Commit 97e3c9fe authored by lintangsutawika's avatar lintangsutawika
Browse files

added lama primed and negated dataset

parent 7852985b
include: _negated_true_false_template_yaml
task: lama_negated_squad_true_false
dataset_name: SQUAD
test_split: high_ranked
include: _negated_true_false_template_yaml
task: lama_negated_trex_true_false
dataset_name: TREx
test_split: high_ranked
include: _normal_true_false_template_yaml
task: lama_normal_conceptnet_true_false
dataset_name: ConceptNet
test_split: high_ranked
include: _normal_true_false_template_yaml
task: lama_normal_google_re_true_false
dataset_name: GoogleRE
test_split: high_ranked
include: _normal_true_false_template_yaml
task: lama_normal_squad_true_false
dataset_name: SQUAD
test_split: high_ranked
include: _normal_true_false_template_yaml
task: lama_normal_trex_true_false
dataset_name: TREx
test_split: high_ranked
group: lama_misprimed_yes_no
dataset_path: lintang/lama_primed_negated
output_type: multiple_choice
# description: "Answer the following question with either a Yes or a No. "
doc_to_text: "Question: {{masked_misprimed[0]|replace('[MASK]', obj_label)}} Yes or No?\n\nAnswer:"
doc_to_target: "Yes"
doc_to_choice: ["Yes", "No"]
should_decontaminate: true
doc_to_decontamination_query: "Question: {{masked_misprimed|replace('[MASK]', obj_label)}} Yes or No?\n\nAnswer:"
metric_list:
- metric: acc
aggregation: mean
higher_is_better: true
- metric: acc_norm
aggregation: mean
higher_is_better: true
metadata:
version: 1.0
group: lama_negated_yes_no
dataset_path: lintang/lama_primed_negated
output_type: multiple_choice
# description: "Answer the following question with either a Yes or a No. "
doc_to_text: "Question: {{masked_negations[0]|replace('[MASK]', obj_label)}} Yes or No?\n\nAnswer:"
doc_to_target: "No"
doc_to_choice: ["Yes", "No"]
should_decontaminate: true
doc_to_decontamination_query: "Question: {{masked_negations|replace('[MASK]', obj_label)}} Yes or No?\n\nAnswer:"
metric_list:
- metric: acc
aggregation: mean
higher_is_better: true
- metric: acc_norm
aggregation: mean
higher_is_better: true
metadata:
version: 1.0
group: lama_normal_yes_no
dataset_path: lintang/lama_primed_negated
output_type: multiple_choice
# description: "Answer the following question with either a Yes or a No. "
doc_to_text: "Question: {{masked_sentences[0]|replace('[MASK]', obj_label)}} Yes or No?\n\nAnswer:"
doc_to_target: "Yes"
doc_to_choice: ["Yes", "No"]
should_decontaminate: true
doc_to_decontamination_query: "Question: {{masked_sentences[0]|replace('[MASK]', obj_label)}} Yes or No?\n\nAnswer:"
metric_list:
- metric: acc
aggregation: mean
higher_is_better: true
- metric: acc_norm
aggregation: mean
higher_is_better: true
metadata:
version: 1.0
include: _misprimed_yes_no_template_yaml
task: lama_misprimed_conceptnet_yes_no
dataset_name: ConceptNet
test_split: high_ranked
include: _misprimed_yes_no_template_yaml
task: lama_misprimed_google_re_yes_no
dataset_name: GoogleRE
test_split: high_ranked
include: _misprimed_yes_no_template_yaml
task: lama_misprimed_squad_yes_no
dataset_name: SQUAD
test_split: high_ranked
include: _misprimed_yes_no_template_yaml
task: lama_misprimed_trex_yes_no
dataset_name: TREx
test_split: high_ranked
include: _negated_yes_no_template_yaml
task: lama_negated_conceptnet_yes_no
dataset_name: ConceptNet
test_split: high_ranked
include: _negated_yes_no_template_yaml
task: lama_negated_google_re_yes_no
dataset_name: GoogleRE
test_split: high_ranked
include: _negated_yes_no_template_yaml
task: lama_negated_squad_yes_no
dataset_name: SQUAD
test_split: high_ranked
include: _negated_yes_no_template_yaml
task: lama_negated_trex_yes_no
dataset_name: TREx
test_split: high_ranked
include: _normal_yes_no_template_yaml
task: lama_normal_conceptnet_yes_no
dataset_name: ConceptNet
test_split: high_ranked
include: _normal_yes_no_template_yaml
task: lama_normal_google_re_yes_no
dataset_name: GoogleRE
test_split: high_ranked
include: _normal_yes_no_template_yaml
task: lama_normal_squad_yes_no
dataset_name: SQUAD
test_split: high_ranked
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment