Automatic dcft datacuration framework upload for webinstruct_stage1_scale_up

config/webinstruct_stage1_scale_up.yaml

@@ -0,0 +1,46 @@
+name: webinstruct_stage1_scale_up
+operators:
+  - id: sample_dclm_refined_web_shards
+    config:
+      type: dclm_refinedweb_source
+      num_shards: 1
+      s3_bucket: commoncrawl
+      s3_prefix: contrib/datacomp/DCLM-refinedweb/
+      seed: 314 # different shard than used for training with default seed 42
+      
+  - id: clean_dclm_refinedweb_docs
+    config:
+      type: function
+      function: data_strategies.WebInstruct.utils.clean_dataset
+      function_config:
+        input_column: text
+        output_column: cleaned_text
+        remove_columns: 
+          - bff_contained_ngram_count_before_dedupe
+          - language_id_whole_page_fasttext
+          - metadata
+          - previous_word_count
+          - warcinfo
+    input_ids:
+      - sample_dclm_refined_web_shards
+
+  - id: classify_fasttext
+    config:
+      type: fasttext
+      fasttext_url: https://huggingface.co/mlfoundations-dev/fasttext-qa-v5/resolve/main/fasttext_classifier_v5.bin
+      input_column: cleaned_text
+      target_label: "__label__QA_doc"
+      filter_threshold: 0.5
+    input_ids:
+      - clean_dclm_refinedweb_docs
+
+  - id: classify_url
+    config:
+      type: function
+      function: data_strategies.WebInstruct.utils.url_filter_fast
+      function_config:
+        input_column: url
+        job_path: dcft/data_strategies/WebInstruct/cache/parallel/
+        filter: True
+    input_ids:
+      - classify_fasttext