Data: MultiDoc2Dial , by Song Feng, Siva Sankalp Patel, Hui Wan, Sachindra Joshi, CC BY 3.0 . The dataset's federal student aid pages in passages: each page's text taken from its HTML, with its headings marked by #s and its list items by -, and its blocks joined into passages of up to 200 words. Each passage has its page, and the page's title without the part number the dataset adds.
Data: MultiDoc2Dial , by Song Feng, Siva Sankalp Patel, Hui Wan, Sachindra Joshi, CC BY 3.0 . The opening question of each dialogue about those pages whose next turn answers it from the same page, with that page: from 500 of the dataset's training dialogues, chosen at random, and its validation dialogues.
Data: MultiDoc2Dial , by Song Feng, Siva Sankalp Patel, Hui Wan, Sachindra Joshi, CC BY 3.0 . The opening question of each dialogue about the dataset's other pages, Social Security's, Veterans Affairs' and New York State's DMV's, whose next turn answers it from the same page, from the same dialogues.
Data: MultiDoc2Dial , by Song Feng, Siva Sankalp Patel, Hui Wan, Sachindra Joshi, CC BY 3.0 . Each passage's vector from nomic-embed-text-v1.5 at 8 bits, embedded after `search_document: `, one text a request, by llama.cpp's server, release b11734, each number times 32,767, rounded.
Data: MultiDoc2Dial , by Song Feng, Siva Sankalp Patel, Hui Wan, Sachindra Joshi, CC BY 3.0 . Each question's vector, the office's and the others', from nomic-embed-text-v1.5 at 8 bits, embedded after `search_query: `, one text a request, by llama.cpp's server, release b11734, each number times 32,767, rounded.
Data: MultiDoc2Dial , by Song Feng, Siva Sankalp Patel, Hui Wan, Sachindra Joshi, CC BY 3.0 . For each of the office's questions, the passages its recorded replies were given, by their numbers, each with its page: the best passage of each of its five best pages by meaning, by the cosines of the recorded vectors.
Data: MultiDoc2Dial , by Song Feng, Siva Sankalp Patel, Hui Wan, Sachindra Joshi, CC BY 3.0 . For each of the office's questions, a reply from each of two releases, `five`, given the question's five passages, and `three`, given the first three, written by Qwen3.5-0.8B at 8 bits through llama.cpp's server, release b11734, on 26 October 2026, at the sampling settings its model card gives for text without thinking, with the question's number for its seed, each with the reason the model stopped and the tokens it wrote.
Model: Qwen3.5-0.8B , by Qwen, Apache License 2.0 .
Model: nomic-embed-text-v1.5 , by Nomic AI, Apache License 2.0 .