Skip to content

added SBERT and LLM as judge - #5

Open
JamesVillanueva-Dev wants to merge 4 commits into
mainfrom
stances
Open

added SBERT and LLM as judge#5
JamesVillanueva-Dev wants to merge 4 commits into
mainfrom
stances

Conversation

@JamesVillanueva-Dev

Copy link
Copy Markdown
Collaborator

SBERT seems to work and LLM uses OPENAI as the judge.

jjdrisco pushed a commit that referenced this pull request Jun 8, 2026
…egrate PR #5

Phase 1: Extract DeBERTaV3 into modular structure
- Move data models to utils/data_models.py
- Extract DeBERTaV3 backend to backends/deberta.py
- Extract scoring logic to scoring/deberta_scoring.py
- Extract reporting to reporting/deberta_report.py
- Remove monolithic metrics.py and reporting.py

Phase 2: Implement StanceAware backend from PR #5
- Add StanceAwareEmbedder to backends/stanceaware.py
- Add StanceAware scoring to scoring/stanceaware_scoring.py
- Add StanceAware reporting to reporting/stanceaware_report.py

Phase 3: Implement LLM Judge backend from PR #5
- Add OpenAIJudge to backends/llm_judge.py
- Add LLM judge scoring to scoring/llm_judge_scoring.py
- Add LLM judge reporting to reporting/llm_judge_report.py

Phase 4: Multi-evaluator CLI orchestration
- Update cli.py to support --evaluators flag
- Run evaluators independently with separate output directories
- Create summary.txt with status of each evaluator

Benefits:
- Modular: Each evaluator independent, can add more easily
- Scalable: Follows same pattern as turn_taking_length
- Backward compatible: Output identical to original implementation
- Flexible: CLI supports selective evaluator execution

All existing tests pass with updated imports.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant