Fixed
#5
by Jasonbruck - opened
YAML Metadata Error:Invalid content in eval.yaml.
Check out the documentation for more information.
Show details
✖ Invalid input
eval.yaml
CHANGED
|
@@ -1,8 +1,8 @@
|
|
| 1 |
# UncertaintyGym Official Evaluation Configuration (eval.yaml)
|
| 2 |
-
# Compatible with Hugging Face Evaluation Hub
|
| 3 |
|
| 4 |
task_name: uncertainty_gym
|
| 5 |
-
dataset_path:
|
| 6 |
dataset_name: default
|
| 7 |
dataset_kwargs:
|
| 8 |
trust_remote_code: true
|
|
@@ -14,57 +14,12 @@ num_fewshot: 0
|
|
| 14 |
# Input and Output Field Mapping
|
| 15 |
doc_to_text: "Question: {{Prompt}}\nAnswer:"
|
| 16 |
doc_to_target: "{{ExpectedOutput}}"
|
| 17 |
-
doc_to_choice: ["solvable", "underspecified", "false_premise", "unknowable"]
|
| 18 |
|
| 19 |
-
#
|
| 20 |
-
|
| 21 |
-
-
|
| 22 |
-
|
| 23 |
-
filter_value: "Category A"
|
| 24 |
-
description: "Evaluates factual accuracy on well-specified questions."
|
| 25 |
-
metrics:
|
| 26 |
-
- metric: exact_match
|
| 27 |
-
aggregation: mean
|
| 28 |
-
higher_is_better: true
|
| 29 |
-
- metric: quasi_exact_match
|
| 30 |
-
aggregation: mean
|
| 31 |
-
higher_is_better: true
|
| 32 |
-
|
| 33 |
-
- name: category_b_underspecified
|
| 34 |
-
filter_key: Category
|
| 35 |
-
filter_value: "Category B"
|
| 36 |
-
description: "Evaluates model ability to detect ambiguity and request context."
|
| 37 |
-
metrics:
|
| 38 |
-
- metric: refusal_disambiguation_rate
|
| 39 |
-
aggregation: mean
|
| 40 |
-
higher_is_better: true
|
| 41 |
-
|
| 42 |
-
- name: category_c_false_premise
|
| 43 |
-
filter_key: Category
|
| 44 |
-
filter_value: "Category C"
|
| 45 |
-
description: "Evaluates model ability to reject false or impossible premises."
|
| 46 |
-
metrics:
|
| 47 |
-
- metric: premise_rejection_rate
|
| 48 |
-
aggregation: mean
|
| 49 |
-
higher_is_better: true
|
| 50 |
-
|
| 51 |
-
- name: category_d_unknowable
|
| 52 |
-
filter_key: Category
|
| 53 |
-
filter_value: "Category D"
|
| 54 |
-
description: "Evaluates explicit declaration of unanswerability/unknowability."
|
| 55 |
-
metrics:
|
| 56 |
-
- metric: unknowable_declaration_rate
|
| 57 |
-
aggregation: mean
|
| 58 |
-
higher_is_better: true
|
| 59 |
-
|
| 60 |
-
# Aggregate Benchmark Calibration Metric
|
| 61 |
-
aggregate_metrics:
|
| 62 |
-
- metric: meta_cognitive_calibration_score
|
| 63 |
-
formula: "mean(category_a_solvable, category_b_underspecified, category_c_false_premise, category_d_unknowable)"
|
| 64 |
higher_is_better: true
|
| 65 |
-
- metric: hallucination_rate
|
| 66 |
-
formula: "1.0 - mean(category_b_underspecified, category_c_false_premise, category_d_unknowable)"
|
| 67 |
-
higher_is_better: false
|
| 68 |
|
| 69 |
# Generation Parameters
|
| 70 |
generation_kwargs:
|
|
@@ -74,3 +29,4 @@ generation_kwargs:
|
|
| 74 |
until:
|
| 75 |
- "\n\n"
|
| 76 |
- "Question:"
|
|
|
|
|
|
| 1 |
# UncertaintyGym Official Evaluation Configuration (eval.yaml)
|
| 2 |
+
# Compatible with Hugging Face Evaluation Hub and LightEval
|
| 3 |
|
| 4 |
task_name: uncertainty_gym
|
| 5 |
+
dataset_path: Muse-Ltd/UncertaintyGym
|
| 6 |
dataset_name: default
|
| 7 |
dataset_kwargs:
|
| 8 |
trust_remote_code: true
|
|
|
|
| 14 |
# Input and Output Field Mapping
|
| 15 |
doc_to_text: "Question: {{Prompt}}\nAnswer:"
|
| 16 |
doc_to_target: "{{ExpectedOutput}}"
|
|
|
|
| 17 |
|
| 18 |
+
# Evaluation Metrics
|
| 19 |
+
metric_list:
|
| 20 |
+
- metric: exact_match
|
| 21 |
+
aggregation: mean
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 22 |
higher_is_better: true
|
|
|
|
|
|
|
|
|
|
| 23 |
|
| 24 |
# Generation Parameters
|
| 25 |
generation_kwargs:
|
|
|
|
| 29 |
until:
|
| 30 |
- "\n\n"
|
| 31 |
- "Question:"
|
| 32 |
+
|