Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,73 +1,79 @@
|
|
| 1 |
import os
|
| 2 |
import json
|
|
|
|
| 3 |
import tempfile
|
|
|
|
|
|
|
| 4 |
import gradio as gr
|
| 5 |
from huggingface_hub import InferenceClient
|
| 6 |
from gtts import gTTS
|
| 7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
try:
|
| 9 |
import PyPDF2
|
| 10 |
except ImportError:
|
| 11 |
PyPDF2 = None
|
| 12 |
|
| 13 |
|
| 14 |
-
# =========================================
|
| 15 |
-
#
|
| 16 |
-
# =========================================
|
| 17 |
|
| 18 |
MODEL_OPTIONS = {
|
| 19 |
-
"Zephyr 7B (
|
| 20 |
-
"Mixtral 8x7B
|
| 21 |
}
|
| 22 |
|
| 23 |
FALLBACK_MODEL = "HuggingFaceH4/zephyr-7b-beta"
|
| 24 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 25 |
PERSONA_PRESETS = {
|
| 26 |
"Balanced Assistant": "You are a helpful, intelligent AI assistant.",
|
| 27 |
-
|
| 28 |
"Forensic Psychologist": """
|
| 29 |
You are acting as a forensic psychologist.
|
| 30 |
Provide structured, objective, evidence-based analysis.
|
| 31 |
Do NOT diagnose real individuals.
|
| 32 |
Separate facts, interpretations, and limitations.
|
| 33 |
Maintain professional neutrality.
|
| 34 |
-
This analysis is educational and not a clinical diagnosis.
|
| 35 |
"""
|
| 36 |
}
|
| 37 |
|
| 38 |
TASK_MODES = {
|
| 39 |
"General Chat": "{user_input}",
|
| 40 |
-
|
| 41 |
"Behavioral Analysis": """
|
| 42 |
-
Conduct a behavioral analysis
|
| 43 |
1. Behavioral Indicators
|
| 44 |
2. Emotional Tone
|
| 45 |
3. Cognitive Patterns
|
| 46 |
-
4.
|
| 47 |
-
5.
|
| 48 |
-
6.
|
| 49 |
-
|
| 50 |
Material:
|
| 51 |
{user_input}
|
| 52 |
""",
|
| 53 |
-
|
| 54 |
-
"Explain Code (Natural Voice)": """
|
| 55 |
Explain the following code clearly and conversationally,
|
| 56 |
-
as if teaching a developer step-by-step
|
| 57 |
|
| 58 |
Code:
|
| 59 |
{user_input}
|
| 60 |
"""
|
| 61 |
}
|
| 62 |
|
| 63 |
-
MAX_CONTEXT_CHARS = 10000
|
| 64 |
-
MAX_HISTORY_PAIRS = 6
|
| 65 |
-
MAX_INPUT_CHARS = 3000
|
| 66 |
|
| 67 |
-
|
| 68 |
-
#
|
| 69 |
-
#
|
| 70 |
-
# =========================================
|
| 71 |
|
| 72 |
def get_client(model_id):
|
| 73 |
token = os.environ.get("HF_TOKEN")
|
|
@@ -76,6 +82,39 @@ def get_client(model_id):
|
|
| 76 |
return InferenceClient(model=model_id, token=token)
|
| 77 |
|
| 78 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 79 |
def extract_text(file):
|
| 80 |
if not file:
|
| 81 |
return ""
|
|
@@ -104,87 +143,112 @@ def text_to_speech(text):
|
|
| 104 |
return temp_file.name
|
| 105 |
|
| 106 |
|
| 107 |
-
def
|
| 108 |
-
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 114 |
)
|
| 115 |
-
return response.choices[0].message.content
|
| 116 |
|
|
|
|
|
|
|
| 117 |
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
| 121 |
|
| 122 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 123 |
|
| 124 |
if not message:
|
| 125 |
return history, None
|
| 126 |
|
| 127 |
-
|
| 128 |
-
|
| 129 |
-
history = history[-MAX_HISTORY_PAIRS * 2:]
|
| 130 |
-
|
| 131 |
-
model_id = MODEL_OPTIONS[model_label]
|
| 132 |
-
system_prompt = PERSONA_PRESETS.get(persona_label, "")
|
| 133 |
-
temperature = 0.4 if "Forensic" in persona_label else 0.7
|
| 134 |
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
system_prompt += f"\n\nUse this document context when relevant:\n\n{file_text}"
|
| 139 |
|
| 140 |
-
|
| 141 |
-
|
|
|
|
|
|
|
| 142 |
|
| 143 |
-
|
| 144 |
-
|
| 145 |
|
| 146 |
-
|
| 147 |
-
|
| 148 |
-
answer = run_model(client, messages, temperature)
|
| 149 |
-
except Exception:
|
| 150 |
-
fallback_client = get_client(FALLBACK_MODEL)
|
| 151 |
-
answer = "⚠️ Fallback model used.\n\n" + run_model(fallback_client, messages, temperature)
|
| 152 |
|
| 153 |
-
|
| 154 |
-
history.append({"role": "assistant", "content": answer})
|
| 155 |
|
| 156 |
-
|
| 157 |
-
|
| 158 |
-
audio_file = text_to_speech(answer)
|
| 159 |
|
| 160 |
-
|
| 161 |
|
| 162 |
-
|
| 163 |
-
history.append({"role": "assistant", "content": f"⚠️ ERROR:\n\n{str(e)}"})
|
| 164 |
-
return history, None
|
| 165 |
|
|
|
|
| 166 |
|
| 167 |
-
# =========================================
|
| 168 |
-
# Utilities
|
| 169 |
-
# =========================================
|
| 170 |
|
| 171 |
def clear_chat():
|
| 172 |
return [], None
|
| 173 |
|
| 174 |
-
def export_chat(history):
|
| 175 |
-
file_path = "conversation.json"
|
| 176 |
-
with open(file_path, "w") as f:
|
| 177 |
-
json.dump(history, f, indent=2)
|
| 178 |
-
return file_path
|
| 179 |
|
| 180 |
-
|
| 181 |
-
# =========================================
|
| 182 |
# UI
|
| 183 |
-
# =========================================
|
| 184 |
|
| 185 |
with gr.Blocks(theme=gr.themes.Soft(), title="Omniscient IRIS") as demo:
|
| 186 |
|
| 187 |
-
gr.Markdown("## Omniscient IRIS — Voice Enabled")
|
| 188 |
|
| 189 |
with gr.Row():
|
| 190 |
|
|
@@ -192,7 +256,7 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Omniscient IRIS") as demo:
|
|
| 192 |
|
| 193 |
chatbot = gr.Chatbot(height=500, type="messages")
|
| 194 |
msg = gr.Textbox(
|
| 195 |
-
placeholder="Enter material for analysis
|
| 196 |
show_label=False
|
| 197 |
)
|
| 198 |
|
|
@@ -221,26 +285,33 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Omniscient IRIS") as demo:
|
|
| 221 |
label="Mode"
|
| 222 |
)
|
| 223 |
|
| 224 |
-
voice_toggle = gr.Checkbox(
|
|
|
|
|
|
|
|
|
|
| 225 |
|
| 226 |
file_upload = gr.File(
|
| 227 |
-
label="Upload Document
|
| 228 |
file_types=[".txt", ".pdf", ".md"]
|
| 229 |
)
|
| 230 |
|
| 231 |
-
|
| 232 |
|
| 233 |
state = gr.State([])
|
| 234 |
|
| 235 |
send_btn.click(
|
| 236 |
generate_response,
|
| 237 |
-
inputs=[msg, state, model_selector,
|
|
|
|
|
|
|
| 238 |
outputs=[chatbot, audio_output],
|
| 239 |
)
|
| 240 |
|
| 241 |
msg.submit(
|
| 242 |
generate_response,
|
| 243 |
-
inputs=[msg, state, model_selector,
|
|
|
|
|
|
|
| 244 |
outputs=[chatbot, audio_output],
|
| 245 |
)
|
| 246 |
|
|
@@ -249,10 +320,10 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Omniscient IRIS") as demo:
|
|
| 249 |
outputs=[chatbot, audio_output],
|
| 250 |
).then(lambda: [], outputs=state)
|
| 251 |
|
| 252 |
-
|
| 253 |
-
|
| 254 |
inputs=chatbot,
|
| 255 |
-
outputs=gr.File(label="Download
|
| 256 |
)
|
| 257 |
|
| 258 |
if __name__ == "__main__":
|
|
|
|
| 1 |
import os
|
| 2 |
import json
|
| 3 |
+
import time
|
| 4 |
import tempfile
|
| 5 |
+
from datetime import datetime
|
| 6 |
+
|
| 7 |
import gradio as gr
|
| 8 |
from huggingface_hub import InferenceClient
|
| 9 |
from gtts import gTTS
|
| 10 |
|
| 11 |
+
from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer
|
| 12 |
+
from reportlab.lib.styles import ParagraphStyle, getSampleStyleSheet
|
| 13 |
+
from reportlab.lib import colors
|
| 14 |
+
from reportlab.lib.units import inch
|
| 15 |
+
|
| 16 |
try:
|
| 17 |
import PyPDF2
|
| 18 |
except ImportError:
|
| 19 |
PyPDF2 = None
|
| 20 |
|
| 21 |
|
| 22 |
+
# =========================================================
|
| 23 |
+
# CONFIGURATION
|
| 24 |
+
# =========================================================
|
| 25 |
|
| 26 |
MODEL_OPTIONS = {
|
| 27 |
+
"Zephyr 7B (Stable)": "HuggingFaceH4/zephyr-7b-beta",
|
| 28 |
+
"Mixtral 8x7B": "mistralai/Mixtral-8x7B-Instruct-v0.1",
|
| 29 |
}
|
| 30 |
|
| 31 |
FALLBACK_MODEL = "HuggingFaceH4/zephyr-7b-beta"
|
| 32 |
|
| 33 |
+
MAX_CONTEXT_CHARS = 10000
|
| 34 |
+
MAX_HISTORY_PAIRS = 6
|
| 35 |
+
MAX_INPUT_CHARS = 3000
|
| 36 |
+
CHAT_LOG_FILE = "chat_master_log.json"
|
| 37 |
+
|
| 38 |
+
|
| 39 |
PERSONA_PRESETS = {
|
| 40 |
"Balanced Assistant": "You are a helpful, intelligent AI assistant.",
|
|
|
|
| 41 |
"Forensic Psychologist": """
|
| 42 |
You are acting as a forensic psychologist.
|
| 43 |
Provide structured, objective, evidence-based analysis.
|
| 44 |
Do NOT diagnose real individuals.
|
| 45 |
Separate facts, interpretations, and limitations.
|
| 46 |
Maintain professional neutrality.
|
|
|
|
| 47 |
"""
|
| 48 |
}
|
| 49 |
|
| 50 |
TASK_MODES = {
|
| 51 |
"General Chat": "{user_input}",
|
|
|
|
| 52 |
"Behavioral Analysis": """
|
| 53 |
+
Conduct a behavioral analysis:
|
| 54 |
1. Behavioral Indicators
|
| 55 |
2. Emotional Tone
|
| 56 |
3. Cognitive Patterns
|
| 57 |
+
4. Risk Observations
|
| 58 |
+
5. Alternative Explanations
|
| 59 |
+
6. Limitations
|
| 60 |
+
|
| 61 |
Material:
|
| 62 |
{user_input}
|
| 63 |
""",
|
| 64 |
+
"Explain Code": """
|
|
|
|
| 65 |
Explain the following code clearly and conversationally,
|
| 66 |
+
as if teaching a developer step-by-step.
|
| 67 |
|
| 68 |
Code:
|
| 69 |
{user_input}
|
| 70 |
"""
|
| 71 |
}
|
| 72 |
|
|
|
|
|
|
|
|
|
|
| 73 |
|
| 74 |
+
# =========================================================
|
| 75 |
+
# INFERENCE LAYER
|
| 76 |
+
# =========================================================
|
|
|
|
| 77 |
|
| 78 |
def get_client(model_id):
|
| 79 |
token = os.environ.get("HF_TOKEN")
|
|
|
|
| 82 |
return InferenceClient(model=model_id, token=token)
|
| 83 |
|
| 84 |
|
| 85 |
+
def run_with_retry(client, messages, temperature, retries=3, delay=2):
|
| 86 |
+
for attempt in range(retries):
|
| 87 |
+
try:
|
| 88 |
+
response = client.chat_completion(
|
| 89 |
+
messages=messages,
|
| 90 |
+
max_tokens=700,
|
| 91 |
+
temperature=temperature,
|
| 92 |
+
top_p=0.95,
|
| 93 |
+
stream=False,
|
| 94 |
+
)
|
| 95 |
+
return response.choices[0].message.content
|
| 96 |
+
except Exception as e:
|
| 97 |
+
if attempt < retries - 1:
|
| 98 |
+
time.sleep(delay)
|
| 99 |
+
else:
|
| 100 |
+
raise e
|
| 101 |
+
|
| 102 |
+
|
| 103 |
+
def generate_ai_response(model_id, messages, temperature):
|
| 104 |
+
try:
|
| 105 |
+
client = get_client(model_id)
|
| 106 |
+
return run_with_retry(client, messages, temperature)
|
| 107 |
+
except Exception:
|
| 108 |
+
fallback_client = get_client(FALLBACK_MODEL)
|
| 109 |
+
return "⚠️ Fallback model used.\n\n" + run_with_retry(
|
| 110 |
+
fallback_client, messages, temperature
|
| 111 |
+
)
|
| 112 |
+
|
| 113 |
+
|
| 114 |
+
# =========================================================
|
| 115 |
+
# UTILITIES
|
| 116 |
+
# =========================================================
|
| 117 |
+
|
| 118 |
def extract_text(file):
|
| 119 |
if not file:
|
| 120 |
return ""
|
|
|
|
| 143 |
return temp_file.name
|
| 144 |
|
| 145 |
|
| 146 |
+
def append_to_master_log(user_msg, assistant_msg):
|
| 147 |
+
entry = {
|
| 148 |
+
"timestamp": datetime.utcnow().isoformat(),
|
| 149 |
+
"user": user_msg,
|
| 150 |
+
"assistant": assistant_msg
|
| 151 |
+
}
|
| 152 |
+
|
| 153 |
+
if os.path.exists(CHAT_LOG_FILE):
|
| 154 |
+
with open(CHAT_LOG_FILE, "r") as f:
|
| 155 |
+
data = json.load(f)
|
| 156 |
+
else:
|
| 157 |
+
data = []
|
| 158 |
+
|
| 159 |
+
data.append(entry)
|
| 160 |
+
|
| 161 |
+
with open(CHAT_LOG_FILE, "w") as f:
|
| 162 |
+
json.dump(data, f, indent=2)
|
| 163 |
+
|
| 164 |
+
|
| 165 |
+
def export_pdf(history):
|
| 166 |
+
file_path = f"transcript_{datetime.utcnow().strftime('%Y%m%d_%H%M%S')}.pdf"
|
| 167 |
+
doc = SimpleDocTemplate(file_path)
|
| 168 |
+
elements = []
|
| 169 |
+
|
| 170 |
+
styles = getSampleStyleSheet()
|
| 171 |
+
user_style = ParagraphStyle(
|
| 172 |
+
"UserStyle",
|
| 173 |
+
parent=styles["Normal"],
|
| 174 |
+
textColor=colors.blue,
|
| 175 |
+
spaceAfter=6
|
| 176 |
+
)
|
| 177 |
+
|
| 178 |
+
assistant_style = ParagraphStyle(
|
| 179 |
+
"AssistantStyle",
|
| 180 |
+
parent=styles["Normal"],
|
| 181 |
+
textColor=colors.darkgreen,
|
| 182 |
+
spaceAfter=12
|
| 183 |
)
|
|
|
|
| 184 |
|
| 185 |
+
elements.append(Paragraph("Omniscient IRIS Transcript", styles["Heading1"]))
|
| 186 |
+
elements.append(Spacer(1, 0.3 * inch))
|
| 187 |
|
| 188 |
+
for msg in history:
|
| 189 |
+
role = msg["role"]
|
| 190 |
+
content = msg["content"].replace("\n", "<br/>")
|
| 191 |
|
| 192 |
+
if role == "user":
|
| 193 |
+
elements.append(Paragraph(f"<b>User:</b> {content}", user_style))
|
| 194 |
+
elif role == "assistant":
|
| 195 |
+
elements.append(Paragraph(f"<b>Assistant:</b> {content}", assistant_style))
|
| 196 |
+
|
| 197 |
+
doc.build(elements)
|
| 198 |
+
return file_path
|
| 199 |
+
|
| 200 |
+
|
| 201 |
+
# =========================================================
|
| 202 |
+
# CORE CHAT LOGIC
|
| 203 |
+
# =========================================================
|
| 204 |
+
|
| 205 |
+
def generate_response(message, history, model_label, persona_label,
|
| 206 |
+
task_mode, uploaded_file, voice_enabled):
|
| 207 |
|
| 208 |
if not message:
|
| 209 |
return history, None
|
| 210 |
|
| 211 |
+
message = message[:MAX_INPUT_CHARS]
|
| 212 |
+
history = history[-MAX_HISTORY_PAIRS * 2:]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 213 |
|
| 214 |
+
model_id = MODEL_OPTIONS[model_label]
|
| 215 |
+
system_prompt = PERSONA_PRESETS[persona_label]
|
| 216 |
+
temperature = 0.4 if "Forensic" in persona_label else 0.7
|
|
|
|
| 217 |
|
| 218 |
+
file_text = extract_text(uploaded_file)
|
| 219 |
+
if file_text:
|
| 220 |
+
file_text = file_text[:MAX_CONTEXT_CHARS]
|
| 221 |
+
system_prompt += f"\n\nDocument Context:\n{file_text}"
|
| 222 |
|
| 223 |
+
messages = [{"role": "system", "content": system_prompt}]
|
| 224 |
+
messages.extend(history)
|
| 225 |
|
| 226 |
+
formatted_input = TASK_MODES[task_mode].format(user_input=message)
|
| 227 |
+
messages.append({"role": "user", "content": formatted_input})
|
|
|
|
|
|
|
|
|
|
|
|
|
| 228 |
|
| 229 |
+
answer = generate_ai_response(model_id, messages, temperature)
|
|
|
|
| 230 |
|
| 231 |
+
history.append({"role": "user", "content": message})
|
| 232 |
+
history.append({"role": "assistant", "content": answer})
|
|
|
|
| 233 |
|
| 234 |
+
append_to_master_log(message, answer)
|
| 235 |
|
| 236 |
+
audio_file = text_to_speech(answer) if voice_enabled else None
|
|
|
|
|
|
|
| 237 |
|
| 238 |
+
return history, audio_file
|
| 239 |
|
|
|
|
|
|
|
|
|
|
| 240 |
|
| 241 |
def clear_chat():
|
| 242 |
return [], None
|
| 243 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 244 |
|
| 245 |
+
# =========================================================
|
|
|
|
| 246 |
# UI
|
| 247 |
+
# =========================================================
|
| 248 |
|
| 249 |
with gr.Blocks(theme=gr.themes.Soft(), title="Omniscient IRIS") as demo:
|
| 250 |
|
| 251 |
+
gr.Markdown("## Omniscient IRIS — Resilient Voice Enabled Assistant")
|
| 252 |
|
| 253 |
with gr.Row():
|
| 254 |
|
|
|
|
| 256 |
|
| 257 |
chatbot = gr.Chatbot(height=500, type="messages")
|
| 258 |
msg = gr.Textbox(
|
| 259 |
+
placeholder="Enter material for analysis...",
|
| 260 |
show_label=False
|
| 261 |
)
|
| 262 |
|
|
|
|
| 285 |
label="Mode"
|
| 286 |
)
|
| 287 |
|
| 288 |
+
voice_toggle = gr.Checkbox(
|
| 289 |
+
label="Enable Voice Output",
|
| 290 |
+
value=True
|
| 291 |
+
)
|
| 292 |
|
| 293 |
file_upload = gr.File(
|
| 294 |
+
label="Upload Document",
|
| 295 |
file_types=[".txt", ".pdf", ".md"]
|
| 296 |
)
|
| 297 |
|
| 298 |
+
pdf_btn = gr.Button("Download PDF Transcript")
|
| 299 |
|
| 300 |
state = gr.State([])
|
| 301 |
|
| 302 |
send_btn.click(
|
| 303 |
generate_response,
|
| 304 |
+
inputs=[msg, state, model_selector,
|
| 305 |
+
persona_selector, task_selector,
|
| 306 |
+
file_upload, voice_toggle],
|
| 307 |
outputs=[chatbot, audio_output],
|
| 308 |
)
|
| 309 |
|
| 310 |
msg.submit(
|
| 311 |
generate_response,
|
| 312 |
+
inputs=[msg, state, model_selector,
|
| 313 |
+
persona_selector, task_selector,
|
| 314 |
+
file_upload, voice_toggle],
|
| 315 |
outputs=[chatbot, audio_output],
|
| 316 |
)
|
| 317 |
|
|
|
|
| 320 |
outputs=[chatbot, audio_output],
|
| 321 |
).then(lambda: [], outputs=state)
|
| 322 |
|
| 323 |
+
pdf_btn.click(
|
| 324 |
+
export_pdf,
|
| 325 |
inputs=chatbot,
|
| 326 |
+
outputs=gr.File(label="Download PDF")
|
| 327 |
)
|
| 328 |
|
| 329 |
if __name__ == "__main__":
|