MarRan85/isla_state_slm_v1

Isla State SLM v1

Isla State SLM v1 is a lightweight text classification model that predicts a user's current regulation / arousal state from short self-expressive text.

The model is designed to be used as a routing signal in conversational systems, not as a diagnostic or clinical tool.

---

Labels

The model outputs one of three classes:

• low
  Low energy or depleted state.
  Examples: exhaustion, burnout, shutdown, low motivation, flat affect.

• base
  Regulated or baseline state.
  Examples: calm, steady, neutral, functional, confident without distress.

• high
  High arousal distress state.
  Examples: anxiety, overwhelm, racing thoughts, agitation, acute stress.

These labels represent state intensity, not emotional valence.

---

Intended Use

• Conversation routing (e.g. selecting grounding vs recovery responses)
• Emotional state detection in wellbeing-oriented applications
• Lightweight first-pass signal before deeper reasoning or LLM responses

---

Not Intended For

• Medical or psychological diagnosis
• Crisis detection or suicide risk assessment
• Use as a standalone decision-maker
• High-stakes or safety-critical systems

---

Example

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

repo_id = "MarRan85/isla_state_slm_v1"
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForSequenceClassification.from_pretrained(repo_id)
model.eval()

text = "I feel overwhelmed and my thoughts are racing."

inputs = tokenizer(text, return_tensors="pt", truncation=True)
with torch.no_grad():
    probs = torch.softmax(model(**inputs).logits, dim=-1)[0]

pred_id = int(torch.argmax(probs))
label = model.config.id2label[pred_id]
confidence = float(torch.max(probs))

print(label, confidence)