Perform batch vector search. Args: query_embeddings: Dictionary mapping query_id to embedding top_k:

Close the Dgraph client connection.

Generate embeddings for BEIR corpus. Args: corpus: BEIR corpus dictionary {doc_id: {"title": ..., "text": ...}}

Generate embeddings for BEIR queries. Args: queries: BEIR queries dictionary {query_id: query_text}

Fetch posting list cache metrics from Dgraph Prometheus endpoint.

getBenchCaches() returns a slice of benchCache's depending on the value of the include params (which is the cache/suite flags passed from main).

Get the default configuration.

Get Dgraph version from health endpoint. Args: host: Dgraph alpha host port: Dgraph alpha HTTP port Returns

Insert a benchmark JSON file as the first data row in the CSV.

Insert documents with embeddings into Dgraph. Args: corpus: BEIR corpus dictionary embeddings: Dicti

Load BEIR dataset, downloading if necessary. Args: dataset_name: Name of the BEIR dataset (e.g., 'scifact') data_dir: Di

Load results from CSV file.

Main evaluation pipeline.

typeName should include the escaped quotes, e.g "\"Person\""

Print a compact summary table.

Print results as a formatted table.

Reads a single line from a buffered reader. The line is read into the passed in buffer to minimize allocations. This is the preferred method for loadi

Drop all data and schema.

======================================================================== Benchmark Code ==============================================================

save writes all logs to the PATH file in CSV format.

Append evaluation results to CSV file.

Set up the Dgraph schema for BEIR documents. Args: hnsw_config: HNSW index configuration

Perform vector similarity search with computed similarity scores. Args: query_embedding: Query embedding vector

======================================================================== Possible workloads ==========================================================

ApproxLen returns the approximate number of UIDs in the UidPack object.

This is taking 16ms. Clone itself takes ~4ms. So, ~12ms for AND.

This is taking 60 microseconds over 1M entries.

Clone is taking ~5ms.

Bitmap runs in 177 ns/op. List search in 252 ns/op. UidPack search is 3024 ns/op.

This is taking 22ms for 1M entries.

Each add is 24ns.

Marshal bitmap is taking 3.2ms. pack is taking <1ms.