eval-harness

model

eval-gemma-3-27b llm

A model record: onboarded once for the project and used by any eval's arms.

weights
open
licence
Gemma, gated
aliases
eval-gemma-3-27b
settings
defaults
onboarding
not onboarded: evh model-onboard
best result per goal
#10 of 30 News topic classification (AG News)
#14 of 28 Encyclopedia topic classification (DBpedia)
#21 of 33 A scorer that agrees with experts on summaries (SummEval)
#14 of 19 Evidence retrieval for scientific claims (SciFact)
#22 of 29 Short news summaries (CNN)
#24 of 28 Named-entity extraction (Few-NERD)

Every model

01deployments and their onboarding

No deployment is recorded for this model. Not onboarded yet: evh model-onboard eval-gemma-3-27b.

02settings history

No change of settings or build is recorded.

03arms

armdeploymentgoalits settingsarm smokeresult
ag_gemma_s_n200_v1 · adjudicated
ag_news_200
——max_tokens 200, temperature 0.0nonecorrect 0.960
cnn_gemma_s_n200_v1 · judged/judge-summ-pick-v1
cnn_dailymail_200
——max_tokens 1200, temperature 0.0nonejudge_won 0.002
cnn_gemma_s_n200_v1 · judged/judge-summ-v1
cnn_dailymail_200
——max_tokens 1200, temperature 0.0nonejudge_score 4.801
cnn_gemma_s_n200_v1 · judged/judge-summ-v2
cnn_dailymail_200
——max_tokens 1200, temperature 0.0nonejudge_score 4.429
cnn_gemma_s_n200_v1 · silver
cnn_dailymail_200
——max_tokens 1200, temperature 0.0nonecoverage 0.350
ag_gemma_s_n200_v1
ag_news_200
—News topic classification (AG News)max_tokens 200, temperature 0.0nonecorrect 0.875
cnn_gemma_s_n200_v1
cnn_dailymail_200
—Short news summaries (CNN)max_tokens 1200, temperature 0.0nonecoverage 0.312
db_gemma_s_n200_v1
dbpedia_280
—Encyclopedia topic classification (DBpedia)max_tokens 200, temperature 0.0nonecorrect 0.975
fn_gemma_s_n200_v1
few_nerd_280
—Named-entity extraction (Few-NERD)max_tokens 600, temperature 0.0nonef1 0.550
sf_gemma_s_n200_v1
scifact_200
—Evidence retrieval for scientific claims (SciFact)max_tokens 700, temperature 0.0nonendcg_10 0.702
se_gemma_s_n100_v1
summeval_100
—A scorer that agrees with experts on summaries (SummEval)max_tokens 1500, temperature 0.0nonetau 0.420