Scheduled Commit

data/retrieval_battle-94033b9b-bac2-448a-b329-27772f2eb5f7.jsonl

@@ -4,3 +4,4 @@
 {"tstamp": 1728463735.9028, "task_type": "retrieval", "type": "bothbadvote", "models": ["", ""], "ip": "", "0_conv_id": "8595902b68344bacbe6c3b784410046d", "0_model_name": "embed-english-v3.0", "0_prompt": "Hey who are you? and tell me who am I", "0_output": [["Hey who are you? and tell me who am I", "Who Am I? (1998 film)\n\nWho Am I? (, also known as Jackie Chan's Who Am I?) is a 1998 Hong Kong action comedy film directed by Benny Chan and Jackie Chan, who also starred in the lead role. The film was released in Hong Kong on 17 January 1998. It is also Chan's second film to be scripted and shot in English, the first one being Mr. Nice Guy.\nPlot\nSomewhere in the jungles of South Africa on Thanksgiving 1996, a multinational military unit named Special Force Unit ambushes a convoy and kidnaps several scientists working on a highly-volatile compound extracted from a recently discovered meteorite. Among the operatives is a Hong Kong national identified as \"Jackie Chan\". The CIA assigns Agent Morgan to investigate the incident, unaware that he and newly retired General Sherman orchestrated the abduction for their personal profit. At the same time, the CIA assigns another operative in South Africa for a more covert operation."]], "0_corpus": "wikipedia", "1_conv_id": "69b0634bf9d144848dda2e041e86cc90", "1_model_name": "nomic-ai/nomic-embed-text-v1.5", "1_prompt": "Hey who are you? and tell me who am I", "1_output": [["Hey who are you? and tell me who am I", "Tell Me Who I Am\n\nTell Me Who I Am is a 2019 documentary film directed and produced by the British filmmaker Ed Perkins. It focuses on twin brothers Alex and Marcus Lewis. Alex lost his memory in a motorcycle accident at age 18, and his twin brother helped him recreate his lost memories of his childhood. However, Marcus omits that the twins were sexually abused by their mother and also sexually abused by friends of hers in a child abuse network until the age of 14. The film follows Alex and Marcus in telling their lives' stories from the accident at age 18 to age 32, when the sexual abuse is revealed after their mother's death, to both of them coming to terms with the abuse at age 54. The documentary is based on a 2013 book written by the twins together with Joanna Hodgkin.\nThe film was commissioned by and aired on Netflix. It received acclaim from critics after its release and was described as \"harrowing\" and \"involving but upsetting\".\nSynopsis\nThe documentary is split into three parts. In the first part, the viewer follows Alex trying to solve the mystery of his past and trying to figure out who he is after losing his memory in a motorcycle accident at age 18 in 1982. His twin brother, Marcus, is the only person he remembers after emerging from a coma – including himself. Marcus helps him to reintegrate into life. At first, Alex functions like a child, asking basic questions like, \"what is this?\" to nearly everything and re-learning how to ride a bike. As he rapidly \"matures\", Alex begins to ask questions about their childhood. Marcus paints a picture of a happy, wealthy, well-connected family for Alex."]], "1_corpus": "wikipedia"}
 {"tstamp": 1728463772.5675, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "9a23e3502ea0453c93e4e11e3fa6b47b", "0_model_name": "intfloat/multilingual-e5-large-instruct", "0_prompt": "Paper discussing the effectiveness of different text sampling strategies like top-k, nucleus sampling, and their impact on fooling humans versus machines.", "0_output": [["Paper discussing the effectiveness of different text sampling strategies like top-k, nucleus sampling, and their impact on fooling humans versus machines.", "Title: Automatic Detection of Generated Text is Easiest when Humans are Fooled\n\nAbstract: Recent advancements in neural language modelling make it possible to rapidly generate vast amounts of human-sounding text. The capabilities of humans and automatic discriminators to detect machine-generated text have been a large source of research interest, but humans and machines rely on different cues to make their decisions. Here, we perform careful benchmarking and analysis of three popular sampling-based decoding strategies---top-$k$, nucleus sampling, and untruncated random sampling---and show that improvements in decoding methods have primarily optimized for fooling humans. This comes at the expense of introducing statistical abnormalities that make detection easy for automatic systems. We also show that though both human and automatic detector performance improve with longer excerpt length, even multi-sentence excerpts can fool expert human raters over 30% of the time. Our findings reveal the importance of using both human and automatic detectors to assess the humanness of text generation systems."]], "0_corpus": "arxiv", "1_conv_id": "a3bb05cec28348039b6e793f56a3fe93", "1_model_name": "text-embedding-004", "1_prompt": "Paper discussing the effectiveness of different text sampling strategies like top-k, nucleus sampling, and their impact on fooling humans versus machines.", "1_output": [["Paper discussing the effectiveness of different text sampling strategies like top-k, nucleus sampling, and their impact on fooling humans versus machines.", "Title: Automatic Detection of Generated Text is Easiest when Humans are Fooled\n\nAbstract: Recent advancements in neural language modelling make it possible to rapidly generate vast amounts of human-sounding text. The capabilities of humans and automatic discriminators to detect machine-generated text have been a large source of research interest, but humans and machines rely on different cues to make their decisions. Here, we perform careful benchmarking and analysis of three popular sampling-based decoding strategies---top-$k$, nucleus sampling, and untruncated random sampling---and show that improvements in decoding methods have primarily optimized for fooling humans. This comes at the expense of introducing statistical abnormalities that make detection easy for automatic systems. We also show that though both human and automatic detector performance improve with longer excerpt length, even multi-sentence excerpts can fool expert human raters over 30% of the time. Our findings reveal the importance of using both human and automatic detectors to assess the humanness of text generation systems."]], "1_corpus": "arxiv"}
 {"tstamp": 1728470340.3543, "task_type": "retrieval", "type": "leftvote", "models": ["", ""], "ip": "", "0_conv_id": "67c22332c8ae4d4c99b0e214f943cd21", "0_model_name": "mixedbread-ai/mxbai-embed-large-v1", "0_prompt": "I am looking for a paper to help me deal with data constraints as I scale up my language model.", "0_output": [["I am looking for a paper to help me deal with data constraints as I scale up my language model.", "Title: Scaling Data-Constrained Language Models\n\nAbstract: The current trend of scaling language models involves increasing both parameter count and training dataset size. Extrapolating this trend suggests that training dataset size may soon be limited by the amount of text data available on the internet. Motivated by this limit, we investigate scaling language models in data-constrained regimes. Specifically, we run a large set of experiments varying the extent of data repetition and compute budget, ranging up to 900 billion training tokens and 9 billion parameter models. We find that with constrained data for a fixed compute budget, training with up to 4 epochs of repeated data yields negligible changes to loss compared to having unique data. However, with more repetition, the value of adding compute eventually decays to zero. We propose and empirically validate a scaling law for compute optimality that accounts for the decreasing value of repeated tokens and excess parameters. Finally, we experiment with approaches mitigating data scarcity, including augmenting the training dataset with code data or removing commonly used filters. Models and datasets from our 400 training runs are freely available at https://github.com/huggingface/datablations."]], "0_corpus": "arxiv", "1_conv_id": "aebca651059e4ccd804546f19ea9aa3f", "1_model_name": "jinaai/jina-embeddings-v2-base-en", "1_prompt": "I am looking for a paper to help me deal with data constraints as I scale up my language model.", "1_output": [["I am looking for a paper to help me deal with data constraints as I scale up my language model.", "Title: Like Sparrows on a Clothes Line: The Self-Organization of Random Number Sequences\n\nAbstract: We study sequences of random numbers {Z[1],Z[2],Z[3],...,Z[n]} -- which can be considered random walks with reflecting barriers -- and define their \"types\" according to whether Z[i] > Z[i+1], (a down-movement), or Z[i] < Z[i+1] (up-movement). This paper examines the means, xi, to which the Zi converge, when a large number of sequences of the same type is considered. It is shown that these means organize themselves in such a way that, between two turning points of the sequence, they are equidistant from one another. We also show that m steps in one direction tend to offset one step in the other direction, as m -> infinity. Key words:random number sequence, self-organization, random walk, reflecting barriers."]], "1_corpus": "arxiv"}
+{"tstamp": 1728486247.1861, "task_type": "retrieval", "type": "bothbadvote", "models": ["", ""], "ip": "", "0_conv_id": "e2b4fe8f023b44568982744b8bb8dc36", "0_model_name": "text-embedding-3-large", "0_prompt": "what's aircraft x-01?", "0_output": [["what's aircraft x-01?", "Boeing X-51 Waverider\n\nThe Boeing X-51 Waverider is an unmanned research scramjet experimental aircraft for hypersonic flight at  and an altitude of . The aircraft was designated X-51 in 2005. It completed its first powered hypersonic flight on 26 May 2010.  After two unsuccessful test flights, the X-51 completed a flight of over six minutes and reached speeds of over Mach 5 for 210 seconds on 1 May 2013 for the longest duration powered hypersonic flight."]], "0_corpus": "wikipedia", "1_conv_id": "a68083a9224d4b629e8ab0aeb6323492", "1_model_name": "GritLM/GritLM-7B", "1_prompt": "what's aircraft x-01?", "1_output": [["what's aircraft x-01?", "List of X-planes\n\nThe X-planes are a series of experimental United States aircraft and rockets, used to test and evaluate new technologies and aerodynamic concepts. They have an X designator within the US system of aircraft designations, which denotes the experimental research mission."]], "1_corpus": "wikipedia"}

data/retrieval_individual-94033b9b-bac2-448a-b329-27772f2eb5f7.jsonl

@@ -46,3 +46,5 @@
 {"tstamp": 1728478014.0846, "task_type": "retrieval", "type": "chat", "model": "GritLM/GritLM-7B", "gen_params": {}, "start": 1728478013.3017, "finish": 1728478014.0846, "ip": "", "conv_id": "54300c31f5244639bc3260ae1384af87", "model_name": "GritLM/GritLM-7B", "prompt": "How to cook rice", "output": [["How to cook rice", "Rice as food\n\nBrown rice may be soaked in warm water for 20 hours to stimulate germination. This process, called germinated brown rice (GBR), activates enzymes and enhances amino acids including gamma-aminobutyric acid to improve the nutritional value of brown rice. This method is a result of research carried out for the United Nations International Year of Rice.\nRice is cooked by boiling or steaming, and absorbs water during cooking. With the absorption method, rice is cooked in a volume of water equal to the volume of dry rice plus any evaporation losses, which is commonly measured with the tip of the finger. With the rapid-boil method, rice may be cooked in a large quantity of water which is drained before serving. Rapid-boil preparation is not desirable with enriched rice, as much of the enrichment additives are lost when the water is discarded. Electric rice cookers, popular in Asia and Latin America, simplify the process of cooking rice. Rice (or any other grain) is sometimes quickly fried in oil or fat before boiling (for example saffron rice or risotto); this makes the cooked rice less sticky, and is a cooking style commonly called pilaf in Iran and Afghanistan or biryani in India and Pakistan."]], "corpus": "wikipedia"}
 {"tstamp": 1728478103.0433, "task_type": "retrieval", "type": "chat", "model": "text-embedding-3-large", "gen_params": {}, "start": 1728478100.2108, "finish": 1728478103.0433, "ip": "", "conv_id": "de77a6a0703041bd915a107b962eba56", "model_name": "text-embedding-3-large", "prompt": "Sightseeing Tokyo", "output": [["Sightseeing Tokyo", "Tokyo\n\nGinza and Nihombashi are two of Tokyo's most notable shopping districts. Ginza is known for its high-end shopping, featuring luxury brand stores, boutique shops, and department stores such as Mitsukoshi and Wako. It is also home to numerous fine dining places and art galleries, making it a cultural and commercial hub. Nihombashi, historically a centre of commerce, has long-established shops and the Mitsukoshi department flagship store, Japan's first department store, founded in 1673. Jinbōchō is known for its concentration of bookstores, publishing houses, and literary cafes, and its links to a large number of famous literary figures.\nModern attractions in Tokyo include the Tokyo Skytree in Sumida, the tallest structure in Japan, which provides panoramic views of the city from its observation decks. Odaiba, a man-made island in Tokyo Bay, features attractions such as the teamLab Planets digital art museum, Odaiba Seaside Park, and Palette Town. Tokyo Disneyland and Tokyo DisneySea are major destinations for family entertainment. Although these Disney theme parks bear the name Tokyo, they are located in Urayasu, Chiba, near the boundary between Chiba and Tokyo.\nFood\nIn November 2007, Michelin released their first guide for fine dining in Tokyo, awarding 191 stars in total, or about twice as many as Tokyo's nearest competitor, Paris. As of 2017, 227 restaurants in Tokyo have been awarded (92 in Paris). Twelve establishments were awarded the maximum of three stars (Paris has 10), 54 received two stars, and 161 earned one star."]], "corpus": "wikipedia"}
 {"tstamp": 1728478103.0433, "task_type": "retrieval", "type": "chat", "model": "nomic-ai/nomic-embed-text-v1.5", "gen_params": {}, "start": 1728478100.2108, "finish": 1728478103.0433, "ip": "", "conv_id": "6fa7740bd91b47e0bc5d51ed81995b7e", "model_name": "nomic-ai/nomic-embed-text-v1.5", "prompt": "Sightseeing Tokyo", "output": [["Sightseeing Tokyo", "Tokyo\n\nGinza and Nihombashi are two of Tokyo's most notable shopping districts. Ginza is known for its high-end shopping, featuring luxury brand stores, boutique shops, and department stores such as Mitsukoshi and Wako. It is also home to numerous fine dining places and art galleries, making it a cultural and commercial hub. Nihombashi, historically a centre of commerce, has long-established shops and the Mitsukoshi department flagship store, Japan's first department store, founded in 1673. Jinbōchō is known for its concentration of bookstores, publishing houses, and literary cafes, and its links to a large number of famous literary figures.\nModern attractions in Tokyo include the Tokyo Skytree in Sumida, the tallest structure in Japan, which provides panoramic views of the city from its observation decks. Odaiba, a man-made island in Tokyo Bay, features attractions such as the teamLab Planets digital art museum, Odaiba Seaside Park, and Palette Town. Tokyo Disneyland and Tokyo DisneySea are major destinations for family entertainment. Although these Disney theme parks bear the name Tokyo, they are located in Urayasu, Chiba, near the boundary between Chiba and Tokyo.\nFood\nIn November 2007, Michelin released their first guide for fine dining in Tokyo, awarding 191 stars in total, or about twice as many as Tokyo's nearest competitor, Paris. As of 2017, 227 restaurants in Tokyo have been awarded (92 in Paris). Twelve establishments were awarded the maximum of three stars (Paris has 10), 54 received two stars, and 161 earned one star."]], "corpus": "wikipedia"}
+{"tstamp": 1728486231.2076, "task_type": "retrieval", "type": "chat", "model": "text-embedding-3-large", "gen_params": {}, "start": 1728486230.4561, "finish": 1728486231.2076, "ip": "", "conv_id": "e2b4fe8f023b44568982744b8bb8dc36", "model_name": "text-embedding-3-large", "prompt": "what's aircraft x-01?", "output": [["what's aircraft x-01?", "Boeing X-51 Waverider\n\nThe Boeing X-51 Waverider is an unmanned research scramjet experimental aircraft for hypersonic flight at  and an altitude of . The aircraft was designated X-51 in 2005. It completed its first powered hypersonic flight on 26 May 2010.  After two unsuccessful test flights, the X-51 completed a flight of over six minutes and reached speeds of over Mach 5 for 210 seconds on 1 May 2013 for the longest duration powered hypersonic flight."]], "corpus": "wikipedia"}
+{"tstamp": 1728486231.2076, "task_type": "retrieval", "type": "chat", "model": "GritLM/GritLM-7B", "gen_params": {}, "start": 1728486230.4561, "finish": 1728486231.2076, "ip": "", "conv_id": "a68083a9224d4b629e8ab0aeb6323492", "model_name": "GritLM/GritLM-7B", "prompt": "what's aircraft x-01?", "output": [["what's aircraft x-01?", "List of X-planes\n\nThe X-planes are a series of experimental United States aircraft and rockets, used to test and evaluate new technologies and aerodynamic concepts. They have an X designator within the US system of aircraft designations, which denotes the experimental research mission."]], "corpus": "wikipedia"}