llm_attribute_runner
Batch execution helpers for hierarchy attribute extraction.
The module exposes a single public entry point, :func:process_hierarchy,
which executes the term-level extraction pipeline over a tabular dataset.
Execution is resumable: each term's result is written to its own JSON file as
soon as it succeeds, so an interrupted run resumes by reprocessing only the
terms without a cached file. Failed terms are left uncached and retried on the
next run.
Terms are processed concurrently by max_workers threads (the work is
dominated by waiting on LLM API calls). Workers only run the extraction;
caching, cost tracking and logging happen in the calling thread.
load_hierarchy_results(path)
Load pipeline results from a JSON list file or a folder of per-term JSON files.
Unreadable per-term files and results without a source_key are skipped with a warning.
Source code in src/ariadne/hierarchy/llm_attribute_runner.py
257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 | |
process_hierarchy(terms, attribute_searcher, reference_searcher, hierarchy_settings, source_term_column='source_term', source_code_column='source_concept_code', source_vocab_column='source_vocabulary_id', source_id_column='source_concept_id', checkpoint_every=5, max_workers=None)
Execute hierarchy extraction over source terms with checkpoint resume.
Parameters
terms
Input rows containing source identifiers and term text.
attribute_searcher
Searcher used by the extraction pipeline for attribute candidates.
reference_searcher
Searcher used for optional reference concept lookups.
hierarchy_settings
Required runtime settings. evaluation.llm_response_folder is used to
store one result file per source term for resume.
source_term_column
Name of the column containing source term text.
source_code_column
Name of the column containing the source concept code.
source_vocab_column
Name of the column containing the source vocabulary id. Combined with
source_code_column to form the composite primary key.
source_id_column
Optional column with the OMOP concept_id (may be absent for new
vocabularies). Carried through when present.
checkpoint_every
Cadence (in terms) for emitting a progress log line. Results are
persisted immediately per term regardless of this value.
max_workers
Number of terms processed concurrently. Defaults to
hierarchy_settings.evaluation.max_workers; 1 runs sequentially.
Returns
list[dict] One result dictionary per processed source code: previously cached results first, then newly processed terms in input order.
Raises
ValueError
If hierarchy_settings is missing.
Source code in src/ariadne/hierarchy/llm_attribute_runner.py
96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 | |