kert 5d8214711d feat(llm): LLM_INSTRUCT_MODEL overrides the instruct model per run ...

Lets eval-tags --live compare judge models (14b vs 32b on the rig)
without editing stack.toml.

2026-11-11 15:01:59 -05:00

..

__init__.py

feat(llm): module scaffold, stack[llm] extra, [llm] config (refs #563)

2026-07-17 08:15:31 -04:00

golden_lineage.yaml

fix(pfs,llm): descriptor stems trimmed to the code's own parenthetical (#700: G0502/G0503 no longer bridge into CCM); excerpts using a family acronym for something else are dropped (CCM = cardiac contractility modulation in the CY2027 NPRM) (refs #700 #691)

2026-09-10 17:48:30 -04:00

golden_themes_v2.yaml

test(llm): golden theme set v2 — 200 comments, 8 dockets, held-out split

2026-11-11 15:01:59 -05:00

golden_themes.yaml

feat(llm): golden-set gate for theme tagging — llm.tageval, tests/llm/golden_themes.yaml, stack llm eval-tags; vocab v2 (refs #577)

2026-09-22 16:58:41 -04:00

test_api.py

feat(llm): Word-exact pages — LibreOffice in the llm image converts .docx to PDF on demand (cached) for the document viewer

2026-09-24 19:12:14 -04:00

test_chunk.py

fix(llm,pfs): thread one code_index into chunk_doc so index and restamp agree (F1)

2026-09-09 20:57:43 -04:00

test_classify.py

test: cover P49 degrade paths so the 99% coverage gate holds (refs #721)

2026-09-11 15:37:56 -04:00

test_client.py

feat(llm,notebooks): httpx-only LlmClient for notebooks and an llm_search example notebook (refs #573)

2026-09-11 17:26:54 -04:00

test_config.py

feat(llm): LLM_INSTRUCT_MODEL overrides the instruct model per run

2026-11-11 15:01:59 -05:00

test_convert.py

feat(llm): Word-exact pages — LibreOffice in the llm image converts .docx to PDF on demand (cached) for the document viewer

2026-09-24 19:12:14 -04:00

test_evidence.py

test: cover P49 degrade paths so the 99% coverage gate holds (refs #721)

2026-09-11 15:37:56 -04:00

test_golden.py

fix(ci,llm): golden runner exits 0 when filing issues; workflow keeps the report on failure; cmd_run tests (refs #692)

2026-09-10 12:12:59 -04:00

test_index_refs.py

fix(llm,pfs): thread one code_index into chunk_doc so index and restamp agree (F1)

2026-09-09 20:57:43 -04:00

test_index.py

fix(llm): filtered vector search returns results — hnsw.iterative_scan = relaxed_order on every connection and at the database level

2026-09-24 20:07:38 -04:00

test_init_lazy.py

feat(llm,notebooks): httpx-only LlmClient for notebooks and an llm_search example notebook (refs #573)

2026-09-11 17:26:54 -04:00

test_iomlink.py

feat(llm): rule discussions ↔ manual sections — cited IOM sections join the chat as companion sources; search links them; rules theme-stamped

2026-09-24 20:25:33 -04:00

test_lineage.py

feat(llm): categorise manual sections by theme — stack llm stamp-themes, /search?theme=, themes on sources

2026-09-24 19:46:04 -04:00

test_links.py

fix(llm): chat reads whole excerpts and a rule hit's next paragraph

2026-11-11 13:39:52 -05:00

test_manuals.py

fix(llm): rules never double as corpus; citations say proposed/final/correction; IOM chapters indexed by section

2026-09-24 19:38:17 -04:00

test_metrics.py

fix(llm): size the chat prompt in tokens and refuse silent host-side truncation

2026-10-24 14:14:38 -04:00

test_migrate.py

perf(llm): btree index on embedding item_key

2026-11-11 12:55:09 -05:00

test_pages.py

fix(llm): rules never double as corpus; citations say proposed/final/correction; IOM chapters indexed by section

2026-09-24 19:38:17 -04:00

test_pdfs.py

feat(llm): the document viewer renders Word (.docx) and plain-text attachments in-page, not just PDFs

2026-09-24 19:02:57 -04:00

test_pool.py

fix(llm): match Ollama model tags exactly in the host pool

2026-11-11 12:47:29 -05:00

test_rag.py

fix(llm): chat reads whole excerpts and a rule hit's next paragraph

2026-11-11 13:39:52 -05:00

test_rerank.py

feat(llm): recency-blended re-ranking of retrieval hits

2026-09-03 10:32:14 -04:00

test_restamp.py

test(llm): restamp pagination-loop mechanics with a fake engine (F8)

2026-09-09 21:04:52 -04:00

test_search.py

fix(llm): search returns each citation label once — the same FR paragraph indexed under two bib items (2019-16041, 2024-14828 each have two rule items) showed twice; keep the closest hit

2026-09-24 20:29:59 -04:00

test_source_refs.py

feat(llm): stack llm index --key also limits the corpus collection

2026-09-09 21:28:04 -04:00

test_source.py

fix(bib): hoard IOM revisions instead of discarding them

2026-11-11 14:20:35 -05:00

test_tageval.py

feat(llm): eval-tags scores a golden split per docket with set thresholds

2026-11-11 14:49:00 -05:00

test_tagger.py

fix(llm): judge theme tags on instruct_model only

2026-11-11 13:12:52 -05:00

test_themes.py

feat(llm): theme filter pushed into the vector store — theme_1..theme_N rank keys beside themes, $or equality filter; stamp threshold 0.60

2026-09-24 19:59:21 -04:00

test_vocab.py

feat(llm): closed theme vocabulary for comment tagging — themes.yaml v1, llm.vocab loader, stack llm vocab, P35 design spec (refs #574)

2026-09-22 16:48:51 -04:00