Automatic dcft datacuration framework upload for webinstruct_stage_2_1k

config/webinstruct_stage_2_1k.yaml

@@ -0,0 +1,62 @@
+name: webinstruct_stage_2_1k
+operators:
+  # - id: load_stage_1
+  #   config:
+  #     type: load_preexisting
+  #     framework_name: webinstruct_stage_1_5_10pct
+  #     load_from_uuid: c8d4905b-ccff-4a68-a1ad-9e52227c2206
+      
+  - id: load_stage_1
+    config:
+      type: load_dataset
+      # Completed stage 1 dataset
+      dataset_id: c8d4905b-ccff-4a68-a1ad-9e52227c2206
+
+  - id: sample_data_5k
+    config:
+      type: sample_data
+      num_samples: 5_000
+      max_shard: 50
+    input_ids:
+      - load_stage_1
+
+  - id: filter_dataset
+    config:
+      type: function
+      function: data_strategies.WebInstruct.utils.filter_dataset
+      function_config:
+        filter_column: "is_educational_and_frequent"
+    input_ids:
+      - sample_data_5k
+
+  - id: truncate_dataset_1k
+    config:
+      type: sample_data
+      num_samples: 1_000
+    input_ids:
+      - filter_dataset
+
+  - id: process_stage_2
+    config:
+      type: function
+      function: data_strategies.WebInstruct.utils.completions_with_remote_backup
+      function_config:
+        user_message_column: text
+        output_column: extracted_text
+        system_message: |- 
+          You are given a set of pre-processed documents, each of which may contain natural question-answer (Q-A) pairs. Your task is to identify and extract these pairs while ignoring unrelated content such as ads, markup, or boilerplate text.
+          Input: Each document contains multiple sections of text. Some of these sections may have clear questions followed by answers, while others may be irrelevant (e.g., ads or noise).
+          Output: Extract the Q-A pairs found within each document. A valid Q-A pair must consist of a clearly defined question and its corresponding answer. If no natural Q-A pair exists in the document, return only the word "void" for that document.
+        reannotation: false
+        gcs_backup_path: gs://dcft-data-gcp/backups/webinstruct_stage_2_1k
+    input_ids:
+      - truncate_dataset_1k
+
+  - id: drop_void
+    config:
+      type: function
+      function: data_strategies.WebInstruct.utils.drop_void
+      function_config:
+        input_column: extracted_text
+    input_ids:
+      - process_stage_2