Scheduled Commit

data/retrieval_battle-cc11c9ee-fc93-4cd6-a4b5-246a7f21038c.jsonl

@@ -22,3 +22,5 @@
 {"tstamp": 1722324069.0641, "task_type": "retrieval", "type": "leftvote", "models": ["", ""], "ip": "", "0_conv_id": "d465b272318c4585b28bdf1eb85d56da", "0_model_name": "text-embedding-004", "0_prompt": "What is Coagulation factor 8?", "0_output": [["What is Coagulation factor 8?", "Factor VIII\n\nCoagulation factor VIII (Factor VIII, FVIII, also known as anti-hemophilic factor (AHF)) is an essential blood clotting protein. In humans, it is encoded by F8 gene. Defects in this gene result in hemophilia A, an X-linked bleeding disorder.\nFactor VIII is produced in the liver's sinusoidal cells and endothelial cells outside the liver throughout the body. This protein circulates in the bloodstream in an inactive form, bound to another molecule called von Willebrand factor, until an injury that damages blood vessels occurs. In response to injury, coagulation factor VIII is activated and separates from von Willebrand factor. The active protein (sometimes written as coagulation factor VIIIa) interacts with another coagulation factor called factor IX. This interaction sets off a chain of additional chemical reactions that form a blood clot."]], "0_corpus": "wikipedia", "1_conv_id": "efa79c47ab3d48ddb151be6f096bae88", "1_model_name": "sentence-transformers/all-MiniLM-L6-v2", "1_prompt": "What is Coagulation factor 8?", "1_output": [["What is Coagulation factor 8?", "Coagulation\n\nContact activation pathway (intrinsic) The contact activation pathway begins with formation of the primary complex on collagen by high-molecular-weight kininogen (HMWK), prekallikrein, and FXII (Hageman factor). Prekallikrein is converted to kallikrein and FXII becomes FXIIa. FXIIa converts FXI into FXIa. Factor XIa activates FIX, which with its co-factor FVIIIa form the tenase complex, which activates FX to FXa. The minor role that the contact activation pathway has in initiating blood clot formation can be illustrated by the fact that individuals with severe deficiencies of FXII, HMWK, and prekallikrein do not have a bleeding disorder. Instead, contact activation system seems to be more involved in inflammation, and innate immunity. Despite this, interference with the pathway may confer protection against thrombosis without a significant bleeding risk.\nFinal common pathway\nThe division of coagulation in two pathways is arbitrary, originating from laboratory tests in which clotting times were measured either after the clotting was initiated by glass, the intrinsic pathway; or clotting was initiated by thromboplastin (a mix of tissue factor and phospholipids), the extrinsic pathway.\nFurther, the final common pathway scheme implies that prothrombin is converted to thrombin only when acted upon by the intrinsic or extrinsic pathways, which is an oversimplification. In fact, thrombin is generated by activated platelets at the initiation of the platelet plug, which in turn promotes more platelet activation."]], "1_corpus": "wikipedia"}
 {"tstamp": 1722325049.2363, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "f437af0ac7124e069c5aa5945335a723", "0_model_name": "BAAI/bge-large-en-v1.5", "0_prompt": "A metric for diversity in rankings in retrieval", "0_output": [["A metric for diversity in rankings in retrieval", "Title: A Ranking Algorithm for Re-finding\n\nAbstract: Re-finding files from a personal computer is a frequent demand to users. When encountered a difficult re-finding task, people may not recall the attributes used by conventional re-finding methods, such as a file's path, file name, keywords etc., the re-finding would fail. We proposed a method to support difficult re-finding tasks. By asking the user a list of questions about the target, such as a document's pages, author numbers, accumulated reading time, last reading location etc. Then use the user's answers to filter out the target. After the user answered a list of questions about the target file, we evaluate the user's familiar degree about the target file based on the answers. We devise a ranking algorithm which sorts the candidates by comparing the user's familiarity degree about the target and the candidates. We also propose a method to generate re-finding tasks artificially based on the user's own document corpus."]], "0_corpus": "arxiv", "1_conv_id": "9332df36b036426a98f95027f37c0247", "1_model_name": "text-embedding-3-large", "1_prompt": "A metric for diversity in rankings in retrieval", "1_output": [["A metric for diversity in rankings in retrieval", "Title: Structural Learning of Diverse Ranking\n\nAbstract: Relevance and diversity are both crucial criteria for an effective search system. In this paper, we propose a unified learning framework for simultaneously optimizing both relevance and diversity. Specifically, the problem is formalized as a structural learning framework optimizing Diversity-Correlated Evaluation Measures (DCEM), such as ERR-IA, a-NDCG and NRBP. Within this framework, the discriminant function is defined to be a bi-criteria objective maximizing the sum of the relevance scores and dissimilarities (or diversity) among the documents. Relevance and diversity features are utilized to define the relevance scores and dissimilarities, respectively. Compared with traditional methods, the advantages of our approach lie in that: (1) Directly optimizing DCEM as the loss function is more fundamental for the task; (2) Our framework does not rely on explicit diversity information such as subtopics, thus is more adaptive to real application; (3) The representation of diversity as the feature-based scoring function is more flexible to incorporate rich diversity-based features into the learning framework. Extensive experiments on the public TREC datasets show that our approach significantly outperforms state-of-the-art diversification approaches, which validate the above advantages."]], "1_corpus": "arxiv"}
 {"tstamp": 1722325274.4844, "task_type": "retrieval", "type": "bothbadvote", "models": ["", ""], "ip": "", "0_conv_id": "8fd0f41e8c554c77a868abae1d83248a", "0_model_name": "intfloat/multilingual-e5-large-instruct", "0_prompt": "Which programming paradigm supposedly eliminates side-effects?", "0_output": [["Which programming paradigm supposedly eliminates side-effects?", "Stackoverflow Stackexchange\n\nQ: I want to call it side-effect programming, but there's gotta be a better term for this What is the name of the following programming paradigm:\nCode is executed based on a set of logical tests resolving to true (the clause). The clause is comprised of operators and operands. Each operand is a value/object.\nInstead of evaluating the clause explicitly, as is done in imperative languages (e.g., normal flow control like if(){}), the clause is declared and bound to the resulting code. When the clause is satisfied, at any point in the future, the code will execute. \nSo, it's basically a dependency tree that re-evaluates whether the code should execute whenever a dependency changes state.\nE.g.,\nwhen(a && b && c < 3 && d.changes())\n{\n  runThisCode();\n}\n\nI'm looking for a more formal name and definition, and I haven't come up with anything after searching for it. It's somewhere between declarative and imperative, but I've never seen a language or paradigm that lets one do this.\nThanks,\nSean\n\nA: Maybe it is dataflow programming? Or reactive programming?\n"]], "0_corpus": "stackexchange", "1_conv_id": "7a4eb2fc2e1c49adb1cf34025bab88e4", "1_model_name": "BAAI/bge-large-en-v1.5", "1_prompt": "Which programming paradigm supposedly eliminates side-effects?", "1_output": [["Which programming paradigm supposedly eliminates side-effects?", "Stackoverflow Stackexchange\n\nQ: I want to call it side-effect programming, but there's gotta be a better term for this What is the name of the following programming paradigm:\nCode is executed based on a set of logical tests resolving to true (the clause). The clause is comprised of operators and operands. Each operand is a value/object.\nInstead of evaluating the clause explicitly, as is done in imperative languages (e.g., normal flow control like if(){}), the clause is declared and bound to the resulting code. When the clause is satisfied, at any point in the future, the code will execute. \nSo, it's basically a dependency tree that re-evaluates whether the code should execute whenever a dependency changes state.\nE.g.,\nwhen(a && b && c < 3 && d.changes())\n{\n  runThisCode();\n}\n\nI'm looking for a more formal name and definition, and I haven't come up with anything after searching for it. It's somewhere between declarative and imperative, but I've never seen a language or paradigm that lets one do this.\nThanks,\nSean\n\nA: Maybe it is dataflow programming? Or reactive programming?\n"]], "1_corpus": "stackexchange"}
+{"tstamp": 1722330580.3861, "task_type": "retrieval", "type": "rightvote", "models": ["", ""], "ip": "", "0_conv_id": "343e583202b94432845a84d92016e273", "0_model_name": "intfloat/e5-mistral-7b-instruct", "0_prompt": "late interaction retrieval engine", "0_output": [["late interaction retrieval engine", "Title: Improving Out-of-Distribution Generalization of Neural Rerankers with Contextualized Late Interaction\n\nAbstract: Recent progress in information retrieval finds that embedding query and document representation into multi-vector yields a robust bi-encoder retriever on out-of-distribution datasets. In this paper, we explore whether late interaction, the simplest form of multi-vector, is also helpful to neural rerankers that only use the [CLS] vector to compute the similarity score. Although intuitively, the attention mechanism of rerankers at the previous layers already gathers the token-level information, we find adding late interaction still brings an extra 5% improvement in average on out-of-distribution datasets, with little increase in latency and no degradation in in-domain effectiveness. Through extensive experiments and analysis, we show that the finding is consistent across different model sizes and first-stage retrievers of diverse natures and that the improvement is more prominent on longer queries."]], "0_corpus": "arxiv", "1_conv_id": "5305603bc27e4f5e9c54451f4c2e4662", "1_model_name": "GritLM/GritLM-7B", "1_prompt": "late interaction retrieval engine", "1_output": [["late interaction retrieval engine", "Title: PLAID: An Efficient Engine for Late Interaction Retrieval\n\nAbstract: Pre-trained language models are increasingly important components across multiple information retrieval (IR) paradigms. Late interaction, introduced with the ColBERT model and recently refined in ColBERTv2, is a popular paradigm that holds state-of-the-art status across many benchmarks. To dramatically speed up the search latency of late interaction, we introduce the Performance-optimized Late Interaction Driver (PLAID). Without impacting quality, PLAID swiftly eliminates low-scoring passages using a novel centroid interaction mechanism that treats every passage as a lightweight bag of centroids. PLAID uses centroid interaction as well as centroid pruning, a mechanism for sparsifying the bag of centroids, within a highly-optimized engine to reduce late interaction search latency by up to 7$\\times$ on a GPU and 45$\\times$ on a CPU against vanilla ColBERTv2, while continuing to deliver state-of-the-art retrieval quality. This allows the PLAID engine with ColBERTv2 to achieve latency of tens of milliseconds on a GPU and tens or just few hundreds of milliseconds on a CPU at large scale, even at the largest scales we evaluate with 140M passages."]], "1_corpus": "arxiv"}
+{"tstamp": 1722330643.9035, "task_type": "retrieval", "type": "leftvote", "models": ["", ""], "ip": "", "0_conv_id": "f57e49c925ec4e97b7cf34fa8fae1544", "0_model_name": "GritLM/GritLM-7B", "0_prompt": "retrieval system that uses an inverted index with lexically matched dense token embeddings", "0_output": [["retrieval system that uses an inverted index with lexically matched dense token embeddings", "Title: COIL: Revisit Exact Lexical Match in Information Retrieval with Contextualized Inverted List\n\nAbstract: Classical information retrieval systems such as BM25 rely on exact lexical match and carry out search efficiently with inverted list index. Recent neural IR models shifts towards soft semantic matching all query document terms, but they lose the computation efficiency of exact match systems. This paper presents COIL, a contextualized exact match retrieval architecture that brings semantic lexical matching. COIL scoring is based on overlapping query document tokens' contextualized representations. The new architecture stores contextualized token representations in inverted lists, bringing together the efficiency of exact match and the representation power of deep language models. Our experimental results show COIL outperforms classical lexical retrievers and state-of-the-art deep LM retrievers with similar or smaller latency."]], "0_corpus": "arxiv", "1_conv_id": "533774b5f05b4ce9a1513bb1b034d6cc", "1_model_name": "embed-english-v3.0", "1_prompt": "retrieval system that uses an inverted index with lexically matched dense token embeddings", "1_output": [["retrieval system that uses an inverted index with lexically matched dense token embeddings", "Title: Efficient Inverted Indexes for Approximate Retrieval over Learned Sparse Representations\n\nAbstract: Learned sparse representations form an attractive class of contextual embeddings for text retrieval. That is so because they are effective models of relevance and are interpretable by design. Despite their apparent compatibility with inverted indexes, however, retrieval over sparse embeddings remains challenging. That is due to the distributional differences between learned embeddings and term frequency-based lexical models of relevance such as BM25. Recognizing this challenge, a great deal of research has gone into, among other things, designing retrieval algorithms tailored to the properties of learned sparse representations, including approximate retrieval systems. In fact, this task featured prominently in the latest BigANN Challenge at NeurIPS 2023, where approximate algorithms were evaluated on a large benchmark dataset by throughput and recall. In this work, we propose a novel organization of the inverted index that enables fast yet effective approximate retrieval over learned sparse embeddings. Our approach organizes inverted lists into geometrically-cohesive blocks, each equipped with a summary vector. During query processing, we quickly determine if a block must be evaluated using the summaries. As we show experimentally, single-threaded query processing using our method, Seismic, reaches sub-millisecond per-query latency on various sparse embeddings of the MS MARCO dataset while maintaining high recall. Our results indicate that Seismic is one to two orders of magnitude faster than state-of-the-art inverted index-based solutions and further outperforms the winning (graph-based) submissions to the BigANN Challenge by a significant margin."]], "1_corpus": "arxiv"}

data/retrieval_individual-cc11c9ee-fc93-4cd6-a4b5-246a7f21038c.jsonl

@@ -62,3 +62,9 @@
 {"tstamp": 1722326183.0687, "task_type": "retrieval", "type": "chat", "model": "Alibaba-NLP/gte-Qwen2-7B-instruct", "gen_params": {}, "start": 1722326180.7859, "finish": 1722326183.0687, "ip": "", "conv_id": "a272df02c97d49d2a15767b334ac5ccb", "model_name": "Alibaba-NLP/gte-Qwen2-7B-instruct", "prompt": "Who wrote about a guide more popular than the Encyclopedia Galactica?", "output": [["Who wrote about a guide more popular than the Encyclopedia Galactica?", "Fictional book\n\nBill Watterson placed fictional children's books in his comic strip Calvin and Hobbes, saying that he could never reveal their contents for they were surely more outrageous in the reader's imagination. For several years, Calvin (perpetually six years old) demands that his father read him Hamster Huey and the Gooey Kablooie as a bedtime story. Occasionally, his father's patience snaps and he introduces new variations, which at least reveal what the original story is not: \"Do you think the townsfolk will ever find Hamster Huey's head?\" An \"actual\" Hamster Huey book was written by Mabel Barr in 2004, years after the strip's conclusion.\n\"Travels With My Cats,\" a Hugo-nominated short story by Mike Resnick first appearing in Asimov's Science Fiction magazine, features a fictional travelogue of the same name.\nPaul Levinson's novel The Plot To Save Socrates features a fictional ancient Platonic Dialogue, without title, that begins \"PERSONS OF THE DIALOGUE: Socrates; Andros, a visitor. SCENE: The Prison of Socrates\".\nThe Encyclopedia Galactica in Isaac Asimov's Foundation series was created in Terminus at the beginning of the Foundation Era. It serves primarily as an introduction to a character, a place or a circumstance to be developed in each chapter. Each quotation contains a copyright disclaimer and cites Terminus as the place of publication. The Encyclopedia also makes an appearance in The Hitchhiker's Guide to the Galaxy by Douglas Adams."]], "corpus": "wikipedia"}
 {"tstamp": 1722326385.221, "task_type": "retrieval", "type": "chat", "model": "GritLM/GritLM-7B", "gen_params": {}, "start": 1722326384.9839, "finish": 1722326385.221, "ip": "", "conv_id": "a642737e88ae42abb145e906bacf917a", "model_name": "GritLM/GritLM-7B", "prompt": "Who wrote about a guide more popular than the Encyclopedia Galactica?", "output": [["Who wrote about a guide more popular than the Encyclopedia Galactica?", "Encyclopedia Galactica\n\nLater instances in fiction Various authors have invoked the Encyclopedia Galactica in both science and science fiction. The first may have been author Frank Holby's short story \"The Strange Case of the Missing Hero\" in the July 1942 issue of Astounding Science Fiction which featured Sebastian Lelong, editor of the Encyclopedia. It was also a common fixture in previous incarnations of the Legion of Super-Heroes comic books, and has appeared in the Star Wars expanded universe and Superman comics set in the future.  The \"Encyclopedia Galactica\" was also mentioned as being a collection of all the knowledge of a galactic Empire in the 1989 science fiction short story \"The Originist\" by American novelist Orson Scott Card, and took place in Isaac Asimov's fictional \"Foundation\" Universe.\nIn the comic science fiction series by Douglas Adams, the Galactica is frequently contrasted with the apparently more popular Hitchhiker's Guide to the Galaxy:\nRobert A. Heinlein mentioned the Encyclopedia in chapter three of To Sail Beyond the Sunset (1987): \"... the computer that led the Lunar Revolution on time line three, code 'Neil Armstrong.' Let's skip the details; it's all in Encyclopedia Galacta (sic) and other books.\"\nIn Arthur C. Clarke's and Gentry Lee's novel Rama II (1989), Nicole des Jardins says to Richard Wakefield, \"Just think, the sum of everything all human beings know or have ever known might be nothing more than an infinitesimal fraction of the Encyclopedia Galactica.\"\n\"Encyclopedia Galactica\" is mentioned by Charlie Sheen's character in The Arrival (1996), and by Jodie Foster's character in Contact (1997)."]], "corpus": "wikipedia"}
 {"tstamp": 1722326385.221, "task_type": "retrieval", "type": "chat", "model": "Alibaba-NLP/gte-Qwen2-7B-instruct", "gen_params": {}, "start": 1722326384.9839, "finish": 1722326385.221, "ip": "", "conv_id": "a226f0bf08ff4c57a57e3e0d18a48fce", "model_name": "Alibaba-NLP/gte-Qwen2-7B-instruct", "prompt": "Who wrote about a guide more popular than the Encyclopedia Galactica?", "output": [["Who wrote about a guide more popular than the Encyclopedia Galactica?", "Fictional book\n\nBill Watterson placed fictional children's books in his comic strip Calvin and Hobbes, saying that he could never reveal their contents for they were surely more outrageous in the reader's imagination. For several years, Calvin (perpetually six years old) demands that his father read him Hamster Huey and the Gooey Kablooie as a bedtime story. Occasionally, his father's patience snaps and he introduces new variations, which at least reveal what the original story is not: \"Do you think the townsfolk will ever find Hamster Huey's head?\" An \"actual\" Hamster Huey book was written by Mabel Barr in 2004, years after the strip's conclusion.\n\"Travels With My Cats,\" a Hugo-nominated short story by Mike Resnick first appearing in Asimov's Science Fiction magazine, features a fictional travelogue of the same name.\nPaul Levinson's novel The Plot To Save Socrates features a fictional ancient Platonic Dialogue, without title, that begins \"PERSONS OF THE DIALOGUE: Socrates; Andros, a visitor. SCENE: The Prison of Socrates\".\nThe Encyclopedia Galactica in Isaac Asimov's Foundation series was created in Terminus at the beginning of the Foundation Era. It serves primarily as an introduction to a character, a place or a circumstance to be developed in each chapter. Each quotation contains a copyright disclaimer and cites Terminus as the place of publication. The Encyclopedia also makes an appearance in The Hitchhiker's Guide to the Galaxy by Douglas Adams."]], "corpus": "wikipedia"}
+{"tstamp": 1722330520.4971, "task_type": "retrieval", "type": "chat", "model": "sentence-transformers/all-MiniLM-L6-v2", "gen_params": {}, "start": 1722330520.2875, "finish": 1722330520.4971, "ip": "", "conv_id": "e8de686f05604683a883fc11feab8e61", "model_name": "sentence-transformers/all-MiniLM-L6-v2", "prompt": "can you travel with expired passport in canada?", "output": [["can you travel with expired passport in canada?", "Askubuntu Stackexchange\n\nQ: How much RAM memory should I allocate for Ubuntu when running a virtual machine? How much memory should I allocate to the host OS (in my case, Ubuntu 12.04) if I run Windows 7 Ultimate on a virtual machine (I want it for some gaming)?\nI have 4 GB of RAM. If I allocate 3512 MB RAM for Windows 7, or if I allocate even more for it, will everything still be ok?\n\nA: Ubuntu 12.04 minimum memory requirements suggest at least 512 megabytes for a desktop install.\nVisit https://www.virtualbox.org/manual/ch01.html#gui-createvm, they give similar recommendations.\nSo I would not go below 512MB for sure. But prefer keeping 1024MB if you see your host swap being actively used though.\n"]], "corpus": "stackexchange"}
+{"tstamp": 1722330520.4971, "task_type": "retrieval", "type": "chat", "model": "BAAI/bge-large-en-v1.5", "gen_params": {}, "start": 1722330520.2875, "finish": 1722330520.4971, "ip": "", "conv_id": "e16f74277f4c4c5b9a443fe994988ac4", "model_name": "BAAI/bge-large-en-v1.5", "prompt": "can you travel with expired passport in canada?", "output": [["can you travel with expired passport in canada?", "Gaming Stackexchange\n\nQ: Can I get a document back after I have returned it? After approving a migrant worker's passport and returning it to them, I realized that it was expired. I still have their work ticket, which appears valid. Is there any way to deny them entry to the country at this point?\n\nA: Nope.\nWell, you just made a big mistake: You approved before checking everything. The game is very harsh in this regard and does not allow you to roll back your decision, once you have given back the document.\nThis is part of the message - Think before you act, even if it takes valuable time ... Or just push your luck.\nAlso: You don't get fined immediately for a simple error - you get two chances every day to make mistakes (intentional or otherwise), and then the fines start kicking in.\n"]], "corpus": "stackexchange"}
+{"tstamp": 1722330571.4973, "task_type": "retrieval", "type": "chat", "model": "intfloat/e5-mistral-7b-instruct", "gen_params": {}, "start": 1722330571.2712, "finish": 1722330571.4973, "ip": "", "conv_id": "343e583202b94432845a84d92016e273", "model_name": "intfloat/e5-mistral-7b-instruct", "prompt": "late interaction retrieval engine", "output": [["late interaction retrieval engine", "Title: Improving Out-of-Distribution Generalization of Neural Rerankers with Contextualized Late Interaction\n\nAbstract: Recent progress in information retrieval finds that embedding query and document representation into multi-vector yields a robust bi-encoder retriever on out-of-distribution datasets. In this paper, we explore whether late interaction, the simplest form of multi-vector, is also helpful to neural rerankers that only use the [CLS] vector to compute the similarity score. Although intuitively, the attention mechanism of rerankers at the previous layers already gathers the token-level information, we find adding late interaction still brings an extra 5% improvement in average on out-of-distribution datasets, with little increase in latency and no degradation in in-domain effectiveness. Through extensive experiments and analysis, we show that the finding is consistent across different model sizes and first-stage retrievers of diverse natures and that the improvement is more prominent on longer queries."]], "corpus": "arxiv"}
+{"tstamp": 1722330571.4973, "task_type": "retrieval", "type": "chat", "model": "GritLM/GritLM-7B", "gen_params": {}, "start": 1722330571.2712, "finish": 1722330571.4973, "ip": "", "conv_id": "5305603bc27e4f5e9c54451f4c2e4662", "model_name": "GritLM/GritLM-7B", "prompt": "late interaction retrieval engine", "output": [["late interaction retrieval engine", "Title: PLAID: An Efficient Engine for Late Interaction Retrieval\n\nAbstract: Pre-trained language models are increasingly important components across multiple information retrieval (IR) paradigms. Late interaction, introduced with the ColBERT model and recently refined in ColBERTv2, is a popular paradigm that holds state-of-the-art status across many benchmarks. To dramatically speed up the search latency of late interaction, we introduce the Performance-optimized Late Interaction Driver (PLAID). Without impacting quality, PLAID swiftly eliminates low-scoring passages using a novel centroid interaction mechanism that treats every passage as a lightweight bag of centroids. PLAID uses centroid interaction as well as centroid pruning, a mechanism for sparsifying the bag of centroids, within a highly-optimized engine to reduce late interaction search latency by up to 7$\\times$ on a GPU and 45$\\times$ on a CPU against vanilla ColBERTv2, while continuing to deliver state-of-the-art retrieval quality. This allows the PLAID engine with ColBERTv2 to achieve latency of tens of milliseconds on a GPU and tens or just few hundreds of milliseconds on a CPU at large scale, even at the largest scales we evaluate with 140M passages."]], "corpus": "arxiv"}
+{"tstamp": 1722330633.1919, "task_type": "retrieval", "type": "chat", "model": "GritLM/GritLM-7B", "gen_params": {}, "start": 1722330632.9349, "finish": 1722330633.1919, "ip": "", "conv_id": "f57e49c925ec4e97b7cf34fa8fae1544", "model_name": "GritLM/GritLM-7B", "prompt": "retrieval system that uses an inverted index with lexically matched dense token embeddings", "output": [["retrieval system that uses an inverted index with lexically matched dense token embeddings", "Title: COIL: Revisit Exact Lexical Match in Information Retrieval with Contextualized Inverted List\n\nAbstract: Classical information retrieval systems such as BM25 rely on exact lexical match and carry out search efficiently with inverted list index. Recent neural IR models shifts towards soft semantic matching all query document terms, but they lose the computation efficiency of exact match systems. This paper presents COIL, a contextualized exact match retrieval architecture that brings semantic lexical matching. COIL scoring is based on overlapping query document tokens' contextualized representations. The new architecture stores contextualized token representations in inverted lists, bringing together the efficiency of exact match and the representation power of deep language models. Our experimental results show COIL outperforms classical lexical retrievers and state-of-the-art deep LM retrievers with similar or smaller latency."]], "corpus": "arxiv"}
+{"tstamp": 1722330633.1919, "task_type": "retrieval", "type": "chat", "model": "embed-english-v3.0", "gen_params": {}, "start": 1722330632.9349, "finish": 1722330633.1919, "ip": "", "conv_id": "533774b5f05b4ce9a1513bb1b034d6cc", "model_name": "embed-english-v3.0", "prompt": "retrieval system that uses an inverted index with lexically matched dense token embeddings", "output": [["retrieval system that uses an inverted index with lexically matched dense token embeddings", "Title: Efficient Inverted Indexes for Approximate Retrieval over Learned Sparse Representations\n\nAbstract: Learned sparse representations form an attractive class of contextual embeddings for text retrieval. That is so because they are effective models of relevance and are interpretable by design. Despite their apparent compatibility with inverted indexes, however, retrieval over sparse embeddings remains challenging. That is due to the distributional differences between learned embeddings and term frequency-based lexical models of relevance such as BM25. Recognizing this challenge, a great deal of research has gone into, among other things, designing retrieval algorithms tailored to the properties of learned sparse representations, including approximate retrieval systems. In fact, this task featured prominently in the latest BigANN Challenge at NeurIPS 2023, where approximate algorithms were evaluated on a large benchmark dataset by throughput and recall. In this work, we propose a novel organization of the inverted index that enables fast yet effective approximate retrieval over learned sparse embeddings. Our approach organizes inverted lists into geometrically-cohesive blocks, each equipped with a summary vector. During query processing, we quickly determine if a block must be evaluated using the summaries. As we show experimentally, single-threaded query processing using our method, Seismic, reaches sub-millisecond per-query latency on various sparse embeddings of the MS MARCO dataset while maintaining high recall. Our results indicate that Seismic is one to two orders of magnitude faster than state-of-the-art inverted index-based solutions and further outperforms the winning (graph-based) submissions to the BigANN Challenge by a significant margin."]], "corpus": "arxiv"}