crfm-helm 0.4.0__py3-none-any.whl → 0.5.10__py3-none-any.whl

helm/benchmark/static/schema_sql.yaml

@@ -0,0 +1,171 @@
+---
+############################################################
+metrics:
+  # Infrastructure metrics:
+  - name: num_perplexity_tokens
+    display_name: '# tokens'
+    description: Average number of tokens in the predicted output (for language modeling, the input too).
+  - name: num_bytes
+    display_name: '# bytes'
+    description: Average number of bytes in the predicted output (for language modeling, the input too).
+
+  - name: num_references
+    display_name: '# ref'
+    description: Number of references.
+  - name: num_train_trials
+    display_name: '# trials'
+    description: Number of trials, where in each trial we choose an independent, random set of training instances.
+  - name: estimated_num_tokens_cost
+    display_name: 'cost'
+    description: An estimate of the number of tokens (including prompt and output completions) needed to perform the request.
+  - name: num_prompt_tokens
+    display_name: '# prompt tokens'
+    description: Number of tokens in the prompt.
+  - name: num_prompt_characters
+    display_name: '# prompt chars'
+    description: Number of characters in the prompt.
+  - name: num_completion_tokens
+    display_name: '# completion tokens'
+    description: Actual number of completion tokens (over all completions).
+  - name: num_output_tokens
+    display_name: '# output tokens'
+    description: Actual number of output tokens.
+  - name: max_num_output_tokens
+    display_name: 'Max output tokens'
+    description: Maximum number of output tokens (overestimate since we might stop earlier due to stop sequences).
+  - name: num_requests
+    display_name: '# requests'
+    description: Number of distinct API requests.
+  - name: num_instances
+    display_name: '# eval'
+    description: Number of evaluation instances.
+  - name: num_train_instances
+    display_name: '# train'
+    description: Number of training instances (e.g., in-context examples).
+  - name: prompt_truncated
+    display_name: truncated
+    description: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).
+  - name: finish_reason_length
+    display_name: finish b/c length
+    description: Fraction of instances where the the output was terminated because of the max tokens limit.
+  - name: finish_reason_stop
+    display_name: finish b/c stop
+    description: Fraction of instances where the the output was terminated because of the stop sequences.
+  - name: finish_reason_endoftext
+    display_name: finish b/c endoftext
+    description: Fraction of instances where the the output was terminated because the end of text token was generated.
+  - name: finish_reason_unknown
+    display_name: finish b/c unknown
+    description: Fraction of instances where the the output was terminated for unknown reasons.
+  - name: num_completions
+    display_name: '# completions'
+    description: Number of completions.
+  - name: predicted_index
+    display_name: Predicted index
+    description: Integer index of the reference (0, 1, ...) that was predicted by the model (for multiple-choice).
+
+  # Accuracy metrics:
+  - name: program_accuracy
+    display_name: Program Accuracy
+    description: Accuracy of the generated programs
+    lower_is_better: false
+  - name: execution_accuracy
+    display_name: Execution Accuracy
+    description: Accuracy of the final result of the generated program
+    lower_is_better: false
+  - name: annotation_financebench_label_correct_answer
+    display_name: Correct Answer
+    description: Whether the final result was correct, as judged by a GPT-4o
+    lower_is_better: false
+  - name: quasi_exact_match
+    display_name: Quasi-exact match
+    short_display_name: EM
+    description: Fraction of instances that the predicted output matches a correct reference up to light processing.
+    lower_is_better: false
+  - name: error_rate
+    display_name: SQL Error Rate
+    short_display_name: SQL Error Rate
+    description: Fraction of generated queries that result in a SQL execution error
+    lower_is_better: true
+  - name: execution_accuracy
+    display_name: Execution Accuracy
+    short_display_name: Execution Accuracy
+    description: Execution Accuracy
+    lower_is_better: false
+
+############################################################
+perturbations: []
+
+############################################################
+metric_groups:
+  - name: accuracy
+    display_name: Accuracy
+    hide_win_rates: true
+    metrics:
+      - name: ${main_name}
+        split: ${main_split}
+
+  - name: efficiency
+    display_name: Efficiency
+    metrics:
+    - name: inference_runtime
+      split: ${main_split}
+
+  - name: general_information
+    display_name: General information
+    hide_win_rates: true
+    metrics:
+    - name: num_instances
+      split: ${main_split}
+    - name: num_train_instances
+      split: ${main_split}
+    - name: prompt_truncated
+      split: ${main_split}
+    - name: num_prompt_tokens
+      split: ${main_split}
+    - name: num_output_tokens
+      split: ${main_split}
+
+############################################################
+run_groups:
+  - name: text_to_sql_scenarios
+    display_name: Text-to-SQL Scenarios
+    description: Text-to-SQL Scenarios
+    category: All scenarios
+    subgroups:
+      - spider
+      - bird_sql
+
+  - name: spider
+    display_name: Spider 1.0 (Test)
+    description: Spider 1.0 (Test)
+    metric_groups:
+      - accuracy
+      - efficiency
+      - general_information
+    environment:
+      main_name: execution_accuracy
+      main_split: valid
+    taxonomy:
+      task: text-to-SQL
+      what: databases from various domains
+      who: expert data scientists
+      when: "?"
+      language: English
+
+  - name: bird_sql
+    display_name: BIRD-SQL (Dev)
+    description: BIRD-SQL (Dev)
+    metric_groups:
+      - accuracy
+      - efficiency
+      - general_information
+    environment:
+      main_name: execution_accuracy
+      main_split: valid
+    taxonomy:
+      task: text-to-SQL
+      what: databases from various domains
+      who: expert data scientists
+      when: "?"
+      language: English

helm/benchmark/static/schema_thai.yaml

@@ -0,0 +1,244 @@
+---
+############################################################
+metrics:
+  # Infrastructure metrics:
+  - name: num_perplexity_tokens
+    display_name: '# tokens'
+    description: Average number of tokens in the predicted output (for language modeling, the input too).
+  - name: num_bytes
+    display_name: '# bytes'
+    description: Average number of bytes in the predicted output (for language modeling, the input too).
+
+  - name: num_references
+    display_name: '# ref'
+    description: Number of references.
+  - name: num_train_trials
+    display_name: '# trials'
+    description: Number of trials, where in each trial we choose an independent, random set of training instances.
+  - name: estimated_num_tokens_cost
+    display_name: 'cost'
+    description: An estimate of the number of tokens (including prompt and output completions) needed to perform the request.
+  - name: num_prompt_tokens
+    display_name: '# prompt tokens'
+    description: Number of tokens in the prompt.
+  - name: num_prompt_characters
+    display_name: '# prompt chars'
+    description: Number of characters in the prompt.
+  - name: num_completion_tokens
+    display_name: '# completion tokens'
+    description: Actual number of completion tokens (over all completions).
+  - name: num_output_tokens
+    display_name: '# output tokens'
+    description: Actual number of output tokens.
+  - name: max_num_output_tokens
+    display_name: 'Max output tokens'
+    description: Maximum number of output tokens (overestimate since we might stop earlier due to stop sequences).
+  - name: num_requests
+    display_name: '# requests'
+    description: Number of distinct API requests.
+  - name: num_instances
+    display_name: '# eval'
+    description: Number of evaluation instances.
+  - name: num_train_instances
+    display_name: '# train'
+    description: Number of training instances (e.g., in-context examples).
+  - name: prompt_truncated
+    display_name: truncated
+    description: Fraction of instances where the prompt itself was truncated (implies that there were no in-context examples).
+  - name: finish_reason_length
+    display_name: finish b/c length
+    description: Fraction of instances where the the output was terminated because of the max tokens limit.
+  - name: finish_reason_stop
+    display_name: finish b/c stop
+    description: Fraction of instances where the the output was terminated because of the stop sequences.
+  - name: finish_reason_endoftext
+    display_name: finish b/c endoftext
+    description: Fraction of instances where the the output was terminated because the end of text token was generated.
+  - name: finish_reason_unknown
+    display_name: finish b/c unknown
+    description: Fraction of instances where the the output was terminated for unknown reasons.
+  - name: num_completions
+    display_name: '# completions'
+    description: Number of completions.
+  - name: predicted_index
+    display_name: Predicted index
+    description: Integer index of the reference (0, 1, ...) that was predicted by the model (for multiple-choice).
+
+  # Accuracy metrics:
+  - name: exact_match
+    display_name: Exact match
+    short_display_name: EM
+    description: Fraction of instances that the predicted output matches a correct reference exactly.
+    lower_is_better: false
+
+############################################################
+perturbations: []
+
+############################################################
+metric_groups:
+  - name: accuracy
+    display_name: Accuracy
+    hide_win_rates: true
+    metrics:
+      - name: ${main_name}
+        split: ${main_split}
+
+  - name: efficiency
+    display_name: Efficiency
+    metrics:
+    - name: inference_runtime
+      split: ${main_split}
+
+  - name: general_information
+    display_name: General information
+    hide_win_rates: true
+    metrics:
+    - name: num_instances
+      split: ${main_split}
+    - name: num_train_instances
+      split: ${main_split}
+    - name: prompt_truncated
+      split: ${main_split}
+    - name: num_prompt_tokens
+      split: ${main_split}
+    - name: num_output_tokens
+      split: ${main_split}
+
+############################################################
+
+run_groups:
+  - name: thai_scenarios
+    display_name: Thai Scenarios
+    description: Thai-language scenarios
+    category: All scenarios
+    subgroups:
+      - thai_exam
+      - thai_exam_onet
+      - thai_exam_ic
+      - thai_exam_tgat
+      - thai_exam_tpat1
+      - thai_exam_a_level
+
+
+  - name: thai_exam
+    display_name: ThaiExam
+    description: >
+      Macro-averaged accuracy on all ThaiExam examinations.
+    metric_groups:
+      - accuracy
+      - efficiency
+      - general_information
+    environment:
+      main_name: exact_match
+      main_split: test
+    taxonomy:
+      task: question answering
+      what: n/a
+      who: n/a
+      when: "?"
+      language: Thai and English
+
+  - name: thai_exam_onet
+    display_name: ONET
+    description: >
+      The Ordinary National Educational Test (ONET) is an examination for students in Thailand.
+      We select the grade-12 ONET exam, which comprises 5 subjects and each question has 5 choices.
+      These subjects are Thai, English, Mathematics, Social Studies, and Science.
+      Amounting to a total of 170 questions and options.
+    metric_groups:
+      - accuracy
+      - efficiency
+      - general_information
+    environment:
+      main_name: exact_match
+      main_split: test
+    taxonomy:
+      task: question answering
+      what: high school / medical school academic knowledge
+      who: n/a
+      when: "?"
+      language: Thai and English
+
+  - name: thai_exam_ic
+    display_name: IC
+    description: >
+      The Investment Consultant (IC) examination, a licensing test for investment professionals in Thailand.
+      Developed by the Stock Exchange of Thailand (SET), features 4 choices per question.
+      We extracted questions for levels 1, 2, and 3 resulting in a total of 95 questions and options.
+    metric_groups:
+      - accuracy
+      - efficiency
+      - general_information
+    environment:
+      main_name: exact_match
+      main_split: test
+    taxonomy:
+      task: question answering
+      what: licensing for investment professionals
+      who: n/a
+      when: "?"
+      language: Thai
+
+  - name: thai_exam_tgat
+    display_name: TGAT
+    description: >
+      The Thai General Aptitude Test (TGAT), a national high school examination in Thailand.
+      Focuses on critical and logical thinking skills.
+      We collected a total of 90 questions and answers. The TGAT consists of four choices per question.
+    metric_groups:
+      - accuracy
+      - efficiency
+      - general_information
+    environment:
+      main_name: exact_match
+      main_split: test
+    taxonomy:
+      task: question answering
+      what: high school level questions on reasoning
+      who: n/a
+      when: "?"
+      language: English
+
+  - name: thai_exam_tpat1
+    display_name: TPAT-1
+    description: >
+      The Thai Professional Aptitude Test 1 (TPAT-1) is a national high school examination in Thailand.
+      The Exam assesses students’ professional skills requirement in medical schools.
+      This subset contains reasoning and medical ethics. We collected a total of 116 questions and answers.
+      The TPAT-1 consists of 5 choices per question.
+    description: TBD
+    metric_groups:
+      - accuracy
+      - efficiency
+      - general_information
+    environment:
+      main_name: exact_match
+      main_split: test
+    taxonomy:
+      task: question answering
+      what: high school / medical school academic knowledge
+      who: n/a
+      when: "?"
+      language: Thai
+
+  - name: thai_exam_a_level
+    display_name: A-Level
+    description: >
+      An academic knowledge assessment examination (Applied Knowledge Level)
+      that covers general foundational subjects taught in schools.
+      The content assessed in this examination aligns with the curriculum guidelines
+      and emphasizes the practical application of knowledge in daily life.
+      We collected a total of 175 questions and answers.
+    metric_groups:
+      - accuracy
+      - efficiency
+      - general_information
+    environment:
+      main_name: exact_match
+      main_split: test
+    taxonomy:
+      task: question answering
+      what: high school academic knowledge
+      who: n/a
+      when: "?"
+      language: Thai and English