Files
brain-of-reese/benchmarks/model_benchmarks.csv
T
ducoterra 542e568d25 docs(agent): record the reverted embed embedding benchmark
The embed alias was reverted to the previous backing model and re-benchmarked
over 3 runs. All gates pass and the numbers reproduce the original baseline
exactly: 768 dims, cosine margin 0.283 (quality 56) versus the 0.217 measured
on the swapped-in model, with throughput back to 109.7-123.3 vec/s.

The margin returning to the identical 0.283 confirms the locked semantic-pair
corpus is a reliable fingerprint for this endpoint, and a raw-vector probe
shows the two models are genuinely different spaces despite both being
unit-normalized 768-dim (max component diff 0.035 on the same input). Any
content embedded while the swapped-in model was live is therefore in an
incompatible space and needs re-embedding.

LLM-Generated: true
2026-10-09 12:41:38 -04:00

3.5 KiB

1datescriptmodelmodegate_statusturnsansweredcapstool_turnsemittedexecutedcontractwall_scontract_denomexecuted_denomtool_turns_denomextra1_colextra1_valextra2_colextra2_val
22026-09-06summarylitequalityFAIL8800005826.95808coherence4.0/5hallucinations4/8
32026-09-06summarylitequalityFAIL8800005325.45308coherence4.0/5hallucinations5/8
42026-09-06embedembeddimensionPASS11000010.1101actual_dim768expected_dim768
52026-09-06embedembedcosinePASS11000010.2101margin0.283quality56
62026-09-06embedembedspeedPASS11000010.5101rate93.8 vec/sbad_vectors0
72026-09-06summaryturboqualityFAIL8800006325.36308coherence4.0/5hallucinations3/8
82026-09-06chatlitefixturePASS101009991140.4121210
92026-09-06chatlitefixturePASS1010010991340.5141410
102026-09-06chatlitederivedFAIL1010010551038.3141410
112026-09-06chatturbofixturePASS101007999113.79910
122026-09-06chatturbofixturePASS101007777112.87710
132026-09-06chatturboderivedFAIL10100555597.25510
142026-10-09embedembeddimensionPASS11000010.2101actual_dim768expected_dim768
152026-10-09embedembedcosinePASS11000010.6101margin0.217quality43
162026-10-09embedembedspeedPASS11000011.4101rate35.6 vec/sbad_vectors0
172026-10-09embedembeddimensionPASS11000010.0101actual_dim768expected_dim768
182026-10-09embedembedcosinePASS11000010.6101margin0.217quality43
192026-10-09embedembedspeedPASS11000011.4101rate35.5 vec/sbad_vectors0
202026-10-09embedembeddimensionPASS11000010.0101actual_dim768expected_dim768
212026-10-09embedembedcosinePASS11000010.6101margin0.217quality43
222026-10-09embedembedspeedPASS11000011.4101rate35.6 vec/sbad_vectors0
232026-10-09embedembeddimensionPASS11000010.1101actual_dim768expected_dim768
242026-10-09embedembedcosinePASS11000010.3101margin0.217quality43
252026-10-09embedembedspeedPASS11000010.8101rate59.1 vec/sbad_vectors0
262026-10-09embedembeddimensionPASS11000010.0101actual_dim768expected_dim768
272026-10-09embedembedcosinePASS11000010.4101margin0.217quality43
282026-10-09embedembedspeedPASS11000010.8101rate61.4 vec/sbad_vectors0
292026-10-09embedembeddimensionPASS11000010.0101actual_dim768expected_dim768
302026-10-09embedembedcosinePASS11000010.3101margin0.217quality43
312026-10-09embedembedspeedPASS11000010.9101rate57.4 vec/sbad_vectors0
322026-10-09embedembeddimensionPASS11000010.2101actual_dim768expected_dim768
332026-10-09embedembedcosinePASS11000010.3101margin0.283quality56
342026-10-09embedembedspeedPASS11000010.5101rate109.7 vec/sbad_vectors0
352026-10-09embedembeddimensionPASS11000010.0101actual_dim768expected_dim768
362026-10-09embedembedcosinePASS11000010.1101margin0.283quality56
372026-10-09embedembedspeedPASS11000010.4101rate123.3 vec/sbad_vectors0
382026-10-09embedembeddimensionPASS11000010.0101actual_dim768expected_dim768
392026-10-09embedembedcosinePASS11000010.2101margin0.283quality56
402026-10-09embedembedspeedPASS11000010.4101rate119.0 vec/sbad_vectors0