skills/summarization/evaluation/tests.yaml (162 lines of code) (raw):
- vars:
input: file://../data/sample-lease1.txt
ground_truth: file://../data/sample-lease1-summary.txt
assert:
- type: icontains-all
value:
- 'parties involved'
- 'property details'
- 'term and rent'
- 'responsibilities'
- 'consent and notices'
- 'special provisions'
- type: python
value: file://custom_evals/llm_eval.py
- type: python
value: file://custom_evals/rouge_eval.py
- type: python
value: file://custom_evals/bleu_eval.py
- vars:
input: file://../data/sample-lease2.txt
ground_truth: file://../data/sample-lease2-summary.txt
assert:
- type: icontains-all
value:
- 'parties involved'
- 'property details'
- 'term and rent'
- 'responsibilities'
- 'consent and notices'
- 'special provisions'
- type: python
value: file://custom_evals/llm_eval.py
- type: python
value: file://custom_evals/rouge_eval.py
- type: python
value: file://custom_evals/bleu_eval.py
- vars:
input: file://../data/sample-lease3.txt
ground_truth: file://../data/sample-lease3-summary.txt
assert:
- type: icontains-all
value:
- 'parties involved'
- 'property details'
- 'term and rent'
- 'responsibilities'
- 'consent and notices'
- 'special provisions'
- type: python
value: file://custom_evals/llm_eval.py
- type: python
value: file://custom_evals/rouge_eval.py
- type: python
value: file://custom_evals/bleu_eval.py
- vars:
input: file://../data/sample-lease4.txt
ground_truth: file://../data/sample-lease4-summary.txt
assert:
- type: icontains-all
value:
- 'parties involved'
- 'property details'
- 'term and rent'
- 'responsibilities'
- 'consent and notices'
- 'special provisions'
- type: python
value: file://custom_evals/llm_eval.py
- type: python
value: file://custom_evals/rouge_eval.py
- type: python
value: file://custom_evals/bleu_eval.py
- vars:
input: file://../data/sample-lease5.txt
ground_truth: file://../data/sample-lease5-summary.txt
assert:
- type: icontains-all
value:
- 'parties involved'
- 'property details'
- 'term and rent'
- 'responsibilities'
- 'consent and notices'
- 'special provisions'
- type: python
value: file://custom_evals/llm_eval.py
- type: python
value: file://custom_evals/rouge_eval.py
- type: python
value: file://custom_evals/bleu_eval.py
- vars:
input: file://../data/sample-lease6.txt
ground_truth: file://../data/sample-lease6-summary.txt
assert:
- type: icontains-all
value:
- 'parties involved'
- 'property details'
- 'term and rent'
- 'responsibilities'
- 'consent and notices'
- 'special provisions'
- type: python
value: file://custom_evals/llm_eval.py
- type: python
value: file://custom_evals/rouge_eval.py
- type: python
value: file://custom_evals/bleu_eval.py
- vars:
input: file://../data/sample-lease7.txt
ground_truth: file://../data/sample-lease7-summary.txt
assert:
- type: icontains-all
value:
- 'parties involved'
- 'property details'
- 'term and rent'
- 'responsibilities'
- 'consent and notices'
- 'special provisions'
- type: python
value: file://custom_evals/llm_eval.py
- type: python
value: file://custom_evals/rouge_eval.py
- type: python
value: file://custom_evals/bleu_eval.py
- vars:
input: file://../data/sample-lease8.txt
ground_truth: file://../data/sample-lease8-summary.txt
assert:
- type: icontains-all
value:
- 'parties involved'
- 'property details'
- 'term and rent'
- 'responsibilities'
- 'consent and notices'
- 'special provisions'
- type: python
value: file://custom_evals/llm_eval.py
- type: python
value: file://custom_evals/rouge_eval.py
- type: python
value: file://custom_evals/bleu_eval.py
- vars:
input: file://../data/sample-lease9.txt
ground_truth: file://../data/sample-lease9-summary.txt
assert:
- type: icontains-all
value:
- 'parties involved'
- 'property details'
- 'term and rent'
- 'responsibilities'
- 'consent and notices'
- 'special provisions'
- type: python
value: file://custom_evals/llm_eval.py
- type: python
value: file://custom_evals/rouge_eval.py
- type: python
value: file://custom_evals/bleu_eval.py