Psychocounsel / app.py
razaali10's picture
Update app.py
e5d5895 verified
Raw History Blame Contribute Delete
4.05 kB
import os
import re
import gradio as gr
from huggingface_hub import InferenceClient
# ----------------------------------------
# HF Setup (use HF Router)
# ----------------------------------------
MODEL_ID = "Psychotherapy-LLM/PsychoCounsel-Llama3-8B"
HF_TOKEN = os.getenv("HF_TOKEN")
if HF_TOKEN:
client = InferenceClient(model=MODEL_ID, token=HF_TOKEN)
else:
client = InferenceClient(model=MODEL_ID)
# ----------------------------------------
# Build prompt
# ----------------------------------------
def build_prompt(client_speech: str, context: str, mode: str) -> str:
if mode == "Brief (5–7 sentences)":
instruction = (
"You are a professional psychotherapist. Respond in 5–7 warm, reflective, "
"empathic sentences. Only output what the therapist says."
)
else:
instruction = (
"You are a professional psychotherapist. Produce a detailed multi-paragraph "
"therapeutic response similar to the Appendix case studies. Start with validation, "
"explore beliefs, discuss values, suggest a small exercise, and end with an open question."
)
if context:
instruction += f" Therapist context: {context}"
return f"""{instruction}
Client Speech:
{client_speech}
Therapist:
"""
# ----------------------------------------
# Generate
# ----------------------------------------
def generate_response(client_speech, context, mode, temperature, top_p):
if not client_speech or not client_speech.strip():
return "⚠️ Please enter client speech."
prompt = build_prompt(client_speech, context, mode)
max_tokens = 220 if mode == "Brief (5–7 sentences)" else 450
try:
output = client.text_generation(
prompt,
max_new_tokens=max_tokens,
temperature=float(temperature),
top_p=float(top_p),
do_sample=True,
return_full_text=False,
)
except Exception as e:
return f"⚠️ HF API Error: {e}"
text = output.strip()
text = text.split("Note:")[0].split("FINAL ANSWER")[0].strip()
# Limit to 7 sentences in brief mode
if mode == "Brief (5–7 sentences)":
sents = re.split(r"(?<=[.!?])\s+", text)
text = " ".join(sents[:7])
return text
# ----------------------------------------
# UI
# ----------------------------------------
DEFAULT_CLIENT_SPEECH = (
"I’ve been having emotional issues for a few years. Nonetheless, these have been somewhat "
"manageable. However, I became increasingly paranoid this winter..."
)
DESCRIPTION = (
"This replicates Appendix-style psychotherapist responses using PsychoCounsel-Llama3-8B.\n\n"
"⚠️ Not for clinical use. Research demo only."
)
with gr.Blocks(title="PsychoCounsel-Llama3-8B — Research Demo") as demo:
gr.Markdown("# 🧠 PsychoCounsel-Llama3-8B — Original / Research Demo")
gr.Markdown(DESCRIPTION)
with gr.Row():
with gr.Column(scale=1):
mode_radio = gr.Radio(
["Brief (5–7 sentences)", "Extended (Appendix-style)"],
value="Extended (Appendix-style)",
label="Response Style"
)
temp = gr.Slider(0.1, 1.0, value=0.6, step=0.05, label="Temperature")
top_p = gr.Slider(0.5, 1.0, value=0.9, step=0.05, label="Top-p")
with gr.Column(scale=2):
speech_box = gr.Textbox(
label="Client Speech",
value=DEFAULT_CLIENT_SPEECH,
lines=12
)
context_box = gr.Textbox(
label="Optional Therapist Context",
value="",
lines=3
)
generate_btn = gr.Button("Generate Therapist Response", variant="primary")
output = gr.Markdown()
generate_btn.click(
generate_response,
inputs=[speech_box, context_box, mode_radio, temp, top_p],
outputs=output,
)
if __name__ == "__main__":
demo.launch()