arm
sf_llama_l_n200_v1 · repeats
One config on scifact_200, scored one way; its repeats averaged per item.
- dataset
- scifact_200
- scored against
- gold/scifact_200 gold
- goal
- Evidence retrieval for scientific claims (SciFact)
- runs
- 1
- headline
- ndcg_10 0.715
- last run
- 2026-09-29 16:49
Inspect itAgainst the baselineWith the 26 arms scored the same way
01metrics
| metric | mean | items | better |
|---|---|---|---|
| ndcg_10 | 0.715 | 200 | higher |
| cost_usd | $0.000846 | 200 | lower |
| duplicate_ids | 0.000 | 200 | neutral |
| found_any | 0.865 | 200 | neutral |
| latency_ms | 11385.9 | 200 | lower |
| llm_named_unknown | 0.025 | 200 | neutral |
| llm_parsed | 0.865 | 200 | neutral |
| mrr_10 | 0.704 | 200 | higher |
| n_relevant | 1.135 | 200 | neutral |
| n_returned | 100.0 | 200 | neutral |
| parsed | 1.000 | 200 | higher |
| rank_of_first | 2.745 | 200 | neutral |
| reasoning_tokens | 0.000 | 200 | neutral |
| recall_10 | 0.792 | 200 | higher |
| recall_100 | 0.859 | 200 | higher |
| tokens_in | 3848.3 | 200 | lower |
| tokens_out | 132.4 | 200 | lower |
| truncated | 0.000 | 200 | neutral |
| unknown_ids | 0.000 | 200 | neutral |
02cost and speed
- cost per item
- $0.000846 · 3,981 tok
- speed per item
- p50 10,149 · p95 28,200 ms
Per item, over its items: cost as dollars and tokens (a local model is free but still uses tokens), speed as the median and the slow tail (p95). A goal that names this arm weighs them in its decision.
03runs
| run | made | items | cost | how |
|---|---|---|---|---|
sf_llama_l_n200_v1_20260929T164943Z | 2026-09-29 16:49 | 200 | $0.169174 | reparse |
04config
config sf_llama_l_n200_v1 · adapter examples/retrieval-scifact/adapter.py · credentials are never shown
- corpus_id
scifact- family
llama- max_tokens
700- model
eval-llama-maverick- price_tier
large- prompt_file
prompt.txt- provider
rerank- reasoning
{"enabled": false}- rerank_k
20- snippet_chars
900- temperature
0.0- top_k
100- usd_per_mtok_in
0.18- usd_per_mtok_out
0.6
05conditions
- data.items_sha256
b97dfe735a78eb4978cde41e2ca9a0bff424de9e6a0a2bdada5d167e3423004d- deployment.model
eval-llama-maverick- deployment.source
adapter- fingerprint.version
5- library.numpy
1.26.4- library.openai
3.19.2- library.scikit-learn
1.9.1- library.sentence-transformers
3.4.1- library.torch
2.2.2- library.transformers
4.55.4- machine.cpu_count
12- machine.label
mac-dev- machine.platform
Darwin/x86_64- machine.processor
i386- machine.python
3.11.8- machine.release
24.6.0- reference.sha256
b522e08017ebc7dd71c5627bccfdd688c5124096df8a5e6454684141bc361d99- scorer.id
examples/retrieval-scifact/adapter.py- scorer.sha256
cd1ffd21ed850b315d8c2e0b8f61d11d0945e21c3c8c0d5b136247dedc285e9c
06the same config, scored other ways
scored against silverscored by a judgescored against a label set, not a reference setunmarked rows are scored against gold
| arm | scored against | headline | mean |
|---|---|---|---|
| sf_llama_l_n200_v1 | gold/scifact_200 gold | ndcg_10 | 0.728 |
07rounds
- Round r3 of Evidence retrieval for scientific claims (SciFact), closed: retired in it.