{% extends "base.html" %}

{% block title %}New Experiment · Sparsifire{% endblock %}

{% block content %}
<div class="mb-4">
  <h6 class="text-uppercase text-muted mb-1 small ls-1">Training Center</h6>
  <h1 class="h3 mb-0">Start New Experiment</h1>
</div>

<div class="panel">
  <div class="row">
    <div class="col-lg-7">
      <form method="post" novalidate>
        {% csrf_token %}
        
        <div class="mb-4">
            {# Renderizza il form usando le classi Bootstrap definite in forms.py #}
            {{ form.as_p }}
        </div>

        <hr class="border-secondary border-opacity-25 my-4">

        <div class="d-flex gap-3">
            <button type="submit" class="btn-accent shadow-sm">
                <i class="bi bi-rocket-takeoff me-2"></i> Create Configuration
            </button>
            <a href="{% url 'sae:run_list' %}" class="btn-ghost text-decoration-none">
                Cancel
            </a>
        </div>
      </form>
    </div>

    <div class="col-lg-5 ps-lg-5 mt-5 mt-lg-0 border-start border-secondary border-opacity-10">
        <h5 class="text-white mb-3">
            <i class="bi bi-sliders me-2 text-accent"></i> Hyperparameters
        </h5>
        <p class="text-muted small mb-4">
            Configure the architecture and training constraints for the Sparse Autoencoder.
        </p>
        
        <ul class="list-unstyled text-muted small">
            <li class="mb-3">
                <strong class="d-block text-white mb-1">Expansion Factor</strong>
                Determines the size of the latent space relative to the input (e.g. 4x means latent_dim = input_dim * 4).
            </li>
            <li class="mb-3">
                <strong class="d-block text-white mb-1">K-Sparsity (Top-K)</strong>
                Enforces strict sparsity by keeping only the top K most active neurons per sample.
            </li>
            <li class="mb-3">
                <strong class="d-block text-white mb-1">Alpha Aux</strong>
                Coefficient for the auxiliary loss, used to revive "dead" neurons during training.
            </li>
        </ul>
    </div>
  </div>
</div>
{% endblock %}