Skip to content

Commit 602da9f

Browse files
refactor: parameterize evaluation configuration variables and remove unused scorers
1 parent 1722276 commit 602da9f

3 files changed

Lines changed: 16 additions & 12 deletions

File tree

cloudbuild.yaml

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -32,6 +32,14 @@ steps:
3232
3333
export EVAL_GCP_PROJECT_ID=$PROJECT_ID
3434
export EVAL_GCP_PROJECT_REGION=us-central1
35+
export GOOGLE_CLOUD_PROJECT=$PROJECT_ID
36+
export CLOUD_SQL_POSTGRES_PROJECT=$PROJECT_ID
37+
export CLOUD_SQL_POSTGRES_INSTANCE=$_CLOUD_SQL_INSTANCE
38+
export CLOUD_SQL_POSTGRES_REGION=$_CLOUD_SQL_REGION
39+
export CLOUD_SQL_POSTGRES_DATABASE=$_CLOUD_SQL_DATABASE
40+
export CLOUD_SQL_POSTGRES_USER=$_CLOUD_SQL_USER
41+
export CLOUD_SQL_POSTGRES_IP_TYPE=$_CLOUD_SQL_IP_TYPE
42+
3543
# Maps the decrypted DB_PASSWORD to the exact variable expected by gemini_cli and extension skills
3644
export CLOUD_SQL_POSTGRES_PASSWORD=$$DB_PASSWORD
3745

evals/model_config.yaml

Lines changed: 8 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -15,18 +15,19 @@
1515
gemini_cli_version: "@google/gemini-cli@0.38.1"
1616
generator: gemini_cli
1717
env:
18-
GOOGLE_CLOUD_PROJECT: "ext-test-cloud-sql-postgres"
18+
GOOGLE_CLOUD_PROJECT: "${GOOGLE_CLOUD_PROJECT}"
19+
1920
GOOGLE_CLOUD_LOCATION: "global"
2021
GOOGLE_GENAI_USE_VERTEXAI: "true"
2122
setup:
2223
extensions:
2324
# Points to the symlink created in cloudbuild.yaml to match the extension ID
2425
"/workspace/cloud-sql-postgresql":
2526
settings:
26-
CLOUD_SQL_POSTGRES_PROJECT: "ext-test-cloud-sql-postgres"
27-
CLOUD_SQL_POSTGRES_INSTANCE: "daily-ci-evals-db"
28-
CLOUD_SQL_POSTGRES_REGION: "us-central1"
29-
CLOUD_SQL_POSTGRES_DATABASE: "postgres"
30-
CLOUD_SQL_POSTGRES_USER: "postgres"
27+
CLOUD_SQL_POSTGRES_PROJECT: "${CLOUD_SQL_POSTGRES_PROJECT}"
28+
CLOUD_SQL_POSTGRES_INSTANCE: "${CLOUD_SQL_POSTGRES_INSTANCE}"
29+
CLOUD_SQL_POSTGRES_REGION: "${CLOUD_SQL_POSTGRES_REGION}"
30+
CLOUD_SQL_POSTGRES_DATABASE: "${CLOUD_SQL_POSTGRES_DATABASE}"
31+
CLOUD_SQL_POSTGRES_USER: "${CLOUD_SQL_POSTGRES_USER}"
3132
CLOUD_SQL_POSTGRES_PASSWORD: '${CLOUD_SQL_POSTGRES_PASSWORD}'
32-
CLOUD_SQL_POSTGRES_IP_TYPE: "PUBLIC"
33+
CLOUD_SQL_POSTGRES_IP_TYPE: "${CLOUD_SQL_POSTGRES_IP_TYPE}"

evals/run_config.yaml

Lines changed: 0 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -23,16 +23,11 @@ model_config: /workspace/evals/model_config.yaml
2323
simulated_user_model_config: /workspace/evals/gemini_2.5_pro_model.yaml
2424

2525
scorers:
26-
# Structural
27-
trajectory_matcher: {}
28-
2926
# Qualitative (Judge-based)
3027
goal_completion:
3128
model_config: /workspace/evals/gemini_2.5_pro_model.yaml
3229
behavioral_metrics:
3330
model_config: /workspace/evals/gemini_2.5_pro_model.yaml
34-
parameter_analysis:
35-
model_config: /workspace/evals/gemini_2.5_pro_model.yaml
3631

3732
# Performance
3833
turn_count: {}

0 commit comments

Comments
 (0)