crfm-helm 0.5.7__py3-none-any.whl → 0.5.9__py3-none-any.whl

helm/benchmark/static/schema_slp.yaml

@@ -0,0 +1,219 @@
+############################################################
+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
+  - name: classification_macro_f1
+    display_name: Macro-F1
+    description: Population-level macro-averaged F1 score.
+    lower_is_better: false
+  - name: classification_micro_f1
+    display_name: Micro-F1
+    description: Population-level micro-averaged F1 score.
+    lower_is_better: false
+  - name: wer_score
+    display_name: Word Error Rate
+    description: Transcription error rate.
+    lower_is_better: true
+  - name: mer_score
+    display_name: Character Error Rate
+    description: Character error rate.
+    lower_is_better: true
+
+############################################################
+perturbations: []
+
+############################################################
+metric_groups:
+  - name: accuracy
+    display_name: Accuracy
+    hide_win_rates: true
+    metrics:
+      - name: exact_match
+        split: ${main_split}
+      - name: classification_macro_f1
+        split: ${main_split}
+      - name: classification_micro_f1
+        split: ${main_split}
+
+  - name: transcription_accuracy
+    display_name: Transcription Accuracy
+    hide_win_rates: true
+    metrics:
+      - name: wer_score
+        split: ${main_split}
+      - name: mer_score
+        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: slp
+    display_name: SLP Scenarios
+    description: SLP-language scenarios
+    category: All scenarios
+    subgroups:
+      - disorder_diagnosis
+      - transcription
+      - symptom_diagnosis
+      - disorder_type_diagnosis
+
+
+  - name: disorder_diagnosis
+    display_name: Disorder Diagnosis Accuracy
+    description: >
+      Macro-averaged accuracy on disorder diagnosis for pediatric speech disorder.
+    metric_groups:
+      - accuracy
+      - efficiency
+      - general_information
+    environment:
+      main_name: classification_micro_f1
+      main_split: test
+    taxonomy:
+      task: classification
+      what: n/a
+      who: n/a
+      when: "?"
+      language: English
+
+  - name: transcription
+    display_name: Transcription Accuracy
+    description: >
+      Model transcription accuracy on understanding disordered pediatric speech
+    metric_groups:
+      - transcription_accuracy
+      - efficiency
+      - general_information
+    environment:
+      main_name: wer_score
+      main_split: test
+    taxonomy:
+      task: transcription
+      what: disordered pediatric speech
+      who: n/a
+      when: "?"
+      language: English
+
+  - name: symptom_diagnosis
+    display_name: Symptom Diagnosis Accuracy
+    description: >
+      Macro-averaged accuracy on symptom diagnosis for pediatric speech disorder.
+    metric_groups:
+      - accuracy
+      - efficiency
+      - general_information
+    environment:
+      main_name: classification_micro_f1
+      main_split: test
+    taxonomy:
+      task: classification
+      what: n/a
+      who: n/a
+      when: "?"
+      language: English
+
+  - name: disorder_type_diagnosis
+    display_name: Disorder Type Diagnosis Accuracy
+    description: >
+      Macro-averaged accuracy on disorder type diagnosis for pediatric speech disorder.
+    metric_groups:
+      - accuracy
+      - efficiency
+      - general_information
+    environment:
+      main_name: classification_micro_f1
+      main_split: test
+    taxonomy:
+      task: classification
+      what: n/a
+      who: n/a
+      when: "?"
+      language: English