Spaces:
Runtime error
Runtime error
push fix
Browse files- app_paligemma.py +106 -3
app_paligemma.py
CHANGED
|
@@ -1,10 +1,113 @@
|
|
|
|
|
| 1 |
import gradio as gr
|
|
|
|
| 2 |
|
| 3 |
-
|
| 4 |
-
|
|
|
|
| 5 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
|
| 7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
if hasattr(demo, 'fns'):
|
| 9 |
for fn in demo.fns.values():
|
| 10 |
fn.api_name = False
|
|
|
|
|
|
|
|
|
| 1 |
+
from gradio_client import Client, handle_file
|
| 2 |
import gradio as gr
|
| 3 |
+
import os
|
| 4 |
|
| 5 |
+
MODELS = {
|
| 6 |
+
"Paligemma-10B": "akhaliq/paligemma2-10b-ft-docci-448"
|
| 7 |
+
}
|
| 8 |
|
| 9 |
+
def create_chat_fn(client, system_prompt, temperature, max_tokens, top_k, rep_penalty, top_p):
|
| 10 |
+
def chat(message, history):
|
| 11 |
+
text = message.get("text", "")
|
| 12 |
+
files = message.get("files", [])
|
| 13 |
+
processed_files = [handle_file(f) for f in files]
|
| 14 |
+
|
| 15 |
+
response = client.predict(
|
| 16 |
+
message={"text": text, "files": processed_files},
|
| 17 |
+
system_prompt=system_prompt,
|
| 18 |
+
temperature=temperature,
|
| 19 |
+
max_new_tokens=max_tokens,
|
| 20 |
+
top_k=top_k,
|
| 21 |
+
repetition_penalty=rep_penalty,
|
| 22 |
+
top_p=top_p,
|
| 23 |
+
api_name="/chat"
|
| 24 |
+
)
|
| 25 |
+
return response
|
| 26 |
+
return chat
|
| 27 |
|
| 28 |
+
def set_client_for_session(model_name, request: gr.Request):
|
| 29 |
+
headers = {}
|
| 30 |
+
if request and hasattr(request, 'headers'):
|
| 31 |
+
x_ip_token = request.headers.get('x-ip-token')
|
| 32 |
+
if x_ip_token:
|
| 33 |
+
headers["X-IP-Token"] = x_ip_token
|
| 34 |
+
|
| 35 |
+
return Client(MODELS[model_name], headers=headers)
|
| 36 |
+
|
| 37 |
+
def safe_chat_fn(message, history, client, system_prompt, temperature,
|
| 38 |
+
max_tokens, top_k, rep_penalty, top_p):
|
| 39 |
+
if client is None:
|
| 40 |
+
return "Error: Client not initialized. Please refresh the page."
|
| 41 |
+
try:
|
| 42 |
+
return create_chat_fn(client, system_prompt, temperature,
|
| 43 |
+
max_tokens, top_k, rep_penalty, top_p)(message, history)
|
| 44 |
+
except Exception as e:
|
| 45 |
+
print(f"Error during chat: {str(e)}")
|
| 46 |
+
return f"Error during chat: {str(e)}"
|
| 47 |
+
|
| 48 |
+
with gr.Blocks() as demo:
|
| 49 |
+
client = gr.State()
|
| 50 |
+
|
| 51 |
+
with gr.Row():
|
| 52 |
+
model_dropdown = gr.Dropdown(
|
| 53 |
+
choices=list(MODELS.keys()),
|
| 54 |
+
value="Paligemma-10B",
|
| 55 |
+
label="Select Model",
|
| 56 |
+
interactive=True
|
| 57 |
+
)
|
| 58 |
+
|
| 59 |
+
with gr.Accordion("Advanced Settings", open=False):
|
| 60 |
+
system_prompt = gr.Textbox(
|
| 61 |
+
value="You are a helpful AI assistant.",
|
| 62 |
+
label="System Prompt"
|
| 63 |
+
)
|
| 64 |
+
with gr.Row():
|
| 65 |
+
temperature = gr.Slider(
|
| 66 |
+
minimum=0.0, maximum=2.0, value=0.7,
|
| 67 |
+
label="Temperature"
|
| 68 |
+
)
|
| 69 |
+
top_p = gr.Slider(
|
| 70 |
+
minimum=0.0, maximum=1.0, value=0.95,
|
| 71 |
+
label="Top P"
|
| 72 |
+
)
|
| 73 |
+
with gr.Row():
|
| 74 |
+
top_k = gr.Slider(
|
| 75 |
+
minimum=1, maximum=100, value=40, step=1,
|
| 76 |
+
label="Top K"
|
| 77 |
+
)
|
| 78 |
+
rep_penalty = gr.Slider(
|
| 79 |
+
minimum=1.0, maximum=2.0, value=1.1,
|
| 80 |
+
label="Repetition Penalty"
|
| 81 |
+
)
|
| 82 |
+
max_tokens = gr.Slider(
|
| 83 |
+
minimum=64, maximum=4096, value=1024, step=64,
|
| 84 |
+
label="Max Tokens"
|
| 85 |
+
)
|
| 86 |
+
|
| 87 |
+
chat_interface = gr.ChatInterface(
|
| 88 |
+
fn=safe_chat_fn,
|
| 89 |
+
additional_inputs=[client, system_prompt, temperature,
|
| 90 |
+
max_tokens, top_k, rep_penalty, top_p],
|
| 91 |
+
multimodal=True
|
| 92 |
+
)
|
| 93 |
+
|
| 94 |
+
# Add model change handler
|
| 95 |
+
model_dropdown.change(
|
| 96 |
+
fn=set_client_for_session,
|
| 97 |
+
inputs=[model_dropdown],
|
| 98 |
+
outputs=[client]
|
| 99 |
+
)
|
| 100 |
+
|
| 101 |
+
# Initialize client on page load
|
| 102 |
+
demo.load(
|
| 103 |
+
fn=set_client_for_session,
|
| 104 |
+
inputs=[gr.State("Paligemma-10B")],
|
| 105 |
+
outputs=[client]
|
| 106 |
+
)
|
| 107 |
+
|
| 108 |
+
# Move the API access check here, after demo is defined
|
| 109 |
if hasattr(demo, 'fns'):
|
| 110 |
for fn in demo.fns.values():
|
| 111 |
fn.api_name = False
|
| 112 |
+
|
| 113 |
+
demo.launch()
|