
|
53 |
+
# Prepare chat history
|
54 |
+
messages = BasicChatHistory()
|
55 |
+
|
56 |
+
for msn in history:
|
57 |
+
user_message = {'role': Roles.user, 'content': msn[0]}
|
58 |
+
assistant_message = {'role': Roles.assistant, 'content': msn[1]}
|
59 |
+
messages.add_message(user_message)
|
60 |
+
messages.add_message(assistant_message)
|
61 |
+
|
62 |
+
# Generate response using Llama model
|
63 |
+
try:
|
64 |
+
stream = llm(
|
65 |
+
prompt=message,
|
66 |
+
max_tokens=max_tokens,
|
67 |
+
temperature=temperature,
|
68 |
+
top_p=top_p,
|
69 |
+
top_k=top_k,
|
70 |
+
stop=["\n"], # Adjust stop conditions as needed
|
71 |
+
echo=False # Ensure only generated text is returned
|
72 |
+
)
|
73 |
+
|
74 |
+
outputs = ""
|
75 |
+
for output in stream:
|
76 |
+
outputs += output["choices"][0]["text"] # Extract text from response
|
77 |
+
yield outputs.strip()
|
78 |
+
|
79 |
+
except Exception as e:
|
80 |
+
yield f"Error: {str(e)}"
|
81 |
+
|
82 |
|
83 |
def create_interface(model_name):
|
84 |
return gr.ChatInterface(
|