{"work":{"id":"a8a49dbd-ad10-4c79-b1aa-3ad5173887ad","openalex_id":"https://openalex.org/W4416051103","doi":"10.48550/arxiv.2508.15487","arxiv_id":"2508.15487","raw_key":null,"title":"Dream 7B: Diffusion Large Language Models","authors":null,"authors_text":"Jiacheng Ye, Zhihui Xie, Lin Zheng, Jiahui Gao, Zirui Wu, Xin Jiang","year":2025,"venue":"cs.CL","abstract":"We introduce Dream 7B, the most powerful open diffusion large language model to date. Unlike autoregressive (AR) models that generate tokens sequentially, Dream 7B employs discrete diffusion modeling to refine sequences in parallel through iterative denoising. Our model consistently outperforms existing diffusion language models on general, mathematical, and coding tasks. Dream 7B demonstrates superior planning abilities and inference flexibility, including arbitrary-order generation, infilling capabilities, and tunable quality-speed trade-offs. These results are achieved through simple yet effective training techniques, including AR-based LLM initialization and context-adaptive token-level noise rescheduling. We release both Dream-Base and Dream-Instruct to facilitate further research in diffusion-based language modeling.","external_url":"https://arxiv.org/abs/2508.15487","cited_by_count":0,"metadata_source":"pith","metadata_fetched_at":"2026-08-05T02:28:24.338817+00:00","pith_arxiv_id":"2508.15487","created_at":"2026-05-10T06:01:13.629300+00:00","updated_at":"2026-08-05T02:28:24.338817+00:00","title_quality_ok":true,"display_title":"Dream 7B: Diffusion Large Language Models","render_title":"Dream 7B: Diffusion Large Language Models"},"hub":{"state":{"work_id":"a8a49dbd-ad10-4c79-b1aa-3ad5173887ad","tier":"super_hub","tier_reason":"100+ Pith inbound or 10,000+ external citations","pith_inbound_count":156,"external_cited_by_count":0,"distinct_field_count":12,"first_pith_cited_at":"2025-09-24T23:59:05+00:00","last_pith_cited_at":"2026-07-08T05:48:57+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":"2026-08-23T19:59:26.332958+00:00","tier_text":"super_hub"},"tier":"super_hub","role_counts":[{"context_role":"background","n":10},{"context_role":"baseline","n":4},{"context_role":"method","n":3}],"polarity_counts":[{"context_polarity":"background","n":9},{"context_polarity":"baseline","n":4},{"context_polarity":"use_method","n":2},{"context_polarity":"extend","n":1},{"context_polarity":"unclear","n":1}],"runs":{"ask_index":{"job_type":"ask_index","status":"succeeded","result":{"title":"Dream 7B: Diffusion Large Language Models","claims":[{"claim_text":"We introduce Dream 7B, the most powerful open diffusion large language model to date. Unlike autoregressive (AR) models that generate tokens sequentially, Dream 7B employs discrete diffusion modeling to refine sequences in parallel through iterative denoising. Our model consistently outperforms existing diffusion language models on general, mathematical, and coding tasks. Dream 7B demonstrates superior planning abilities and inference flexibility, including arbitrary-order generation, infilling capabilities, and tunable quality-speed trade-offs. These results are achieved through simple yet ef","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"attributes form and commit across denoising steps remains underexplored, leaving open whether temporally informed intervention can avoid this quality cost. We take an interpretability-first approach to address this gap (Figure 1). We train SAEs on four DLMs spanning two training objectives, three architectures, and a 60× scale range (MDLM [2], SEDD [3], DREAM [4], and LLaDA [5]) and use the resulting features to characterize how attributes evolve across the denoising trajectory. Our analysis (§3","claim_type":"method","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"Model Parameters MMMU val MMMU-Prostandard MMEsum RealWorldQA MMStar AI2D ChartQA AutoRegressive Vision-Language Models Qwen3-VL [4] 4B 47.9 35.0 2297 70.5 56.9 81.0 80.9 Qwen3-VL [4] 8B 53.0 36.0 2379 69.5 59.9 83.5 84.0 InternVL3.5 [19] 4B 57.4 38.2 2236 66.7 65.6 80.6 86.2 InternVL3.5 [19] 8B 57.2 41.0 2359 63.1 66.3 82.1 87.0 Diffusion Vision-Language Models LLaDA-V [26] 8B 48.8 35.4 1998 63.4 60.4 77.8 78.2 Dream-VL [25] 7B 51.6 25.0 2179 67.7 59.9 80.4 86.2 LaviDa [11] 8B 44.2 28.6 1711 40","claim_type":"baseline","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"Classifier-free guidance (CFG) (Ho & Salimans, 2022) controls conditional diffusion sampling by amplifying the effect of conditioning information during the reverse process. Letcdenote the condition (e.g., a prompt). For a corrupted sequencex t, the mask predictor produces token-level log-probabilities ℓi cond(x) = logp θ(x|x t, c), ℓ i uncond(x) = logp θ(x|x t),(2) for each masked positioni. In dLLMs, CFG is implemented by defining guided logits as ℓi CFG(x) =ℓ i uncond(x) + (1 +γ) ℓi cond(x)−ℓ","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"state is either too far from Ut or too close to Ut, we set a fixed window size W (e.g., W= 32 ). Consequently, during training, we only need to randomly sample the initial timestep t and its context Ut; the target stateU t′ is then determined byWnewly unmasked tokens. 7 4 Experiments Models.We employ two state-of-the-art diffusion language models as our primary experimental testbeds: Dream-7B-Instruct [5] and LLaDA-8B-Instruct [6]. These models serve as the baselines upon which we apply our post","claim_type":"baseline","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"for token prediction, which iteratively generate high-confidence outputs based on contextual information, which are known as masked diffusion language models[10, 25]. LLaDA [1] is the first framework to apply masked text generation to diffusion language models. It employs a bidirectional transformer[ 26] with 8B parameters trained from scratch. Similarly, Dream[2] is built on the weights of pre-trained autoregressive models. Both frameworks achieve comparable performance to autoregressive models","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"[64, 60, 65, 99], and recent work has explored their application to text generation. Among existing paradigms, masked diffusion language models (MDLMs) [ 70, 4, 66, 105, 51] have emerged as a promising alternative to AR-LLMs by modeling language in discrete space through masked token 9 prediction. Building on this formulation, LLaDA [58] and Dream [92] scale MDLMs to the billion- parameter regime with large-scale pretraining, demonstrating their practical potential. LLaDA-2.0 [10] and LLaDA-MoE ","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks Dream 7B: Diffusion Large Language Models because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (10 contexts).","role_counts":[{"n":10,"context_role":"background"},{"n":4,"context_role":"baseline"},{"n":3,"context_role":"method"}]},"error":null,"updated_at":"2026-06-29T08:18:36.964861+00:00"},"author_expand":{"job_type":"author_expand","status":"succeeded","result":{"authors_linked":[{"id":"0d0c7252-a302-4ac2-9d42-268dae454452","orcid":null,"display_name":"Jiacheng Ye"},{"id":"ba26ac54-ef58-4395-b646-59d819ab4917","orcid":null,"display_name":"Zhihui Xie"},{"id":"975ed4ee-f78b-4c9d-8c36-2f4491e6324f","orcid":null,"display_name":"Lin Zheng"},{"id":"00b1412f-a306-478c-9164-71b867962c12","orcid":null,"display_name":"Jiahui Gao"},{"id":"0b4d4c60-5e67-4da1-a3ff-3927f3c3fe63","orcid":null,"display_name":"Zirui Wu"},{"id":"0240650d-c7fc-4da4-b9b2-fc2763fb404c","orcid":null,"display_name":"Xin Jiang"}]},"error":null,"updated_at":"2026-06-29T08:18:37.673974+00:00"},"context_extract":{"job_type":"context_extract","status":"succeeded","result":{"enqueued_papers":25},"error":null,"updated_at":"2026-05-14T17:49:51.961634+00:00"},"graph_features":{"job_type":"graph_features","status":"succeeded","result":{"co_cited":[{"title":"Large Language Diffusion Models","work_id":"cce0f4b3-ed4d-4375-b84d-3f01316016c1","shared_citers":27},{"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","shared_citers":18},{"title":"Llada2.0: Scaling up diffusion language models to 100b","work_id":"a1b1080d-0a91-44a4-8f70-2bf3e7a27e0b","shared_citers":13},{"title":"Fast-dllm: Training-free acceleration of diffusion LLM by enabling KV cache and parallel decoding.CoRR, abs/2505.22618","work_id":"9f6c2a70-9830-48ae-b181-6b5b1cbfae97","shared_citers":11},{"title":"Evaluating Large Language Models Trained on Code","work_id":"042493e9-b26f-4b4e-bbde-382072ca9b08","shared_citers":10},{"title":"Seed diffusion: A large-scale diffusion language model with high-speed inference","work_id":"7412f5f3-8e71-41c1-9c69-d4ca250b18fa","shared_citers":10},{"title":"Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution","work_id":"fcc1dcd6-aa26-420e-86d2-dc87b127ddd5","shared_citers":9},{"title":"Program Synthesis with Large Language Models","work_id":"fd241a05-03b9-4de2-9588-9d77ce176125","shared_citers":8},{"title":"dparallel: Learnable parallel decoding for dllms.arXiv preprint arXiv:2509.26488","work_id":"92d4d572-3089-4267-aa76-3c257ab6e6ad","shared_citers":7},{"title":"Llada-v: Large language diffusion models with visual instruction tuning","work_id":"0cc20892-a1cb-4674-af31-8b884e2a3a79","shared_citers":7},{"title":"Qwen3 Technical Report","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","shared_citers":7},{"title":"The Llama 3 Herd of Models","work_id":"1549a635-88af-4ac1-acfe-51ae7bb53345","shared_citers":7},{"title":"A survey on diffusion language models","work_id":"e44cc99a-c47c-4c26-9848-e709df566178","shared_citers":6},{"title":"Block diffusion: Interpolating between autoregressive and diffusion language models","work_id":"b34ab928-6ffb-4028-b13c-395a8924d76b","shared_citers":6},{"title":"d1: Scaling reasoning in diffusion large language models via reinforcement learning","work_id":"570ed81c-2193-43d1-8537-c6bdb7cd8112","shared_citers":6},{"title":"dllm-cache: Accelerating diffusion large language models with adaptive caching","work_id":"fd1cde49-2043-4004-96e2-a9cca486ec0a","shared_citers":6},{"title":"Fast-dllm v2: Efficient block-diffusion llm","work_id":"248fd492-8bf7-4ce9-a1df-130b5b976e36","shared_citers":6},{"title":"Llada 1.5: Variance-reduced preference optimization for large language diffusion models","work_id":"ebe72b3e-b18c-4784-8c3d-d7bfda67e098","shared_citers":6},{"title":"Measuring Mathematical Problem Solving With the MATH Dataset","work_id":"50652ac6-fb7c-4675-a2c2-159c241feb17","shared_citers":6},{"title":"DeepSeek-V3 Technical Report","work_id":"57d2791d-2219-4c31-a077-afc04b12a75c","shared_citers":5},{"title":"Diffusion llms can do faster-than-ar inference via discrete diffusion forcing.arXiv preprint arXiv:2508.09192","work_id":"3fd87c40-91ee-403b-9781-58b4e2feb625","shared_citers":5},{"title":"LLaDA2.1 : Speeding up text diffusion via token editing","work_id":"e3c404e2-a097-4443-8acd-3b073ce2156e","shared_citers":5},{"title":"Mercury: Ultra-fast language models based on diffusion","work_id":"09638e55-9958-4407-94da-0a6fbc082ebc","shared_citers":5},{"title":"Sdar: A synergistic diffusion-autoregression paradigm for scalable sequence generation","work_id":"78561109-9df2-4b95-b2c6-4136d3384b98","shared_citers":5}],"time_series":[{"n":40,"year":2026}],"dependency_candidates":[]},"error":null,"updated_at":"2026-05-14T17:49:38.804047+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":"2026-05-14T17:50:01.784604+00:00"},"role_polarity":{"job_type":"role_polarity","status":"succeeded","result":{"title":"Dream 7B: Diffusion Large Language Models","claims":[{"claim_text":"We introduce Dream 7B, the most powerful open diffusion large language model to date. Unlike autoregressive (AR) models that generate tokens sequentially, Dream 7B employs discrete diffusion modeling to refine sequences in parallel through iterative denoising. Our model consistently outperforms existing diffusion language models on general, mathematical, and coding tasks. Dream 7B demonstrates superior planning abilities and inference flexibility, including arbitrary-order generation, infilling capabilities, and tunable quality-speed trade-offs. These results are achieved through simple yet ef","claim_type":"abstract","evidence_strength":"source_metadata"},{"claim_text":"attributes form and commit across denoising steps remains underexplored, leaving open whether temporally informed intervention can avoid this quality cost. We take an interpretability-first approach to address this gap (Figure 1). We train SAEs on four DLMs spanning two training objectives, three architectures, and a 60× scale range (MDLM [2], SEDD [3], DREAM [4], and LLaDA [5]) and use the resulting features to characterize how attributes evolve across the denoising trajectory. Our analysis (§3","claim_type":"method","confidence":0.95,"evidence_strength":"citation_context"},{"claim_text":"Model Parameters MMMU val MMMU-Prostandard MMEsum RealWorldQA MMStar AI2D ChartQA AutoRegressive Vision-Language Models Qwen3-VL [4] 4B 47.9 35.0 2297 70.5 56.9 81.0 80.9 Qwen3-VL [4] 8B 53.0 36.0 2379 69.5 59.9 83.5 84.0 InternVL3.5 [19] 4B 57.4 38.2 2236 66.7 65.6 80.6 86.2 InternVL3.5 [19] 8B 57.2 41.0 2359 63.1 66.3 82.1 87.0 Diffusion Vision-Language Models LLaDA-V [26] 8B 48.8 35.4 1998 63.4 60.4 77.8 78.2 Dream-VL [25] 7B 51.6 25.0 2179 67.7 59.9 80.4 86.2 LaviDa [11] 8B 44.2 28.6 1711 40","claim_type":"baseline","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"Classifier-free guidance (CFG) (Ho & Salimans, 2022) controls conditional diffusion sampling by amplifying the effect of conditioning information during the reverse process. Letcdenote the condition (e.g., a prompt). For a corrupted sequencex t, the mask predictor produces token-level log-probabilities ℓi cond(x) = logp θ(x|x t, c), ℓ i uncond(x) = logp θ(x|x t),(2) for each masked positioni. In dLLMs, CFG is implemented by defining guided logits as ℓi CFG(x) =ℓ i uncond(x) + (1 +γ) ℓi cond(x)−ℓ","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"state is either too far from Ut or too close to Ut, we set a fixed window size W (e.g., W= 32 ). Consequently, during training, we only need to randomly sample the initial timestep t and its context Ut; the target stateU t′ is then determined byWnewly unmasked tokens. 7 4 Experiments Models.We employ two state-of-the-art diffusion language models as our primary experimental testbeds: Dream-7B-Instruct [5] and LLaDA-8B-Instruct [6]. These models serve as the baselines upon which we apply our post","claim_type":"baseline","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"for token prediction, which iteratively generate high-confidence outputs based on contextual information, which are known as masked diffusion language models[10, 25]. LLaDA [1] is the first framework to apply masked text generation to diffusion language models. It employs a bidirectional transformer[ 26] with 8B parameters trained from scratch. Similarly, Dream[2] is built on the weights of pre-trained autoregressive models. Both frameworks achieve comparable performance to autoregressive models","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"},{"claim_text":"[64, 60, 65, 99], and recent work has explored their application to text generation. Among existing paradigms, masked diffusion language models (MDLMs) [ 70, 4, 66, 105, 51] have emerged as a promising alternative to AR-LLMs by modeling language in discrete space through masked token 9 prediction. Building on this formulation, LLaDA [58] and Dream [92] scale MDLMs to the billion- parameter regime with large-scale pretraining, demonstrating their practical potential. LLaDA-2.0 [10] and LLaDA-MoE ","claim_type":"background","confidence":0.9,"evidence_strength":"citation_context"}],"why_cited":"Pith tracks Dream 7B: Diffusion Large Language Models because it crossed a citation-hub threshold. Current citing contexts most often use it as background evidence (10 contexts).","role_counts":[{"n":10,"context_role":"background"},{"n":4,"context_role":"baseline"},{"n":3,"context_role":"method"}]},"error":null,"updated_at":"2026-06-29T08:18:36.962443+00:00"},"summary_claims":{"job_type":"summary_claims","status":"succeeded","result":{"title":"Dream 7B: Diffusion Large Language Models","claims":[{"claim_text":"We introduce Dream 7B, the most powerful open diffusion large language model to date. Unlike autoregressive (AR) models that generate tokens sequentially, Dream 7B employs discrete diffusion modeling to refine sequences in parallel through iterative denoising. Our model consistently outperforms existing diffusion language models on general, mathematical, and coding tasks. Dream 7B demonstrates superior planning abilities and inference flexibility, including arbitrary-order generation, infilling capabilities, and tunable quality-speed trade-offs. These results are achieved through simple yet ef","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks Dream 7B: Diffusion Large Language Models because it crossed a citation-hub threshold.","role_counts":[]},"error":null,"updated_at":"2026-05-14T17:49:58.654351+00:00"}},"summary":{"title":"Dream 7B: Diffusion Large Language Models","claims":[{"claim_text":"We introduce Dream 7B, the most powerful open diffusion large language model to date. Unlike autoregressive (AR) models that generate tokens sequentially, Dream 7B employs discrete diffusion modeling to refine sequences in parallel through iterative denoising. Our model consistently outperforms existing diffusion language models on general, mathematical, and coding tasks. Dream 7B demonstrates superior planning abilities and inference flexibility, including arbitrary-order generation, infilling capabilities, and tunable quality-speed trade-offs. These results are achieved through simple yet ef","claim_type":"abstract","evidence_strength":"source_metadata"}],"why_cited":"Pith tracks Dream 7B: Diffusion Large Language Models because it crossed a citation-hub threshold.","role_counts":[]},"graph":{"co_cited":[{"title":"Large Language Diffusion Models","work_id":"cce0f4b3-ed4d-4375-b84d-3f01316016c1","shared_citers":27},{"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","shared_citers":18},{"title":"Llada2.0: Scaling up diffusion language models to 100b","work_id":"a1b1080d-0a91-44a4-8f70-2bf3e7a27e0b","shared_citers":13},{"title":"Fast-dllm: Training-free acceleration of diffusion LLM by enabling KV cache and parallel decoding.CoRR, abs/2505.22618","work_id":"9f6c2a70-9830-48ae-b181-6b5b1cbfae97","shared_citers":11},{"title":"Evaluating Large Language Models Trained on Code","work_id":"042493e9-b26f-4b4e-bbde-382072ca9b08","shared_citers":10},{"title":"Seed diffusion: A large-scale diffusion language model with high-speed inference","work_id":"7412f5f3-8e71-41c1-9c69-d4ca250b18fa","shared_citers":10},{"title":"Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution","work_id":"fcc1dcd6-aa26-420e-86d2-dc87b127ddd5","shared_citers":9},{"title":"Program Synthesis with Large Language Models","work_id":"fd241a05-03b9-4de2-9588-9d77ce176125","shared_citers":8},{"title":"dparallel: Learnable parallel decoding for dllms.arXiv preprint arXiv:2509.26488","work_id":"92d4d572-3089-4267-aa76-3c257ab6e6ad","shared_citers":7},{"title":"Llada-v: Large language diffusion models with visual instruction tuning","work_id":"0cc20892-a1cb-4674-af31-8b884e2a3a79","shared_citers":7},{"title":"Qwen3 Technical Report","work_id":"25a4e30c-1232-48e7-9925-02fa12ba7c9e","shared_citers":7},{"title":"The Llama 3 Herd of Models","work_id":"1549a635-88af-4ac1-acfe-51ae7bb53345","shared_citers":7},{"title":"A survey on diffusion language models","work_id":"e44cc99a-c47c-4c26-9848-e709df566178","shared_citers":6},{"title":"Block diffusion: Interpolating between autoregressive and diffusion language models","work_id":"b34ab928-6ffb-4028-b13c-395a8924d76b","shared_citers":6},{"title":"d1: Scaling reasoning in diffusion large language models via reinforcement learning","work_id":"570ed81c-2193-43d1-8537-c6bdb7cd8112","shared_citers":6},{"title":"dllm-cache: Accelerating diffusion large language models with adaptive caching","work_id":"fd1cde49-2043-4004-96e2-a9cca486ec0a","shared_citers":6},{"title":"Fast-dllm v2: Efficient block-diffusion llm","work_id":"248fd492-8bf7-4ce9-a1df-130b5b976e36","shared_citers":6},{"title":"Llada 1.5: Variance-reduced preference optimization for large language diffusion models","work_id":"ebe72b3e-b18c-4784-8c3d-d7bfda67e098","shared_citers":6},{"title":"Measuring Mathematical Problem Solving With the MATH Dataset","work_id":"50652ac6-fb7c-4675-a2c2-159c241feb17","shared_citers":6},{"title":"DeepSeek-V3 Technical Report","work_id":"57d2791d-2219-4c31-a077-afc04b12a75c","shared_citers":5},{"title":"Diffusion llms can do faster-than-ar inference via discrete diffusion forcing.arXiv preprint arXiv:2508.09192","work_id":"3fd87c40-91ee-403b-9781-58b4e2feb625","shared_citers":5},{"title":"LLaDA2.1 : Speeding up text diffusion via token editing","work_id":"e3c404e2-a097-4443-8acd-3b073ce2156e","shared_citers":5},{"title":"Mercury: Ultra-fast language models based on diffusion","work_id":"09638e55-9958-4407-94da-0a6fbc082ebc","shared_citers":5},{"title":"Sdar: A synergistic diffusion-autoregression paradigm for scalable sequence generation","work_id":"78561109-9df2-4b95-b2c6-4136d3384b98","shared_citers":5}],"time_series":[{"n":40,"year":2026}],"dependency_candidates":[]},"authors":[{"id":"0d0c7252-a302-4ac2-9d42-268dae454452","orcid":null,"display_name":"Jiacheng Ye","source":"manual","import_confidence":0.72},{"id":"00b1412f-a306-478c-9164-71b867962c12","orcid":null,"display_name":"Jiahui Gao","source":"manual","import_confidence":0.72},{"id":"975ed4ee-f78b-4c9d-8c36-2f4491e6324f","orcid":null,"display_name":"Lin Zheng","source":"manual","import_confidence":0.72},{"id":"0240650d-c7fc-4da4-b9b2-fc2763fb404c","orcid":null,"display_name":"Xin Jiang","source":"manual","import_confidence":0.72},{"id":"ba26ac54-ef58-4395-b646-59d819ab4917","orcid":null,"display_name":"Zhihui Xie","source":"manual","import_confidence":0.72},{"id":"0b4d4c60-5e67-4da1-a3ff-3927f3c3fe63","orcid":null,"display_name":"Zirui Wu","source":"manual","import_confidence":0.72}]}}