gemma-e2b-rlcd / reports /demo-workload-web-validation.json
larkooo's picture
Publish Gemma E2B RLCD with multimodal checkpoint and parallel scoring
53e24ca verified
Raw History Blame Contribute Delete
1.07 kB
{
"scope": "Browser integration checks after installing the workload examples. Timings are individual observations, not pooled with the repeated benchmark.",
"presets_checked": {
"support_matrix": 28,
"inbox_matrix": 32,
"ticket_flags": 32,
"catalog_choices": 4
},
"initial_check": {
"complete": true,
"comparison_rows": 28,
"different_fields": [
"public_status_page_incident",
"assigned_agent_tier"
],
"parallel_seconds": 2.456,
"normal_seconds": 12.004,
"issue": "Browser editor materialized different default yes/no meanings than the benchmark."
},
"fix": "Presets now include the core default yes/no meanings explicitly. A regression test verifies the complete generated task matches each benchmark case after editor-style serialization.",
"final_check": {
"complete": true,
"comparison_rows": 28,
"different_fields": [],
"parallel_seconds": 3.407,
"normal_seconds": 8.573,
"console_errors": []
},
"tests": "89 passed; lint and JavaScript syntax checks passed."
}