File size: 3,351 Bytes
0bca32c 5fd8123 0bca32c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 | ---
license: mit
task_categories:
- feature-extraction
- text-retrieval
tags:
- embeddings
- retrieval
- rag
- benchmarks
- embedding-models
size_categories:
- n<1K
pretty_name: "Embedding Eval Results — Round 1, Stage 1, Chunked A/B"
---
# Embedding Eval Results
The committed results from a real embedding-model benchmark that embarrassed a leaderboard's recommendation.
## What's in here
Four CSV files representing four evaluation runs on a personal Obsidian vault:
| File | Notes | Queries | Models | Purpose |
|---|---|---|---|---|
| `20260618-233912.csv` | 54 | 29 | 7 | Round 1 — toy slice. **Saturated benchmark.** |
| `20260619-025330.csv` | 995 | 150 | 7 | Stage 1 — real pile. The ranking inverted. |
| `20260620-142445-wholenote.csv` | 996 | 450 | 7 | Stage 1 enlarged — held the new order. |
| `20260620-172625-chunked.csv` | 18,279 chunks | 450 | 1 | Chunked A/B — chunking hurt every metric. |
## What the data shows
Round 1 on 54 notes: all 7 models looked great. recall@5 above 90% for every candidate. The "winner" by leaderboard-style ranking was `nomic-embed-text`.
Stage 1 on 995 notes: the ranking inverted. The mid-pack model `embeddinggemma` took first on every metric. The round-one "winner" sank to mid-pack. The default the system had been quietly using, `all-minilm`, dropped to last.
Same code. Same models. Same scoring. The only thing that changed was the size of the pile.
## Columns
Each CSV has the same columns:
```
name,provider,model,dim,recall@1,recall@5,recall@10,mrr@10,q_latency_ms,approx_cost_usd
```
- `name`: human-readable label (matches the model except for the legacy `all-minilm` row, which is annotated)
- `provider`: `ollama` (local) or `modal` (the project's own GPU service) — see sources below
- `model`: the actual model name passed to the embedder
- `dim`: embedding dimension
- `recall@1, recall@5, recall@10`: fraction of queries where the source note landed in the top-k
- `mrr@10`: mean reciprocal rank
- `q_latency_ms`: query-side embedding latency (note embedding happens separately)
- `approx_cost_usd`: zero for local models; the cheapest paid model cost roughly a few cents per run
## Source
The full eval harness is in the [pi-vault-mind](https://github.com/kylebrodeur/pi-vault-mind) repo at `eval/run_eval.py`. The companion blog post is ["Pick the Model From Your Own Data"](https://kylebrodeur.substack.com) (KE-1).
The simplified standalone version of the harness, written for the blog post, is at [kylebrodeur/embed-eval-on-your-vault](https://github.com/kylebrodeur/embed-eval-on-your-vault). It is a single-file Python script with no dependencies beyond `python3` that runs against any vault of `.md` or `.txt` notes.
## Reproducibility
Each CSV is timestamped (UTC). The harness commits the result on every run. Re-running with the same model/provider/dataset should produce the same numbers within rounding. Embedding models do drift across versions, so if you re-run after a model bump, expect ±1-2% on the metrics.
## Citation
If you use these numbers in another piece, cite the model row and the run timestamp. The full provenance is in the [pi-vault-mind](https://github.com/kylebrodeur/pi-vault-mind) repo commit log (the 2026-06-18 to 2026-06-20 range).
## License
MIT. Numbers are facts; the framing is the author's.
|