Update app.py
Browse files
app.py
CHANGED
|
@@ -1,29 +1,18 @@
|
|
| 1 |
import openai, gradio as gr, json, plotly.graph_objects as go
|
|
|
|
| 2 |
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
Choose the single best modality for each reply:
|
| 6 |
-
|
| 7 |
-
1. Image – when an illustration, diagram, or photo-realistic scene answers best.
|
| 8 |
-
Respond **only** with JSON:
|
| 9 |
-
{"type":"image","prompt":"<detailed prompt for DALL-E-3>"}
|
| 10 |
-
|
| 11 |
-
2. Chart – when the user requests or clearly needs data-visualisation.
|
| 12 |
-
Respond **only** with JSON:
|
| 13 |
-
{"type":"chart",
|
| 14 |
-
"title":"<chart title>",
|
| 15 |
-
"data":[
|
| 16 |
-
{"x":[...], "y":[...], "label":"<series name>"},
|
| 17 |
-
...
|
| 18 |
-
]}
|
| 19 |
-
|
| 20 |
-
3. Text – for every other case.
|
| 21 |
-
Respond with plain language (no JSON).
|
| 22 |
|
| 23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 24 |
"""
|
| 25 |
|
| 26 |
-
# ---------- Helper functions ---------- #
|
| 27 |
def build_messages(history, user_msg):
|
| 28 |
messages = [{"role": "system", "content": SYSTEM_PROMPT}]
|
| 29 |
for u, a in history:
|
|
@@ -32,7 +21,6 @@ def build_messages(history, user_msg):
|
|
| 32 |
messages.append({"role": "user", "content": user_msg})
|
| 33 |
return messages
|
| 34 |
|
| 35 |
-
|
| 36 |
def multimodal_chat(api_key, user_msg, history):
|
| 37 |
if not api_key:
|
| 38 |
raise gr.Error("🔑 Please paste your OpenAI API key first.")
|
|
@@ -40,27 +28,29 @@ def multimodal_chat(api_key, user_msg, history):
|
|
| 40 |
|
| 41 |
history = history or []
|
| 42 |
messages = build_messages(history, user_msg)
|
| 43 |
-
|
| 44 |
response = openai.chat.completions.create(
|
| 45 |
-
model="gpt-4o
|
| 46 |
messages=messages,
|
| 47 |
-
temperature=0.
|
| 48 |
)
|
| 49 |
assistant_content = response.choices[0].message.content.strip()
|
| 50 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 51 |
img_url, fig = None, None
|
| 52 |
-
try:
|
| 53 |
parsed = json.loads(assistant_content)
|
| 54 |
if parsed.get("type") == "image":
|
| 55 |
dalle = openai.images.generate(
|
| 56 |
model="dall-e-3",
|
| 57 |
-
prompt=parsed.get("prompt", "high
|
| 58 |
n=1,
|
| 59 |
size="1024x1024",
|
| 60 |
)
|
| 61 |
img_url = dalle.data[0].url
|
| 62 |
history.append([user_msg, f""])
|
| 63 |
-
|
| 64 |
elif parsed.get("type") == "chart":
|
| 65 |
fig = go.Figure()
|
| 66 |
for s in parsed["data"]:
|
|
@@ -74,41 +64,35 @@ def multimodal_chat(api_key, user_msg, history):
|
|
| 74 |
)
|
| 75 |
fig.update_layout(title=parsed.get("title", "Chart"))
|
| 76 |
history.append([user_msg, parsed.get("title", "Chart below")])
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
history.append([user_msg, assistant_content])
|
| 80 |
-
|
| 81 |
-
|
| 82 |
history.append([user_msg, assistant_content])
|
| 83 |
|
| 84 |
return history, img_url, fig
|
| 85 |
|
| 86 |
-
|
| 87 |
-
# ---------- Gradio UI ---------- #
|
| 88 |
-
with gr.Blocks(title="ZEN Multimodal Assistant") as demo:
|
| 89 |
gr.Markdown(
|
| 90 |
"""
|
| 91 |
# 🧠 ZEN Multimodal Assistant
|
| 92 |
-
Paste your OpenAI API key
|
| 93 |
-
|
| 94 |
"""
|
| 95 |
)
|
| 96 |
-
|
| 97 |
api_key = gr.Textbox(label="OpenAI API Key", type="password", placeholder="sk-...")
|
| 98 |
chatbot = gr.Chatbot(label="Conversation")
|
| 99 |
with gr.Row():
|
| 100 |
user_msg = gr.Textbox(placeholder="Ask me anything…", label="Your message", scale=4)
|
| 101 |
send_btn = gr.Button("Send", variant="primary")
|
| 102 |
-
|
| 103 |
img_out = gr.Image(label="Generated image")
|
| 104 |
chart_out = gr.Plot(label="Interactive chart")
|
| 105 |
|
| 106 |
def respond(api_key, user_msg, chat_history):
|
| 107 |
chat_history, img_url, fig = multimodal_chat(api_key, user_msg, chat_history)
|
| 108 |
-
|
| 109 |
img_update = gr.update(value=img_url) if img_url else gr.update(value=None)
|
| 110 |
fig_update = gr.update(value=fig) if fig else gr.update(value=None)
|
| 111 |
-
|
| 112 |
return chat_history, img_update, fig_update
|
| 113 |
|
| 114 |
send_btn.click(
|
|
|
|
| 1 |
import openai, gradio as gr, json, plotly.graph_objects as go
|
| 2 |
+
from pathlib import Path
|
| 3 |
|
| 4 |
+
# --- Style: load from local CSS file ---
|
| 5 |
+
CUSTOM_CSS = Path("style.css").read_text()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
|
| 7 |
+
SYSTEM_PROMPT = """
|
| 8 |
+
You are ZEN Multimodal Assistant by ZEN AI Co.
|
| 9 |
+
Choose only ONE of these output modes per reply:
|
| 10 |
+
- Image: when a visual or illustration is most useful. Respond only with JSON: {"type":"image","prompt":"<prompt for DALL-E-3>"}
|
| 11 |
+
- Chart: when a user requests or needs a data visualization. Respond only with JSON: {"type":"chart","title":"<chart title>","data":[{"x":[...], "y":[...], "label":"<series name>"}]}
|
| 12 |
+
- Text: for all other situations, reply with a helpful, complete, conversational answer. Never reply with the word "text" or any label, just the response itself. Never reply in JSON unless for image or chart.
|
| 13 |
+
Never use markdown code fences, never add comments.
|
| 14 |
"""
|
| 15 |
|
|
|
|
| 16 |
def build_messages(history, user_msg):
|
| 17 |
messages = [{"role": "system", "content": SYSTEM_PROMPT}]
|
| 18 |
for u, a in history:
|
|
|
|
| 21 |
messages.append({"role": "user", "content": user_msg})
|
| 22 |
return messages
|
| 23 |
|
|
|
|
| 24 |
def multimodal_chat(api_key, user_msg, history):
|
| 25 |
if not api_key:
|
| 26 |
raise gr.Error("🔑 Please paste your OpenAI API key first.")
|
|
|
|
| 28 |
|
| 29 |
history = history or []
|
| 30 |
messages = build_messages(history, user_msg)
|
|
|
|
| 31 |
response = openai.chat.completions.create(
|
| 32 |
+
model="gpt-4o",
|
| 33 |
messages=messages,
|
| 34 |
+
temperature=0.6,
|
| 35 |
)
|
| 36 |
assistant_content = response.choices[0].message.content.strip()
|
| 37 |
|
| 38 |
+
# Defensive: Never show the word "text" alone
|
| 39 |
+
if assistant_content.lower() == "text":
|
| 40 |
+
assistant_content = "(I'm sorry, I didn't understand. Could you rephrase?)"
|
| 41 |
+
|
| 42 |
img_url, fig = None, None
|
| 43 |
+
try: # Try to parse as JSON for image or chart
|
| 44 |
parsed = json.loads(assistant_content)
|
| 45 |
if parsed.get("type") == "image":
|
| 46 |
dalle = openai.images.generate(
|
| 47 |
model="dall-e-3",
|
| 48 |
+
prompt=parsed.get("prompt", "high quality illustration, cinematic, best quality"),
|
| 49 |
n=1,
|
| 50 |
size="1024x1024",
|
| 51 |
)
|
| 52 |
img_url = dalle.data[0].url
|
| 53 |
history.append([user_msg, f""])
|
|
|
|
| 54 |
elif parsed.get("type") == "chart":
|
| 55 |
fig = go.Figure()
|
| 56 |
for s in parsed["data"]:
|
|
|
|
| 64 |
)
|
| 65 |
fig.update_layout(title=parsed.get("title", "Chart"))
|
| 66 |
history.append([user_msg, parsed.get("title", "Chart below")])
|
| 67 |
+
else:
|
| 68 |
+
# If unexpected JSON, fallback to text
|
| 69 |
+
history.append([user_msg, str(assistant_content)])
|
| 70 |
+
except (json.JSONDecodeError, KeyError, TypeError):
|
| 71 |
+
# If not JSON, treat as text
|
| 72 |
history.append([user_msg, assistant_content])
|
| 73 |
|
| 74 |
return history, img_url, fig
|
| 75 |
|
| 76 |
+
with gr.Blocks(title="ZEN Multimodal Assistant", css=CUSTOM_CSS) as demo:
|
|
|
|
|
|
|
| 77 |
gr.Markdown(
|
| 78 |
"""
|
| 79 |
# 🧠 ZEN Multimodal Assistant
|
| 80 |
+
Paste your OpenAI API key (never saved).
|
| 81 |
+
This assistant intelligently responds with **text, an image, or an interactive chart**.
|
| 82 |
"""
|
| 83 |
)
|
|
|
|
| 84 |
api_key = gr.Textbox(label="OpenAI API Key", type="password", placeholder="sk-...")
|
| 85 |
chatbot = gr.Chatbot(label="Conversation")
|
| 86 |
with gr.Row():
|
| 87 |
user_msg = gr.Textbox(placeholder="Ask me anything…", label="Your message", scale=4)
|
| 88 |
send_btn = gr.Button("Send", variant="primary")
|
|
|
|
| 89 |
img_out = gr.Image(label="Generated image")
|
| 90 |
chart_out = gr.Plot(label="Interactive chart")
|
| 91 |
|
| 92 |
def respond(api_key, user_msg, chat_history):
|
| 93 |
chat_history, img_url, fig = multimodal_chat(api_key, user_msg, chat_history)
|
|
|
|
| 94 |
img_update = gr.update(value=img_url) if img_url else gr.update(value=None)
|
| 95 |
fig_update = gr.update(value=fig) if fig else gr.update(value=None)
|
|
|
|
| 96 |
return chat_history, img_update, fig_update
|
| 97 |
|
| 98 |
send_btn.click(
|