Publish transformation report

validation_report.json

@@ -2,19 +2,29 @@
   "source_dataset": "facebook/community-alignment-dataset",
   "source_revision": "97343c7f6399fcbea430ed0f37c1768281a78d56",
   "source_rows": 90256,
-  "published_rows": 83193,
+  "published_rows": 22067,
   "skipped_rows_without_followup": 7063,
+  "filtered_rows": {
+    "non_english": 54913,
+    "prompt_token_count": 6211,
+    "target_token_count": 2
+  },
   "target_turn_counts": {
-    "2": 42513,
-    "3": 21243,
-    "4": 19437
+    "2": 14364,
+    "3": 5505,
+    "4": 2198
   },
   "language_counts": {
-    "en": 28280,
-    "fr": 13451,
-    "hi": 12699,
-    "it": 13638,
-    "pt": 15125
+    "en": 22067
   },
-  "format_contract": "prompt starts with `Conversation:\\n`, ends with `Human: `, target is the untouched deepest human follow-up, and text is TASK_DESCRIPTION + prompt + target"
+  "format_contract": "prompt starts with `Conversation:\\n`, ends with `Human: `, target is the untouched deepest human follow-up, and text is TASK_DESCRIPTION + prompt + target",
+  "language_contract": "assigned_lang is one of en/eng/english; source-provided language metadata is used rather than inferring demographics or balancing",
+  "length_contract": {
+    "tokenizer": "marin-community/isoflop-1ep20-d2304-L23-B128-dclm-default-4e21eb",
+    "tokenizer_revision": "c932b4da5f56c721082e1a678ecfb8c0f4e9d4c7",
+    "max_prompt_tokens": 400,
+    "max_target_tokens": 256,
+    "max_text_tokens": 2048,
+    "add_special_tokens": false
+  }
 }