Upload configuration file

config/filtered_pdf_flow.yaml

@@ -0,0 +1,75 @@
+operators:
+  - id: load_seed_code
+    config:
+      type: load_preexisting
+      framework_name: raw_parsed_pdfs_for_verifiable_qa
+  - id: filtered_qas
+    config:
+      shard: True
+      num_shards: 1
+      type: function
+      function: data_strategies.32b_exploit.pdf_helper.filter_valid_qas
+    input_ids:
+      - load_seed_code
+  - id: validate_questions
+    config:
+      type: completions
+      model: gpt-4o-mini
+      map: binary_classifier
+      map_config:
+        user_message: |
+          Yes or No, is the question an answerable difficult science, math, or coding question? If the question refers to content (figures, equations, additional text) that you cannot see, then it is unanswerable.
+          Provide your reasoning.
+          Question:
+          {{improved_question_solution}}
+        output_column: qa_validation_outputs
+      temperature: 0.0  # From stage 1
+      top_p: 1.0
+      presence_penalty: 0.0
+    input_ids:
+      - filtered_qas
+
+  - id: filter_by_validation_output
+    config:
+      type: function
+      function: data_strategies.commons.select_rows
+      function_config:
+        conditions:
+          qa_validation_outputs: True
+    input_ids:
+      - validate_questions
+  - id: fuzzy_dedup
+    config:
+      type: cpu_function
+      num_cpus: 80
+      function: data_strategies.DeepSeekR1.utils.fuzzy_dedup
+      function_config:
+        column: improved_question_solution
+        similarity_threshold: 95.0
+    input_ids:
+      -  filter_by_validation_output
+  - id: decontaminate
+    config:
+      type: cpu_function
+      num_cpus: 80
+      function:  data_strategies.DeepSeekR1.utils.decontaminate_table
+      function_config:
+        column: improved_question_solution
+        eval_datasets:
+          - HuggingFaceH4/MATH-500
+          - Maxwell-Jia/AIME_2024
+          - AI-MO/aimo-validation-amc
+          - livecodebench/code_generation_lite
+        eval_columns:
+          - problem
+          - Problem
+          - problem
+          - question_content
+        eval_splits:
+          - test
+          - train
+          - train
+          - test
+        similarity_threshold: 95.0
+    input_ids:
+      - fuzzy_dedup