community_alignment_modified / validation_report.json
Alberto1231's picture
Document explicit validation splits
a586b19 verified
Raw History Blame Contribute Delete
4.62 kB
{
"source_dataset": "facebook/community-alignment-dataset",
"source_revision": "97343c7f6399fcbea430ed0f37c1768281a78d56",
"source_rows": 90256,
"published_rows": 2514,
"training_rows": 2509,
"fewshot_rows": 5,
"fewshot_conversation_ids": [
529851820193270,
1397186111302571,
9607340472689046,
1202852077472500,
1043746071135066
],
"skipped_rows_without_followup": 7063,
"filtered_rows": {
"content_non_english": 24,
"prompt_token_count": 6211,
"source_non_english": 54913,
"target_token_count": 2
},
"deduplication": {
"normalization": "Unicode NFKC + casefold + collapsed whitespace",
"eligible_rows_before_deduplication": 22043,
"unique_content_groups": 22034,
"duplicate_content_groups": 8,
"duplicate_rows_removed": 9,
"largest_duplicate_group": 3,
"seed": 42
},
"target_turn_counts": {
"2": 14332,
"3": 5504,
"4": 2198
},
"language_counts": {
"en": 22034
},
"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 and langid==1.1.6 classifies the combined complete conversation as English; no 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
},
"demonstration_selection": {
"selection_version": "prism_aligned_stratified_v1",
"selection_mode": "fixed stratum quotas",
"seed": 42,
"identity_column": "conversation_id",
"stratum_column": "target_turn",
"stratum_quotas": {
"2": 2,
"3": 2,
"4": 1
},
"minimum_target_words": 30,
"maximum_target_words": 200,
"maximum_prompt_plus_target_tokens": 256,
"eligible_rows_by_stratum": {
"2": 987,
"3": 29,
"4": 1
},
"selected_ids": [
"529851820193270",
"1397186111302571",
"9607340472689046",
"1202852077472500",
"1043746071135066"
],
"selected_strata": [
"2",
"2",
"3",
"3",
"4"
],
"selected_target_words": [
48,
31,
40,
35,
35
],
"query_rows_excluded_by_duplicate_content": 0
},
"validation_split": {
"format_version": "lm_eval_five_shot_v1",
"format_contract": "lm-eval description + five prompt/target demonstrations joined by two newlines + query prompt + query target",
"fewshot_count": 5,
"fewshot_delimiter": "\\n\\n",
"identity_column": "conversation_id",
"fewshot_ids": [
"529851820193270",
"1397186111302571",
"9607340472689046",
"1202852077472500",
"1043746071135066"
],
"source_tokens_excluding_eos": 650269,
"processed_tokens_including_eos": 3194395,
"selected_tokens_including_eos": 3194395,
"five_shot_overhead_tokens_per_document": 1013,
"selected_complete_documents": 2509,
"max_input_tokens": 1445,
"max_document_tokens": 1507,
"input_token_limit": 1792,
"document_token_limit": 2048,
"add_special_tokens": false,
"eos_tokens_per_document": 1,
"demonstration_selection": {
"selection_version": "prism_aligned_stratified_v1",
"selection_mode": "fixed stratum quotas",
"seed": 42,
"identity_column": "conversation_id",
"stratum_column": "target_turn",
"stratum_quotas": {
"2": 2,
"3": 2,
"4": 1
},
"minimum_target_words": 30,
"maximum_target_words": 200,
"maximum_prompt_plus_target_tokens": 256,
"eligible_rows_by_stratum": {
"2": 987,
"3": 29,
"4": 1
},
"selected_ids": [
"529851820193270",
"1397186111302571",
"9607340472689046",
"1202852077472500",
"1043746071135066"
],
"selected_strata": [
"2",
"2",
"3",
"3",
"4"
],
"selected_target_words": [
48,
31,
40,
35,
35
],
"query_rows_excluded_by_duplicate_content": 0
},
"source_selection": {
"source_token_budget_excluding_eos": 650000,
"selected_source_tokens_excluding_eos": 650269,
"source_token_overshoot": 269,
"selected_complete_source_documents": 2509,
"boundary_source_document_tokens": 342
}
}
}