{"work":{"id":"f22dae5a-27e2-41d0-a061-c4286418dee3","openalex_id":null,"doi":null,"arxiv_id":"2401.14196","raw_key":null,"title":"DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence","authors":null,"authors_text":"Daya Guo, Qihao Zhu, Dejian Yang, Zhenda Xie, Kai Dong, Wentao Zhang","year":2024,"venue":"cs.SE","abstract":"The rapid development of large language models has revolutionized code intelligence in software development. However, the predominance of closed-source models has restricted extensive research and development. To address this, we introduce the DeepSeek-Coder series, a range of open-source code models with sizes from 1.3B to 33B, trained from scratch on 2 trillion tokens. These models are pre-trained on a high-quality project-level code corpus and employ a fill-in-the-blank task with a 16K window to enhance code generation and infilling. Our extensive evaluations demonstrate that DeepSeek-Coder not only achieves state-of-the-art performance among open-source code models across multiple benchmarks but also surpasses existing closed-source models like Codex and GPT-3.5. Furthermore, DeepSeek-Coder models are under a permissive license that allows for both research and unrestricted commercial use.","external_url":"https://arxiv.org/abs/2401.14196","cited_by_count":null,"metadata_source":"pith","metadata_fetched_at":"2026-05-25T08:20:31.507063+00:00","pith_arxiv_id":"2401.14196","created_at":"2026-05-08T17:28:42.017187+00:00","updated_at":"2026-05-25T08:20:31.507063+00:00","title_quality_ok":true,"display_title":"DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence","render_title":"DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence"},"hub":{"state":{"work_id":"f22dae5a-27e2-41d0-a061-c4286418dee3","tier":"super_hub","tier_reason":"100+ Pith inbound or 10,000+ external citations","pith_inbound_count":134,"external_cited_by_count":null,"distinct_field_count":12,"first_pith_cited_at":"2024-02-02T13:42:50+00:00","last_pith_cited_at":"2026-05-21T18:14:09+00:00","author_build_status":"needed","summary_status":"needed","contexts_status":"needed","graph_status":"needed","ask_index_status":"needed","reader_status":"not_needed","recognition_status":"not_needed","updated_at":"2026-06-04T15:57:45.139380+00:00","tier_text":"super_hub"},"tier":"super_hub","role_counts":[{"context_role":"background","n":30},{"context_role":"method","n":6},{"context_role":"dataset","n":2},{"context_role":"baseline","n":1}],"polarity_counts":[{"context_polarity":"background","n":29},{"context_polarity":"use_method","n":6},{"context_polarity":"use_dataset","n":2},{"context_polarity":"baseline","n":1},{"context_polarity":"support","n":1}],"runs":{"ask_index":{"job_type":"ask_index","status":"succeeded","result":{"title":"DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence","claims":[{"claim_text":"The rapid development of large language models has revolutionized code intelligence in software development. However, the predominance of closed-source models has restricted extensive research and development. To address this, we introduce the DeepSeek-Coder series, a range of open-source code models with sizes from 1.3B to 33B, trained from scratch on 2 trillion tokens. These models are pre-trained on a high-quality project-level code corpus and employ a fill-in-the-blank task with a 16K window to enhance code generation and infilling. Our extensive evaluations demonstrate that DeepSeek-Coder","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"Spatial reasoning with vision-language models in ego-centric multi-view scenes.arXiv preprint arXiv:2509.06266, 2025. [34] Raktim Gautam Goswami, Prashanth Krishnamurthy, Yann LeCun, and Farshad Khorrami. Osvi-wm: One-shot visualimitationforunseentasksusingworld-model-guidedtrajectorygeneration. arXivpreprintarXiv:2505.20425, 2025. OpenDCAI Technical Report 19 [35] Daya Guo, Qihao Zhu, Dejian Yang, Zhenda Xie, Kai Dong, Wentao Zhang, Guanting Chen, Xiao Bi, Yu Wu, YK Li, et al. Deepseek-coder: W","claim_type":"background","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"#Func means the average number of functions in each example. #Repo means whether benchmarks have repo-level evaluation metrics. Benchmark Task Input InformationDifferent Difficult LevelsSource Data Leaking EvaluationGranularity#Req #Diag #Func Class File #LOC#Real-R #NumData Time#Repo Class/#Func Tests Concode [19] Function✓ ✗ 2,000 0 0 - ✓- 2018 ✗ ✗-CoNaLA [44] Statement✗ ✗ 500 0 0 1 ✓- 2018 ✗ ✗-APPS [17] Function✓ ✗ 5,000 0 0 21.4 ✗- 2021 ✗ ✓13.2HumanEval [11]Function✓ ✗ 164 0 0 11.5 ✗- 2021 ✗","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"perimental environment. This approach allows us to isolate the effects of our data filtering strategies by eliminating confounding variables from prior training stages, which is a common practice in this line of pre-training data filtering research [4, 7, 32, 40]. For the learning rate strategy, we followed prior work [40] and adopted the Warmup-Stable-Decay (WSD) [14] learning rate sched- ule. This approach ensures that the learning rate remains stable at 1𝑒 −4 for the majority of the training ","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"metrics, and reporting practices further complicates comparisons across studies [27-29] leaving both researchers and practitioners uncertain about how to adapt and assess prompt strategies in practice. These challenges are amplified by the heterogeneity of LLM architectures-encoder-only (e.g.,CodeBERT [ 30], GraphCodeBERT [31]), encoder-decoder (e.g.,CodeT5 [ 32], PLBART [33]), and decoder-only (e.g.,GPT-4 [ 15], CodeLlama [22], DeepSeek-Coder [34])-each with distinct capabilities and constraint","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"DSL toolchain requirements. To investigate feasibility and cost-benefit trade-offs of adapting LLMs, we evaluate two instruction-tuned LLMs (Qwen2.5-Coder and DeepSeek-Coder, 7B) under three configurations: (i) baseline zero-shot prompting, (ii) one-shot in-context learning (ICL) with DSL rules and an example [ 3], and (iii) parameter-efficient fine- tuning using QLoRA [18]. Because generic similarity metrics alone are insufficient for repository-scale DSL changes, we combine stan- dard measures","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"KeywordsLLM Safety · Malicious Code Generation · Benchmark Dataset · Multi-Judge Classification · Fleiss Kappa 1 Introduction Evaluations of whether large language models (LLMs) refuse malicious requests have become a central activity in AI safety research. In coding-specialized LLMs, open-weight systems such as Code Llama [1], StarCoder and StarCoder2 [2, 3], DeepSeek-Coder [4], and Qwen2.5-Coder [5] now match or exceed closed-model performance on benchmarks such as HumanEval [6]. In this setti","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (23 contexts).","role_counts":[{"n":23,"context_role":"background"},{"n":5,"context_role":"method"},{"n":2,"context_role":"dataset"},{"n":1,"context_role":"baseline"}]},"error":null,"updated_at":"2026-05-19T08:11:39.706952+00:00"},"author_expand":{"job_type":"author_expand","status":"succeeded","result":{"authors_linked":[{"id":"a13a64d4-5fb6-4a23-86ae-52ba08a175a9","orcid":null,"display_name":"Daya Guo"},{"id":"b7f65b1d-099e-4295-a8c1-64ecb925c661","orcid":null,"display_name":"Qihao Zhu"},{"id":"3b1d7e71-775d-41c7-a01c-132771e3162f","orcid":null,"display_name":"Dejian Yang"},{"id":"b2cd9f86-c0a7-44c8-8b7a-6fd8cea42fa6","orcid":null,"display_name":"Zhenda Xie"},{"id":"f93ab3fb-8ba4-442b-b38b-63a91d981d58","orcid":null,"display_name":"Kai Dong"},{"id":"0239b48b-727a-4313-a4b3-cb9387521b7f","orcid":null,"display_name":"Wentao Zhang"}]},"error":null,"updated_at":"2026-05-19T08:11:39.700463+00:00"},"context_extract":{"job_type":"context_extract","status":"succeeded","result":{"enqueued_papers":25},"error":null,"updated_at":"2026-05-14T08:57:54.291286+00:00"},"graph_features":{"job_type":"graph_features","status":"succeeded","result":{"co_cited":[{"title":"Evaluating Large Language Models Trained on Code","work_id":"042493e9-b26f-4b4e-bbde-382072ca9b08","shared_citers":40},{"title":"Code Llama: Open Foundation Models for Code","work_id":"e73bffa4-7620-47ac-9327-259a60db52ca","shared_citers":35},{"title":"Qwen2.5-Coder Technical Report","work_id":"09ba463d-6377-4017-9801-444ffb94b056","shared_citers":31},{"title":"Program Synthesis with Large Language Models","work_id":"fd241a05-03b9-4de2-9588-9d77ce176125","shared_citers":21},{"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","shared_citers":16},{"title":"DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models","work_id":"c5006563-f3ec-438a-9e35-b7b484f34828","shared_citers":14},{"title":"StarCoder: may the source be with you!","work_id":"7e9c3d6e-d6f7-4763-9ef6-de471506c58f","shared_citers":14},{"title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","work_id":"e6b75ad5-2877-4168-97c8-710407094d20","shared_citers":13},{"title":"DeepSeek-V3 Technical Report","work_id":"57d2791d-2219-4c31-a077-afc04b12a75c","shared_citers":13},{"title":"StarCoder 2 and The Stack v2: The Next Generation","work_id":"2495cc72-f326-4c23-8a39-d9a08cf583e4","shared_citers":12},{"title":"LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code","work_id":"ea9e51ce-1e75-4182-92d8-4d25f70d2ee4","shared_citers":11},{"title":"Qwen3 Technical Report","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","shared_citers":11},{"title":"Proximal Policy Optimization Algorithms","work_id":"240c67fe-d14d-4520-91c1-38a4e272ca19","shared_citers":10},{"title":"Qwen Technical Report","work_id":"bb1fd52f-6b2f-437c-9516-37bdf6eb9be8","shared_citers":10},{"title":"Measuring Mathematical Problem Solving With the MATH Dataset","work_id":"50652ac6-fb7c-4675-a2c2-159c241feb17","shared_citers":9},{"title":"Mistral 7B","work_id":"eb5e1305-ad11-4875-ad8d-ad8b8f697599","shared_citers":9},{"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","shared_citers":9},{"title":"arXiv preprint arXiv:2009.08366 , year=","work_id":"05f7fd92-3de8-42af-9707-70010caa34c6","shared_citers":8},{"title":"LLaMA: Open and Efficient Foundation Language Models","work_id":"c018fc23-6f3f-4035-9d02-28a2173b2b9d","shared_citers":8},{"title":"Qwen2.5 Technical Report","work_id":"d8432992-4980-4a81-85c7-9fa2c2b87f85","shared_citers":8},{"title":"The Llama 3 Herd of Models","work_id":"1549a635-88af-4ac1-acfe-51ae7bb53345","shared_citers":8},{"title":"arXiv preprint arXiv:2401.03065 , year=","work_id":"0daed386-84ea-40ec-bdeb-546f8991fca5","shared_citers":7},{"title":"Llama 2: Open Foundation and Fine-Tuned Chat Models","work_id":"68a5177f-d644-44c1-bd4f-4e5278c22f5d","shared_citers":7},{"title":"Measuring Coding Challenge Competence With APPS","work_id":"c014c12f-1080-4cb2-ae03-ab6b7c09445c","shared_citers":7}],"time_series":[{"n":5,"year":2024},{"n":2,"year":2025},{"n":64,"year":2026}],"dependency_candidates":[]},"error":null,"updated_at":"2026-05-14T08:57:59.017735+00:00"},"identity_refresh":{"job_type":"identity_refresh","status":"succeeded","result":{"items":[{"title":"Qwen3 Technical Report","outcome":"unchanged","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","resolver":"local_arxiv","confidence":0.98,"old_work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e"}],"counts":{"fixed":0,"merged":0,"unchanged":1,"quarantined":0,"needs_external_resolution":0},"errors":[],"attempted":1},"error":null,"updated_at":"2026-05-14T08:57:50.315699+00:00"},"role_polarity":{"job_type":"role_polarity","status":"succeeded","result":{"title":"DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence","claims":[{"claim_text":"The rapid development of large language models has revolutionized code intelligence in software development. However, the predominance of closed-source models has restricted extensive research and development. To address this, we introduce the DeepSeek-Coder series, a range of open-source code models with sizes from 1.3B to 33B, trained from scratch on 2 trillion tokens. These models are pre-trained on a high-quality project-level code corpus and employ a fill-in-the-blank task with a 16K window to enhance code generation and infilling. Our extensive evaluations demonstrate that DeepSeek-Coder","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"Spatial reasoning with vision-language models in ego-centric multi-view scenes.arXiv preprint arXiv:2509.06266, 2025. [34] Raktim Gautam Goswami, Prashanth Krishnamurthy, Yann LeCun, and Farshad Khorrami. Osvi-wm: One-shot visualimitationforunseentasksusingworld-model-guidedtrajectorygeneration. arXivpreprintarXiv:2505.20425, 2025. OpenDCAI Technical Report 19 [35] Daya Guo, Qihao Zhu, Dejian Yang, Zhenda Xie, Kai Dong, Wentao Zhang, Guanting Chen, Xiao Bi, Yu Wu, YK Li, et al. Deepseek-coder: W","claim_type":"background","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"#Func means the average number of functions in each example. #Repo means whether benchmarks have repo-level evaluation metrics. Benchmark Task Input InformationDifferent Difficult LevelsSource Data Leaking EvaluationGranularity#Req #Diag #Func Class File #LOC#Real-R #NumData Time#Repo Class/#Func Tests Concode [19] Function✓ ✗ 2,000 0 0 - ✓- 2018 ✗ ✗-CoNaLA [44] Statement✗ ✗ 500 0 0 1 ✓- 2018 ✗ ✗-APPS [17] Function✓ ✗ 5,000 0 0 21.4 ✗- 2021 ✗ ✓13.2HumanEval [11]Function✓ ✗ 164 0 0 11.5 ✗- 2021 ✗","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"perimental environment. This approach allows us to isolate the effects of our data filtering strategies by eliminating confounding variables from prior training stages, which is a common practice in this line of pre-training data filtering research [4, 7, 32, 40]. For the learning rate strategy, we followed prior work [40] and adopted the Warmup-Stable-Decay (WSD) [14] learning rate sched- ule. This approach ensures that the learning rate remains stable at 1𝑒 −4 for the majority of the training ","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"metrics, and reporting practices further complicates comparisons across studies [27-29] leaving both researchers and practitioners uncertain about how to adapt and assess prompt strategies in practice. These challenges are amplified by the heterogeneity of LLM architectures-encoder-only (e.g.,CodeBERT [ 30], GraphCodeBERT [31]), encoder-decoder (e.g.,CodeT5 [ 32], PLBART [33]), and decoder-only (e.g.,GPT-4 [ 15], CodeLlama [22], DeepSeek-Coder [34])-each with distinct capabilities and constraint","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"DSL toolchain requirements. To investigate feasibility and cost-benefit trade-offs of adapting LLMs, we evaluate two instruction-tuned LLMs (Qwen2.5-Coder and DeepSeek-Coder, 7B) under three configurations: (i) baseline zero-shot prompting, (ii) one-shot in-context learning (ICL) with DSL rules and an example [ 3], and (iii) parameter-efficient fine- tuning using QLoRA [18]. Because generic similarity metrics alone are insufficient for repository-scale DSL changes, we combine stan- dard measures","claim_type":"method","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"KeywordsLLM Safety · Malicious Code Generation · Benchmark Dataset · Multi-Judge Classification · Fleiss Kappa 1 Introduction Evaluations of whether large language models (LLMs) refuse malicious requests have become a central activity in AI safety research. In coding-specialized LLMs, open-weight systems such as Code Llama [1], StarCoder and StarCoder2 [2, 3], DeepSeek-Coder [4], and Qwen2.5-Coder [5] now match or exceed closed-model performance on benchmarks such as HumanEval [6]. In this setti","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (23 contexts).","role_counts":[{"n":23,"context_role":"background"},{"n":5,"context_role":"method"},{"n":2,"context_role":"dataset"},{"n":1,"context_role":"baseline"}]},"error":null,"updated_at":"2026-05-19T08:11:39.712230+00:00"},"summary_claims":{"job_type":"summary_claims","status":"succeeded","result":{"title":"DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence","claims":[{"claim_text":"The rapid development of large language models has revolutionized code intelligence in software development. However, the predominance of closed-source models has restricted extensive research and development. To address this, we introduce the DeepSeek-Coder series, a range of open-source code models with sizes from 1.3B to 33B, trained from scratch on 2 trillion tokens. These models are pre-trained on a high-quality project-level code corpus and employ a fill-in-the-blank task with a 16K window to enhance code generation and infilling. Our extensive evaluations demonstrate that DeepSeek-Coder","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence because it crossed a citation-hub threshold.","role_counts":[]},"error":null,"updated_at":"2026-05-14T08:57:52.328540+00:00"}},"summary":{"title":"DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence","claims":[{"claim_text":"The rapid development of large language models has revolutionized code intelligence in software development. However, the predominance of closed-source models has restricted extensive research and development. To address this, we introduce the DeepSeek-Coder series, a range of open-source code models with sizes from 1.3B to 33B, trained from scratch on 2 trillion tokens. These models are pre-trained on a high-quality project-level code corpus and employ a fill-in-the-blank task with a 16K window to enhance code generation and infilling. Our extensive evaluations demonstrate that DeepSeek-Coder","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence because it crossed a citation-hub threshold.","role_counts":[]},"graph":{"co_cited":[{"title":"Evaluating Large Language Models Trained on Code","work_id":"042493e9-b26f-4b4e-bbde-382072ca9b08","shared_citers":40},{"title":"Code Llama: Open Foundation Models for Code","work_id":"e73bffa4-7620-47ac-9327-259a60db52ca","shared_citers":35},{"title":"Qwen2.5-Coder Technical Report","work_id":"09ba463d-6377-4017-9801-444ffb94b056","shared_citers":31},{"title":"Program Synthesis with Large Language Models","work_id":"fd241a05-03b9-4de2-9588-9d77ce176125","shared_citers":21},{"title":"GPT-4 Technical Report","work_id":"b928e041-6991-4c08-8c81-0359e4097c7b","shared_citers":16},{"title":"DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models","work_id":"c5006563-f3ec-438a-9e35-b7b484f34828","shared_citers":14},{"title":"StarCoder: may the source be with you!","work_id":"7e9c3d6e-d6f7-4763-9ef6-de471506c58f","shared_citers":14},{"title":"DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning","work_id":"e6b75ad5-2877-4168-97c8-710407094d20","shared_citers":13},{"title":"DeepSeek-V3 Technical Report","work_id":"57d2791d-2219-4c31-a077-afc04b12a75c","shared_citers":13},{"title":"StarCoder 2 and The Stack v2: The Next Generation","work_id":"2495cc72-f326-4c23-8a39-d9a08cf583e4","shared_citers":12},{"title":"LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code","work_id":"ea9e51ce-1e75-4182-92d8-4d25f70d2ee4","shared_citers":11},{"title":"Qwen3 Technical Report","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","shared_citers":11},{"title":"Proximal Policy Optimization Algorithms","work_id":"240c67fe-d14d-4520-91c1-38a4e272ca19","shared_citers":10},{"title":"Qwen Technical Report","work_id":"bb1fd52f-6b2f-437c-9516-37bdf6eb9be8","shared_citers":10},{"title":"Measuring Mathematical Problem Solving With the MATH Dataset","work_id":"50652ac6-fb7c-4675-a2c2-159c241feb17","shared_citers":9},{"title":"Mistral 7B","work_id":"eb5e1305-ad11-4875-ad8d-ad8b8f697599","shared_citers":9},{"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","shared_citers":9},{"title":"arXiv preprint arXiv:2009.08366 , year=","work_id":"05f7fd92-3de8-42af-9707-70010caa34c6","shared_citers":8},{"title":"LLaMA: Open and Efficient Foundation Language Models","work_id":"c018fc23-6f3f-4035-9d02-28a2173b2b9d","shared_citers":8},{"title":"Qwen2.5 Technical Report","work_id":"d8432992-4980-4a81-85c7-9fa2c2b87f85","shared_citers":8},{"title":"The Llama 3 Herd of Models","work_id":"1549a635-88af-4ac1-acfe-51ae7bb53345","shared_citers":8},{"title":"arXiv preprint arXiv:2401.03065 , year=","work_id":"0daed386-84ea-40ec-bdeb-546f8991fca5","shared_citers":7},{"title":"Llama 2: Open Foundation and Fine-Tuned Chat Models","work_id":"68a5177f-d644-44c1-bd4f-4e5278c22f5d","shared_citers":7},{"title":"Measuring Coding Challenge Competence With APPS","work_id":"c014c12f-1080-4cb2-ae03-ab6b7c09445c","shared_citers":7}],"time_series":[{"n":5,"year":2024},{"n":2,"year":2025},{"n":64,"year":2026}],"dependency_candidates":[]},"authors":[{"id":"a13a64d4-5fb6-4a23-86ae-52ba08a175a9","orcid":null,"display_name":"Daya Guo","source":"manual","import_confidence":0.72},{"id":"3b1d7e71-775d-41c7-a01c-132771e3162f","orcid":null,"display_name":"Dejian Yang","source":"manual","import_confidence":0.72},{"id":"f93ab3fb-8ba4-442b-b38b-63a91d981d58","orcid":null,"display_name":"Kai Dong","source":"manual","import_confidence":0.72},{"id":"b7f65b1d-099e-4295-a8c1-64ecb925c661","orcid":null,"display_name":"Qihao Zhu","source":"manual","import_confidence":0.72},{"id":"0239b48b-727a-4313-a4b3-cb9387521b7f","orcid":null,"display_name":"Wentao Zhang","source":"manual","import_confidence":0.72},{"id":"b2cd9f86-c0a7-44c8-8b7a-6fd8cea42fa6","orcid":null,"display_name":"Zhenda Xie","source":"manual","import_confidence":0.72}]}}