mba/chunking_strategies_evaluation
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

chore(config): sync env example and context for ±3 defaults

Browse Source 
Why:
- d5ccec2 locked ±3 in code; env template and domain glossary were still at 0/0.

Changes:
- NEIGHBOR_PREV/NEXT=3 in .env.example
- CONTEXT.md Neighbor Expansion defaults updated to 3/3

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Mahdi Bazrafshan
2026-08-22 15:32:29 +03:30
parent dc76ca67b8
commit 47e4846270
2 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
.env.example
View File

@@ -14,9 +14,9 @@ OLLAMA_BASE_URL=http://192.168.10.10:11435
# Retrieval defaults # Retrieval defaults
TOP_K=5 TOP_K=5
# Neighbor Expansion for fixed_size only (ADR-0023); 0 = off # Neighbor Expansion for fixed_size only (ADR-0023); 3/3 = stabilized default after benchmark
NEIGHBOR_PREV=0 NEIGHBOR_PREV=3
NEIGHBOR_NEXT=0 NEIGHBOR_NEXT=3
# LLM generation parameters # LLM generation parameters
TEMPERATURE=0.0 TEMPERATURE=0.0

2
CONTEXT.md
View File

@@ -87,7 +87,7 @@ The configured catalog of Embedding Models the operator may assign to Boundary o
_Avoid_: Model list, provider catalog, available embeddings _Avoid_: Model list, provider catalog, available embeddings
**Neighbor Expansion**: **Neighbor Expansion**:
Query-time widening of retrieved context for the `fixed_size` Strategy only: for every top-k hit, also include a configurable number of previous and next chunks in document order within the same document. Neighbors are added on top of the top-k set (context may grow beyond k). The same chunk id appears at most once in the LLM context. Final LLM context is sorted by document order (`chunk_index`). Neighbors are not re-ranked as independent hits. Counts (`neighbor_prev` / `neighbor_next`) are set per Query or Experiment like `top_k`; defaults are `0`/`0` (off). Missing neighbors at document edges are skipped. Knobs are ignored for non-`fixed_size` Strategies. Dashboard labels Experiments with a compact `±P/N` badge (tooltip explains prev/next; muted when `fixed_size` was not in the run). Experiments list and Compare show these knobs; Compare warns when selected Experiments differ on Neighbor Expansion, Corpus Embedding Model, or Boundary Embedding Model (when recorded) so operators do not misread cross-run rankings. Distinct from how a Strategy cuts text at process time. Query-time widening of retrieved context for the `fixed_size` Strategy only: for every top-k hit, also include a configurable number of previous and next chunks in document order within the same document. Neighbors are added on top of the top-k set (context may grow beyond k). The same chunk id appears at most once in the LLM context. Final LLM context is sorted by document order (`chunk_index`). Neighbors are not re-ranked as independent hits. Counts (`neighbor_prev` / `neighbor_next`) are set per Query or Experiment like `top_k`; benchmarker defaults are `3`/`3` (symmetric ±3 after Strategy finalization); `0`/`0` turns expansion off. Missing neighbors at document edges are skipped. Knobs are ignored for non-`fixed_size` Strategies. Dashboard labels Experiments with a compact `±P/N` badge (tooltip explains prev/next; muted when `fixed_size` was not in the run). Experiments list and Compare show these knobs; Compare warns when selected Experiments differ on Neighbor Expansion, Corpus Embedding Model, or Boundary Embedding Model (when recorded) so operators do not misread cross-run rankings. Distinct from how a Strategy cuts text at process time.
_Avoid_: Context windowing, chunk padding, sliding window retrieval, adjacent chunk merge _Avoid_: Context windowing, chunk padding, sliding window retrieval, adjacent chunk merge
**Expansion Tree**: **Expansion Tree**: