ivere27's picture
Initial release
ac68cef
Raw
History Blame Contribute Delete
12.6 kB
{
"format": "tiny_receipt_vqa_family_examples_v1",
"scope": "bpe1536_explicit_and_auto_answer_exact_smoke_test",
"model": {
"tokenizer": "byte_fallback_bpe",
"vocab_size": 1536,
"target_schema": "direct_answer_no_value_v1"
},
"source": {
"dataset": "synthetic validation split",
"heldout_used": false,
"selection": "visually clear English and Korean receipts that pass all seven listed family checks plus one phone-position check under explicit and AUTO routing in both deployment variants"
},
"families": {
"tested": [
"phone",
"address",
"store",
"item_row",
"item_math",
"item_lookup",
"math"
],
"trained_structured_families": [
"phone",
"address",
"store",
"item_row",
"item_math",
"item_lookup"
],
"execution_only_untrained_slots": [
"math"
],
"excluded_untrained_slots": [
"other"
],
"note": "The training distribution contains records for the six structured families only. The explicit math case verifies shared-backbone and adapter-slot execution; it is not evidence of separately trained math-family accuracy."
},
"validation": {
"case_count": 16,
"claim": "answer exact; selected-family checks are reported separately",
"explicit_family_override": {
"fp32": {
"passed": 16,
"total": 16,
"selected_family_matches": 16
},
"int8_w8a8": {
"passed": 16,
"total": 16,
"selected_family_matches": 16
}
},
"learned_auto_router": {
"fp32": {
"passed": 16,
"total": 16,
"listed_family_matches": 14
},
"int8_w8a8": {
"passed": 16,
"total": 16,
"listed_family_matches": 14
},
"selected_family_note": "AUTO selects trained item_math for both listed math questions and still produces the exact answers."
},
"math_note": "The selected explicit math smoke cases produce the exact structured outputs, but math still has zero family-specific training records and is not a family-level accuracy claim."
},
"examples": [
{
"id": "receipt_en",
"language": "en",
"image": "receipt_en.jpg",
"image_sha256": "24d7dde7d28bec844fee9134e6cc8d3180ee85315d6d493cc39c7cf0e1ebf328",
"source_split": "synth/data/val",
"source_id": "000868",
"receipt": {
"store": "Best price Mall",
"address": "Seoul Hwasun Jodonan St. 837",
"phone": "653-3831",
"items": [
{
"name": "Packed Yogurt",
"unit_price": 600,
"quantity": 3,
"total": 1800
},
{
"name": "HACCP Salt",
"unit_price": 100,
"quantity": 4,
"total": 400
}
]
},
"cases": [
{
"family": "phone",
"question": "What is the store phone number?",
"expected_answer": "6533831",
"validated_answers": {
"fp32": "6533831",
"int8_w8a8": "6533831"
},
"validated_auto": {
"fp32": {
"answer": "6533831",
"selected_family": "phone"
},
"int8_w8a8": {
"answer": "6533831",
"selected_family": "phone"
}
}
},
{
"family": "address",
"question": "What is the street number in the store address?",
"expected_answer": "837",
"validated_answers": {
"fp32": "837",
"int8_w8a8": "837"
},
"validated_auto": {
"fp32": {
"answer": "837",
"selected_family": "address"
},
"int8_w8a8": {
"answer": "837",
"selected_family": "address"
}
}
},
{
"family": "store",
"question": "What is the store name?",
"expected_answer": "Best price Mall",
"validated_answers": {
"fp32": "Best price Mall",
"int8_w8a8": "Best price Mall"
},
"validated_auto": {
"fp32": {
"answer": "Best price Mall",
"selected_family": "store"
},
"int8_w8a8": {
"answer": "Best price Mall",
"selected_family": "store"
}
}
},
{
"family": "item_row",
"question": "What is the name of item 1?",
"expected_answer": "Packed Yogurt",
"validated_answers": {
"fp32": "Packed Yogurt",
"int8_w8a8": "Packed Yogurt"
},
"validated_auto": {
"fp32": {
"answer": "Packed Yogurt",
"selected_family": "item_row"
},
"int8_w8a8": {
"answer": "Packed Yogurt",
"selected_family": "item_row"
}
}
},
{
"family": "item_math",
"question": "What is the sum of item totals?",
"expected_answer": "2200",
"validated_answers": {
"fp32": "2200",
"int8_w8a8": "2200"
},
"validated_auto": {
"fp32": {
"answer": "2200",
"selected_family": "item_math"
},
"int8_w8a8": {
"answer": "2200",
"selected_family": "item_math"
}
}
},
{
"family": "item_lookup",
"question": "What is the quantity of Packed Yogurt?",
"expected_answer": "3",
"validated_answers": {
"fp32": "3",
"int8_w8a8": "3"
},
"validated_auto": {
"fp32": {
"answer": "3",
"selected_family": "item_lookup"
},
"int8_w8a8": {
"answer": "3",
"selected_family": "item_lookup"
}
}
},
{
"family": "math",
"question": "Calculate item 1 unit price multiplied by quantity.",
"expected_answer": "1800",
"validated_answers": {
"fp32": "1800",
"int8_w8a8": "1800"
},
"validated_auto": {
"fp32": {
"answer": "1800",
"selected_family": "item_math"
},
"int8_w8a8": {
"answer": "1800",
"selected_family": "item_math"
}
}
}
],
"additional_cases": [
{
"case_type": "phone_digit_position",
"family": "phone",
"question": "What is the second number of the store's phone number?",
"expected_answer": "5",
"validated_answers": {
"fp32": "5",
"int8_w8a8": "5"
},
"validated_auto": {
"fp32": {
"answer": "5",
"selected_family": "phone"
},
"int8_w8a8": {
"answer": "5",
"selected_family": "phone"
}
}
}
]
},
{
"id": "receipt_ko",
"language": "ko",
"image": "receipt_ko.jpg",
"image_sha256": "0c9e7b52a5205eddd7944046a6db35eac2b2034535bfef34edc23b666e41d5ff",
"source_split": "synth/data/val",
"source_id": "002267",
"receipt": {
"store": "์ฐฉํ•œ ๋„๋งค์Šˆํผ๋งˆ์ผ“",
"address": "์˜์ •๋ถ€์‹œ ๋„๋ด‰๊ตฌ ์—„์น˜๊ธธ 875",
"phone": "911-3125",
"items": [
{
"name": "๊ณผ์ผ๋ฅ˜",
"unit_price": 100,
"quantity": 4,
"total": 400
},
{
"name": "๋”๋ธ”์ƒท ๋…น์ฐจ",
"unit_price": 400,
"quantity": 8,
"total": 3200
}
]
},
"cases": [
{
"family": "phone",
"question": "๊ฐ€๊ฒŒ ์ „ํ™”๋ฒˆํ˜ธ๋Š” ๋ฌด์—‡์ž…๋‹ˆ๊นŒ?",
"expected_answer": "9113125",
"validated_answers": {
"fp32": "9113125",
"int8_w8a8": "9113125"
},
"validated_auto": {
"fp32": {
"answer": "9113125",
"selected_family": "phone"
},
"int8_w8a8": {
"answer": "9113125",
"selected_family": "phone"
}
}
},
{
"family": "address",
"question": "์˜์ˆ˜์ฆ์˜ ๊ฐ€๊ฒŒ ์œ„์น˜์—์„œ ๋„๋กœ๋ช… ๋’ค ์ˆซ์ž๋Š” ๋ฌด์—‡์ž…๋‹ˆ๊นŒ?",
"expected_answer": "875",
"validated_answers": {
"fp32": "875",
"int8_w8a8": "875"
},
"validated_auto": {
"fp32": {
"answer": "875",
"selected_family": "address"
},
"int8_w8a8": {
"answer": "875",
"selected_family": "address"
}
}
},
{
"family": "store",
"question": "๊ฐ€๊ฒŒ ์ด๋ฆ„์€ ๋ฌด์—‡์ž…๋‹ˆ๊นŒ?",
"expected_answer": "์ฐฉํ•œ ๋„๋งค์Šˆํผ๋งˆ์ผ“",
"validated_answers": {
"fp32": "์ฐฉํ•œ ๋„๋งค์Šˆํผ๋งˆ์ผ“",
"int8_w8a8": "์ฐฉํ•œ ๋„๋งค์Šˆํผ๋งˆ์ผ“"
},
"validated_auto": {
"fp32": {
"answer": "์ฐฉํ•œ ๋„๋งค์Šˆํผ๋งˆ์ผ“",
"selected_family": "store"
},
"int8_w8a8": {
"answer": "์ฐฉํ•œ ๋„๋งค์Šˆํผ๋งˆ์ผ“",
"selected_family": "store"
}
}
},
{
"family": "item_row",
"question": "1๋ฒˆ์งธ ํ’ˆ๋ชฉ๋ช…์€ ๋ฌด์—‡์ž…๋‹ˆ๊นŒ?",
"expected_answer": "๊ณผ์ผ๋ฅ˜",
"validated_answers": {
"fp32": "๊ณผ์ผ๋ฅ˜",
"int8_w8a8": "๊ณผ์ผ๋ฅ˜"
},
"validated_auto": {
"fp32": {
"answer": "๊ณผ์ผ๋ฅ˜",
"selected_family": "item_row"
},
"int8_w8a8": {
"answer": "๊ณผ์ผ๋ฅ˜",
"selected_family": "item_row"
}
}
},
{
"family": "item_math",
"question": "ํ’ˆ๋ชฉ ์ดํ•ฉ์„ ๋ชจ๋‘ ๋”ํ•˜๋ฉด ์–ผ๋งˆ์ž…๋‹ˆ๊นŒ?",
"expected_answer": "3600",
"validated_answers": {
"fp32": "3600",
"int8_w8a8": "3600"
},
"validated_auto": {
"fp32": {
"answer": "3600",
"selected_family": "item_math"
},
"int8_w8a8": {
"answer": "3600",
"selected_family": "item_math"
}
}
},
{
"family": "item_lookup",
"question": "๊ณผ์ผ๋ฅ˜ ๊ฐœ์ˆ˜๋Š” ๋ช‡ ๊ฐœ์ž…๋‹ˆ๊นŒ?",
"expected_answer": "4",
"validated_answers": {
"fp32": "4",
"int8_w8a8": "4"
},
"validated_auto": {
"fp32": {
"answer": "4",
"selected_family": "item_lookup"
},
"int8_w8a8": {
"answer": "4",
"selected_family": "item_lookup"
}
}
},
{
"family": "math",
"question": "1๋ฒˆ์งธ ํ’ˆ๋ชฉ์˜ ๊ฐ€๊ฒฉ ๊ณฑํ•˜๊ธฐ ๊ฐœ์ˆ˜๋Š” ์–ผ๋งˆ์ž…๋‹ˆ๊นŒ?",
"expected_answer": "400",
"validated_answers": {
"fp32": "400",
"int8_w8a8": "400"
},
"validated_auto": {
"fp32": {
"answer": "400",
"selected_family": "item_math"
},
"int8_w8a8": {
"answer": "400",
"selected_family": "item_math"
}
}
}
],
"additional_cases": [
{
"case_type": "phone_digit_position",
"family": "phone",
"question": "๊ฐ€๊ฒŒ ์ „ํ™”๋ฒˆํ˜ธ์˜ ๋’ค์—์„œ 3๋ฒˆ์งธ ์ˆซ์ž๋Š” ๋ฌด์—‡์ž…๋‹ˆ๊นŒ?",
"expected_answer": "1",
"validated_answers": {
"fp32": "1",
"int8_w8a8": "1"
},
"validated_auto": {
"fp32": {
"answer": "1",
"selected_family": "phone"
},
"int8_w8a8": {
"answer": "1",
"selected_family": "phone"
}
}
}
]
}
]
}