Scheduled Commit

data/retrieval_battle-23cee96f-6768-4f6e-9a62-131dedf90618.jsonl

@@ -51,3 +51,4 @@
 {"tstamp": 1732260923.5542, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "d2a907b49d7a4ea29e1bb60e39906d8e", "0_model_name": "sentence-transformers/all-MiniLM-L6-v2", "0_prompt": "how to use AI to review risks in CVE, CWE, and CVSS reports", "0_output": [["how to use AI to review risks in CVE, CWE, and CVSS reports", "Title: Asymptotic Liapunov exponents spectrum for an extended chaotic coupled map lattice\n\nAbstract: The scaling hypothesis for the coupled chotic map lattices (CML) is formulated. Scaling properties of the CML in the regime of extensive chaos observed numerically before is justified analytically. The asymptotic Liapunov exponents spectrum for coupled piece-wise linear chaotic maps lattice is found."]], "0_corpus": "arxiv", "1_conv_id": "50e3001d99904789bd8461375a546fb4", "1_model_name": "mixedbread-ai/mxbai-embed-large-v1", "1_prompt": "how to use AI to review risks in CVE, CWE, and CVSS reports", "1_output": [["how to use AI to review risks in CVE, CWE, and CVSS reports", "Title: Artificial Intelligence Techniques for Security Vulnerability Prevention\n\nAbstract: Computer security has been a concern for decades and artificial intelligence techniques have been applied to the area for nearly as long. Most of the techniques are being applied to the detection of attacks to running systems, but recent improvements in machine learning (for example, in natural language processing) have enabled the opportunity to process software and specifications to detect vulnerabilities in a system before it is deployed. This paper presents a survey of artificial intelligence techniques (including machine learning) to detect or repair security vulnerabilities before product introduction. In the surveyed papers, techniques are presented for using NLP to analyze requirements documents for security standard completeness, performing neural fuzz testing of software, generating exploits to detect risk, and more. We categorize current techniques into 3 groups: vulnerability detection, vulnerability repair, and specification analysis. Generally, while AI techniques have become quite useful in this area, we show that AI techniques still tend to be limited in scope, providing a collection of tools which can augment but not replace careful system development to reduce vulnerability risks."]], "1_corpus": "arxiv"}
 {"tstamp": 1732280260.3504, "task_type": "retrieval", "type": "tievote", "models": ["", ""], "ip": "", "0_conv_id": "0b2adfc534844111924a36740ae06b6e", "0_model_name": "embed-english-v3.0", "0_prompt": "Banana", "0_output": [["Banana", "Banana\n\nA banana is an elongated, edible fruit – botanically a berry – produced by several kinds of large herbaceous flowering plants in the genus Musa. In some countries, cooking bananas are called plantains, distinguishing them from dessert bananas. The fruit is variable in size, color, and firmness, but is usually elongated and curved, with soft flesh rich in starch covered with a peel, which may have a variety of colors when ripe. It grows upward in clusters near the top of the plant. Almost all modern edible seedless (parthenocarp) cultivated bananas come from two wild species – Musa acuminata and Musa balbisiana, or hybrids of them.\nMusa species are native to tropical Indomalaya and Australia; they were probably domesticated in New Guinea. They are grown in 135 countries, primarily for their fruit, and to a lesser extent to make banana paper and textiles, while some are grown as ornamental plants. The world's largest producers of bananas in 2022 were India and China, which together accounted for approximately 26% of total production. Bananas are eaten raw or cooked in recipes varying from curries to banana chips, fritters, fruit preserves, or simply baked or steamed."]], "0_corpus": "wikipedia", "1_conv_id": "a64edda93da641f6b8cc4488b70c87ca", "1_model_name": "Salesforce/SFR-Embedding-2_R", "1_prompt": "Banana", "1_output": [["Banana", "Banana\n\nA banana is an elongated, edible fruit – botanically a berry – produced by several kinds of large herbaceous flowering plants in the genus Musa. In some countries, cooking bananas are called plantains, distinguishing them from dessert bananas. The fruit is variable in size, color, and firmness, but is usually elongated and curved, with soft flesh rich in starch covered with a peel, which may have a variety of colors when ripe. It grows upward in clusters near the top of the plant. Almost all modern edible seedless (parthenocarp) cultivated bananas come from two wild species – Musa acuminata and Musa balbisiana, or hybrids of them.\nMusa species are native to tropical Indomalaya and Australia; they were probably domesticated in New Guinea. They are grown in 135 countries, primarily for their fruit, and to a lesser extent to make banana paper and textiles, while some are grown as ornamental plants. The world's largest producers of bananas in 2022 were India and China, which together accounted for approximately 26% of total production. Bananas are eaten raw or cooked in recipes varying from curries to banana chips, fritters, fruit preserves, or simply baked or steamed."]], "1_corpus": "wikipedia"}
 {"tstamp": 1732294887.1615, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "75e025a1f5794a1a96eaaea2daccee10", "0_model_name": "nomic-ai/nomic-embed-text-v1.5", "0_prompt": "What are some attempts to scale context window without tons of ram?", "0_output": [["What are some attempts to scale context window without tons of ram?", "Title: Efficient Contextformer: Spatio-Channel Window Attention for Fast Context Modeling in Learned Image Compression\n\nAbstract: Entropy estimation is essential for the performance of learned image compression. It has been demonstrated that a transformer-based entropy model is of critical importance for achieving a high compression ratio, however, at the expense of a significant computational effort. In this work, we introduce the Efficient Contextformer (eContextformer) - a computationally efficient transformer-based autoregressive context model for learned image compression. The eContextformer efficiently fuses the patch-wise, checkered, and channel-wise grouping techniques for parallel context modeling, and introduces a shifted window spatio-channel attention mechanism. We explore better training strategies and architectural designs and introduce additional complexity optimizations. During decoding, the proposed optimization techniques dynamically scale the attention span and cache the previous attention computations, drastically reducing the model and runtime complexity. Compared to the non-parallel approach, our proposal has ~145x lower model complexity and ~210x faster decoding speed, and achieves higher average bit savings on Kodak, CLIC2020, and Tecnick datasets. Additionally, the low complexity of our context model enables online rate-distortion algorithms, which further improve the compression performance. We achieve up to 17% bitrate savings over the intra coding of Versatile Video Coding (VVC) Test Model (VTM) 16.2 and surpass various learning-based compression models."]], "0_corpus": "arxiv", "1_conv_id": "297057f55de7429da712df6355cf1f60", "1_model_name": "intfloat/multilingual-e5-large-instruct", "1_prompt": "What are some attempts to scale context window without tons of ram?", "1_output": [["What are some attempts to scale context window without tons of ram?", "Title: Training-Free Exponential Extension of Sliding Window Context with Cascading KV Cache\n\nAbstract: The context window within a transformer provides a form of active memory for the current task, which can be useful for few-shot learning and conditional generation, both which depend heavily on previous context tokens. However, as the context length grows, the computational cost increases quadratically. Recent works have shown that saving a few initial tokens along with a fixed-sized sliding window leads to stable streaming generation with linear complexity in transformer-based Large Language Models (LLMs). However, they make suboptimal use of the fixed window by naively evicting all tokens unconditionally from the key-value (KV) cache once they reach the end of the window, resulting in tokens being forgotten and no longer able to affect subsequent predictions. To overcome this limitation, we propose a novel mechanism for storing longer sliding window contexts with the same total cache size by keeping separate cascading sub-cache buffers whereby each subsequent buffer conditionally accepts a fraction of the relatively more important tokens evicted from the previous buffer. Our method results in a dynamic KV cache that can store tokens from the more distant past than a fixed, static sliding window approach. Our experiments show improvements of 5.6% on long context generation (LongBench), 1.2% in streaming perplexity (PG19), and 0.6% in language understanding (MMLU STEM) using LLMs given the same fixed cache size. Additionally, we provide an efficient implementation that improves the KV cache latency from 1.33ms per caching operation to 0.54ms, a 59% speedup over previous work."]], "1_corpus": "arxiv"}
+{"tstamp": 1732296435.0733, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "6746b0a2290345d2966396e96a2fab9d", "0_model_name": "voyage-multilingual-2", "0_prompt": "What is the average time of a formula 1 pit stop in 2024?", "0_output": [["What is the average time of a formula 1 pit stop in 2024?", "Pit stop\n\nOther services performed in routine pit stops include removing debris from radiator air intakes, cleaning the windscreen, and making adjustments to tyre pressure, suspension settings, and aerodynamic devices to optimize the vehicle's performance for the current conditions.  In endurance racing, scheduled driver changes and brake pad replacements are also considered \"routine\" service when done as part of a scheduled pit stop.\nAn unscheduled pit stop may see other services performed; because unscheduled stops are often due to damage or mechanical problems, they may include emergency repairs and replacement of parts.\nIn some forms of racing, a team may be ordered to bring a vehicle into the pits as a penalty and either drive through the pit lane at the maximum permitted speed or remain stationary in their pit box for a specified period of time. During this time no services can be performed.\nFormula One\nIn Formula One, mid-race refuelling has been banned since 2010, and cars make pit stops with the primary purpose of changing tyres. Teams sometimes also make adjustments to the front and rear wings and perform minor repairs, most commonly replacing the nose and front wing assembly. A pit stop typically takes 2 to 3 seconds to complete. McLaren holds the current world record for the fastest pit stop, with a 1.80-second stop performed at the 2023 Qatar Grand Prix on Lando Norris. Pit strategies generally call for between one and four scheduled stops, depending on the circuit. The drives between pit stops are commonly known as 'stints'."]], "0_corpus": "wikipedia", "1_conv_id": "8c683eae585b4286a3ca61d83d0639a3", "1_model_name": "intfloat/multilingual-e5-large-instruct", "1_prompt": "What is the average time of a formula 1 pit stop in 2024?", "1_output": [["What is the average time of a formula 1 pit stop in 2024?", "Pit stop\n\n2019 DXC Technology 600: Having led the race from pole position, Rahal Letterman Lanigan Racing's Takuma Sato entered the pits for his first stop at the end of lap 62 and overshot his pit box, hooking the inside air hoses and lollipop on his car and launching the inside front tire changer into the next pit box. The crewman was uninjured, but the lengthy pit stop and subsequent penalty for hitting pit equipment left Sato 3 laps down and out of contention.\n2019 British Grand Prix: The Red Bull Racing team completed a 1.91 pit stop for Pierre Gasly, making it the fourth fastest pit stop in F1 and overall.\n2019 German Grand Prix: The Red Bull Racing team completed a pit stop for Max Verstappen in 1.88 seconds, making it the third fastest pit stop in F1 and overall.\n2019 Brazilian Grand Prix: The Red Bull Racing team completed a pit stop for Max Verstappen in 1.82 seconds, making it the second fastest pit stop ever in F1 and overall.\n2021 Monaco Grand Prix: Valtteri Bottas of the Mercedes team was forced to retire from the race after his wheel could not be removed from the car due to a faulty wheel nut.\n2023 Qatar Grand Prix: The McLaren team completed a pit stop for Lando Norris in 1.80 seconds, making it the fastest pit stop ever in F1 and overall."]], "1_corpus": "wikipedia"}

data/retrieval_individual-23cee96f-6768-4f6e-9a62-131dedf90618.jsonl

@@ -142,3 +142,5 @@
 {"tstamp": 1732280241.7246, "task_type": "retrieval", "type": "chat", "model": "Salesforce/SFR-Embedding-2_R", "gen_params": {}, "start": 1732280241.4072, "finish": 1732280241.7246, "ip": "", "conv_id": "a64edda93da641f6b8cc4488b70c87ca", "model_name": "Salesforce/SFR-Embedding-2_R", "prompt": "Banana", "output": [["Banana", "Banana\n\nA banana is an elongated, edible fruit – botanically a berry – produced by several kinds of large herbaceous flowering plants in the genus Musa. In some countries, cooking bananas are called plantains, distinguishing them from dessert bananas. The fruit is variable in size, color, and firmness, but is usually elongated and curved, with soft flesh rich in starch covered with a peel, which may have a variety of colors when ripe. It grows upward in clusters near the top of the plant. Almost all modern edible seedless (parthenocarp) cultivated bananas come from two wild species – Musa acuminata and Musa balbisiana, or hybrids of them.\nMusa species are native to tropical Indomalaya and Australia; they were probably domesticated in New Guinea. They are grown in 135 countries, primarily for their fruit, and to a lesser extent to make banana paper and textiles, while some are grown as ornamental plants. The world's largest producers of bananas in 2022 were India and China, which together accounted for approximately 26% of total production. Bananas are eaten raw or cooked in recipes varying from curries to banana chips, fritters, fruit preserves, or simply baked or steamed."]], "corpus": "wikipedia"}
 {"tstamp": 1732294839.6508, "task_type": "retrieval", "type": "chat", "model": "nomic-ai/nomic-embed-text-v1.5", "gen_params": {}, "start": 1732294839.4352, "finish": 1732294839.6508, "ip": "", "conv_id": "75e025a1f5794a1a96eaaea2daccee10", "model_name": "nomic-ai/nomic-embed-text-v1.5", "prompt": "What are some attempts to scale context window without tons of ram?", "output": [["What are some attempts to scale context window without tons of ram?", "Title: Efficient Contextformer: Spatio-Channel Window Attention for Fast Context Modeling in Learned Image Compression\n\nAbstract: Entropy estimation is essential for the performance of learned image compression. It has been demonstrated that a transformer-based entropy model is of critical importance for achieving a high compression ratio, however, at the expense of a significant computational effort. In this work, we introduce the Efficient Contextformer (eContextformer) - a computationally efficient transformer-based autoregressive context model for learned image compression. The eContextformer efficiently fuses the patch-wise, checkered, and channel-wise grouping techniques for parallel context modeling, and introduces a shifted window spatio-channel attention mechanism. We explore better training strategies and architectural designs and introduce additional complexity optimizations. During decoding, the proposed optimization techniques dynamically scale the attention span and cache the previous attention computations, drastically reducing the model and runtime complexity. Compared to the non-parallel approach, our proposal has ~145x lower model complexity and ~210x faster decoding speed, and achieves higher average bit savings on Kodak, CLIC2020, and Tecnick datasets. Additionally, the low complexity of our context model enables online rate-distortion algorithms, which further improve the compression performance. We achieve up to 17% bitrate savings over the intra coding of Versatile Video Coding (VVC) Test Model (VTM) 16.2 and surpass various learning-based compression models."]], "corpus": "arxiv"}
 {"tstamp": 1732294839.6508, "task_type": "retrieval", "type": "chat", "model": "intfloat/multilingual-e5-large-instruct", "gen_params": {}, "start": 1732294839.4352, "finish": 1732294839.6508, "ip": "", "conv_id": "297057f55de7429da712df6355cf1f60", "model_name": "intfloat/multilingual-e5-large-instruct", "prompt": "What are some attempts to scale context window without tons of ram?", "output": [["What are some attempts to scale context window without tons of ram?", "Title: Training-Free Exponential Extension of Sliding Window Context with Cascading KV Cache\n\nAbstract: The context window within a transformer provides a form of active memory for the current task, which can be useful for few-shot learning and conditional generation, both which depend heavily on previous context tokens. However, as the context length grows, the computational cost increases quadratically. Recent works have shown that saving a few initial tokens along with a fixed-sized sliding window leads to stable streaming generation with linear complexity in transformer-based Large Language Models (LLMs). However, they make suboptimal use of the fixed window by naively evicting all tokens unconditionally from the key-value (KV) cache once they reach the end of the window, resulting in tokens being forgotten and no longer able to affect subsequent predictions. To overcome this limitation, we propose a novel mechanism for storing longer sliding window contexts with the same total cache size by keeping separate cascading sub-cache buffers whereby each subsequent buffer conditionally accepts a fraction of the relatively more important tokens evicted from the previous buffer. Our method results in a dynamic KV cache that can store tokens from the more distant past than a fixed, static sliding window approach. Our experiments show improvements of 5.6% on long context generation (LongBench), 1.2% in streaming perplexity (PG19), and 0.6% in language understanding (MMLU STEM) using LLMs given the same fixed cache size. Additionally, we provide an efficient implementation that improves the KV cache latency from 1.33ms per caching operation to 0.54ms, a 59% speedup over previous work."]], "corpus": "arxiv"}
+{"tstamp": 1732296382.7925, "task_type": "retrieval", "type": "chat", "model": "voyage-multilingual-2", "gen_params": {}, "start": 1732296380.6347, "finish": 1732296382.7925, "ip": "", "conv_id": "6746b0a2290345d2966396e96a2fab9d", "model_name": "voyage-multilingual-2", "prompt": "What is the average time of a formula 1 pit stop in 2024?", "output": [["What is the average time of a formula 1 pit stop in 2024?", "Pit stop\n\nOther services performed in routine pit stops include removing debris from radiator air intakes, cleaning the windscreen, and making adjustments to tyre pressure, suspension settings, and aerodynamic devices to optimize the vehicle's performance for the current conditions.  In endurance racing, scheduled driver changes and brake pad replacements are also considered \"routine\" service when done as part of a scheduled pit stop.\nAn unscheduled pit stop may see other services performed; because unscheduled stops are often due to damage or mechanical problems, they may include emergency repairs and replacement of parts.\nIn some forms of racing, a team may be ordered to bring a vehicle into the pits as a penalty and either drive through the pit lane at the maximum permitted speed or remain stationary in their pit box for a specified period of time. During this time no services can be performed.\nFormula One\nIn Formula One, mid-race refuelling has been banned since 2010, and cars make pit stops with the primary purpose of changing tyres. Teams sometimes also make adjustments to the front and rear wings and perform minor repairs, most commonly replacing the nose and front wing assembly. A pit stop typically takes 2 to 3 seconds to complete. McLaren holds the current world record for the fastest pit stop, with a 1.80-second stop performed at the 2023 Qatar Grand Prix on Lando Norris. Pit strategies generally call for between one and four scheduled stops, depending on the circuit. The drives between pit stops are commonly known as 'stints'."]], "corpus": "wikipedia"}
+{"tstamp": 1732296382.7925, "task_type": "retrieval", "type": "chat", "model": "intfloat/multilingual-e5-large-instruct", "gen_params": {}, "start": 1732296380.6347, "finish": 1732296382.7925, "ip": "", "conv_id": "8c683eae585b4286a3ca61d83d0639a3", "model_name": "intfloat/multilingual-e5-large-instruct", "prompt": "What is the average time of a formula 1 pit stop in 2024?", "output": [["What is the average time of a formula 1 pit stop in 2024?", "Pit stop\n\n2019 DXC Technology 600: Having led the race from pole position, Rahal Letterman Lanigan Racing's Takuma Sato entered the pits for his first stop at the end of lap 62 and overshot his pit box, hooking the inside air hoses and lollipop on his car and launching the inside front tire changer into the next pit box. The crewman was uninjured, but the lengthy pit stop and subsequent penalty for hitting pit equipment left Sato 3 laps down and out of contention.\n2019 British Grand Prix: The Red Bull Racing team completed a 1.91 pit stop for Pierre Gasly, making it the fourth fastest pit stop in F1 and overall.\n2019 German Grand Prix: The Red Bull Racing team completed a pit stop for Max Verstappen in 1.88 seconds, making it the third fastest pit stop in F1 and overall.\n2019 Brazilian Grand Prix: The Red Bull Racing team completed a pit stop for Max Verstappen in 1.82 seconds, making it the second fastest pit stop ever in F1 and overall.\n2021 Monaco Grand Prix: Valtteri Bottas of the Mercedes team was forced to retire from the race after his wheel could not be removed from the car due to a faulty wheel nut.\n2023 Qatar Grand Prix: The McLaren team completed a pit stop for Lando Norris in 1.80 seconds, making it the fastest pit stop ever in F1 and overall."]], "corpus": "wikipedia"}