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
This commit is contained in:
@@ -29,3 +29,12 @@ date,script,model,mode,gate_status,turns,answered,caps,tool_turns,emitted,execut
|
||||
2026-10-09,embed,embed,dimension,PASS,1,1,0,0,0,0,1,0.0,1,0,1,actual_dim,768,expected_dim,768
|
||||
2026-10-09,embed,embed,cosine,PASS,1,1,0,0,0,0,1,0.3,1,0,1,margin,0.217,quality,43
|
||||
2026-10-09,embed,embed,speed,PASS,1,1,0,0,0,0,1,0.9,1,0,1,rate,57.4 vec/s,bad_vectors,0
|
||||
2026-10-09,embed,embed,dimension,PASS,1,1,0,0,0,0,1,0.2,1,0,1,actual_dim,768,expected_dim,768
|
||||
2026-10-09,embed,embed,cosine,PASS,1,1,0,0,0,0,1,0.3,1,0,1,margin,0.283,quality,56
|
||||
2026-10-09,embed,embed,speed,PASS,1,1,0,0,0,0,1,0.5,1,0,1,rate,109.7 vec/s,bad_vectors,0
|
||||
2026-10-09,embed,embed,dimension,PASS,1,1,0,0,0,0,1,0.0,1,0,1,actual_dim,768,expected_dim,768
|
||||
2026-10-09,embed,embed,cosine,PASS,1,1,0,0,0,0,1,0.1,1,0,1,margin,0.283,quality,56
|
||||
2026-10-09,embed,embed,speed,PASS,1,1,0,0,0,0,1,0.4,1,0,1,rate,123.3 vec/s,bad_vectors,0
|
||||
2026-10-09,embed,embed,dimension,PASS,1,1,0,0,0,0,1,0.0,1,0,1,actual_dim,768,expected_dim,768
|
||||
2026-10-09,embed,embed,cosine,PASS,1,1,0,0,0,0,1,0.2,1,0,1,margin,0.283,quality,56
|
||||
2026-10-09,embed,embed,speed,PASS,1,1,0,0,0,0,1,0.4,1,0,1,rate,119.0 vec/s,bad_vectors,0
|
||||
|
||||
|
Reference in New Issue
Block a user