{% extends "base.html" %} {% block title %}{{ current_interp.label|default:feature.label|default:"Feature Detail" }} · Run #{{ run.id }}{%endblock %} {% block extra_head %} {% endblock %} {% block content %}
SAE MODEL #{{ run.id }}

FEATURE {{ feature.feature_index }}

{% if feature.label %} / {{ feature.label }} {% endif %}
Refresh Stats

This page shows the detailed analysis of a specific feature in the Sparse Autoencoder (SAE). It includes the semantic interpretation generated by the LLM, the top activating documents that trigger this feature, and statistical metrics like density and activation distribution.

Semantic Interpretation
{% if current_interp %} {{ current_interp.llm_model }} #{{ current_interp.id }} {% endif %}

{{ current_interp.description|default:feature.description|default:"No interpretation available." }}

STATISTICS & DISTRIBUTION
Statistics Explained

Density
Fraction of documents where this feature activates at least once.
Higher density = feature appears frequently across the dataset.

Peak Activation
Maximum activation value observed for this feature over all documents.

Activation Histogram
Distribution of all activation values for this feature. Each bar represents a bin (range) of values, and its height shows how many documents fall in that range.

Density
{{ feature.density|floatformat:5 }}
Peak Act
{{ feature.max_activation|floatformat:2 }}
{% if feature.activation_histogram.counts %}
{% for count in feature.activation_histogram.counts %} {% widthratio count max_hist_count 100 as height_pct %}
{% endfor %}
0.0 Max
{% endif %}
Global Context
TOP ACTIVATING CONTEXTS
{% for doc in feature.example_docs %}
DOC #{{ doc.external_id|default:doc.id }} ANALYZE
ACT: {{ doc.act|floatformat:4 }}
{{ doc.text|truncatechars:350 }}
{% if feature.max_activation %}
{% widthratio doc.act feature.max_activation 100 as act_pct %}
{% endif %}
{% empty %}
No cached examples found. Please scan the dataset.
{% endfor %}
{% if feature.example_docs|length > 5 %}
{% endif %}
{% if feature.family_parent.exists or feature.family_children.exists %}
TOPOLOGY
    {% for family in feature.family_parent.all %}
  • ROOT OF FAMILY #{{ family.id }}
    {% for child in family.children_features.all|slice:":5" %}
    {{ child.label|default:"Unknown" }} #{{ child.feature_index }}
    {% endfor %} {% if family.children_features.count > 5 %}
    + {{ family.children_features.count|add:"-5" }} others
    {% endif %}
    • {% endfor %} {% for family in feature.family_children.all %}
  • #{{ family.parent_feature.feature_index }}
    • {% endfor %}
{% endif %}
CORRELATIONS
W – Weight-space · similarity between decoder vectors (features that point in similar directions in the embedding space).
    {% for item in feature.correlated_features|slice:":6" %}
  • {{ item.label }} {{ item.score|floatformat:2 }}
    • {% empty %}
  • No weight-space correlations.
    • {% endfor %}
D – Data / co-activation · features that tend to fire together on the same documents (co-occurrence in the dataset).
    {% for item in feature.co_occurring_features|slice:":6" %}
  • {{ item.label }} {{ item.score|floatformat:2 }}
    • {% empty %}
  • No data co-occurrence.
    • {% endfor %}
{% include "explorer/includes/modals.html" %} {{ scatter_data|json_script:"scatter-data-json" }} {% endblock %} {% block extra_js %} {% endblock %}