An attribution graph as Neuronpedia's schema (graph-schema.json) and circuit-tracer's create_graph_files describe it.
More...
#include <attribution_graph.hpp>
|
| std::string | slug |
| |
| std::string | scan |
| | The model id, for example "gemma-2-2b"; Neuronpedia uses it to find the model.
|
| |
| std::string | prompt |
| |
| std::vector< std::string > | prompt_tokens |
| |
| std::optional< double > | node_threshold |
| | The viewer's default pruning threshold (circuit-tracer uses 0.8).
|
| |
| std::string | description |
| | Optional metadata.info text.
|
| |
| std::vector< Node > | nodes |
| |
| std::vector< Link > | links |
| |
An attribution graph as Neuronpedia's schema (graph-schema.json) and circuit-tracer's create_graph_files describe it.
- Note
- The viewer places a node by its
layer ("E" for embeddings, then "0", "1", ..., and the logit layer last) and its ctx_idx (token position), and draws a link's weight as its width and sign.
◆ description
| std::string pulsatrix::AttributionGraph::description |
Optional metadata.info text.
◆ links
| std::vector<Link> pulsatrix::AttributionGraph::links |
◆ node_threshold
| std::optional<double> pulsatrix::AttributionGraph::node_threshold |
The viewer's default pruning threshold (circuit-tracer uses 0.8).
◆ nodes
| std::vector<Node> pulsatrix::AttributionGraph::nodes |
◆ prompt
| std::string pulsatrix::AttributionGraph::prompt |
◆ prompt_tokens
| std::vector<std::string> pulsatrix::AttributionGraph::prompt_tokens |
◆ scan
| std::string pulsatrix::AttributionGraph::scan |
The model id, for example "gemma-2-2b"; Neuronpedia uses it to find the model.
◆ slug
| std::string pulsatrix::AttributionGraph::slug |
The documentation for this struct was generated from the following file: