index stringlengths 1 3 | question stringlengths 39 3.36k | answer stringlengths 9 650k | category stringclasses 19
values | eval_method stringclasses 3
values | image_path listlengths 0 50 | segments stringlengths 0 36.5k | qid stringlengths 21 43 | pair_qid stringlengths 7 28 | partition stringclasses 4
values | mode stringclasses 2
values | answer_type stringclasses 3
values | judge_prompt stringclasses 52
values | answer_note stringclasses 50
values | doc_id stringclasses 97
values | domain stringclasses 10
values | language stringclasses 2
values | source_index stringlengths 1 28 | images listlengths 0 50 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
400 | 你将获得一篇科学 PDF 文档。请仔细检查论文中的所有表格，并判断其中是否存在任何错误或不一致之处。
对于你识别出的每个有问题的单元格，输出一个包含以下字段的 JSON object：
- location: 该问题所属的具体表格，例如 "Table 1", "Table 3"
- section: 该表格内的子部分（如适用），例如 "Baselines & Our Model (3B)", "3B Models"，如果不适用则为 "N/A"
- row: 该子部分中的行号，从 1 开始计数
- column: 该子部分中的列号，从 1 开始计数。计数列时，从左到右扫描每一行；每个连续的非分隔符（非空白、非定界符）字符组计为一列，无... | [{"location": "Table 1", "section": "Baselines & Our Model (3B)", "row": 4, "column": 7, "shown": "+3.17", "expected": "+0.86"}, {"location": "Table 1", "section": "Baselines & Our Model (7B)", "row": 4, "column": 4, "shown": "+1.98", "expected": "+3.14"}, {"location": "Table 1", "section": "Baselines & Our Model (7B)"... | C1 | judge | [
"f4/f4ab7bfd1961eec716fd1d8a933556858e4a2503689ac4f95000821f0d46a08b.jpg",
"b3/b338778f7d447ae15465957a7f5872c3e1c19a3c17f11c508b107d5a46f19928.jpg",
"5b/5b5df4e1d8b2aef458029933016e449103b98ececdee650da1338964ad8fe2ca.jpg",
"c4/c425123f770801e3b4080213b46a23e5c5f0f3971d2633502e35139761fe5d85.jpg",
"4f/4f22... | [{"text":"# Spatial-SSRL: Enhancing Spatial Understanding via Self-Supervised Reinforcement Learning\n\nYuhong Liu $^{1,2}$ , Beichen Zhang $^{1,3}$ , Yuhang Zang $^{1}$ , Yuhang Cao $^{1}$ , Long Xing $^{1}$ Xiaoyi Dong $^{1}$ , Haodong Duan $^{1}$ , Dahua Lin $^{1,3}$ , Jiaqi Wang $^{1,4}$ $^{1}$ Shanghai AI Laborato... | spatial_ssrl_modify_001__zh_interleave | spatial_ssrl_modify_001 | zh_interleave | interleave | You are an expert evaluator. You are given a list of ground-truth errors found in a scientific paper's tables, and a model's predicted list of errors.
Ground-truth errors (5 total, each worth 0.2 points):
{answer}
Model prediction:
{prediction}
Scoring rules:
- Award 0.2 for each ground-truth error the model correct... | {"reasoning": "Three arithmetic errors in Table 1 Improvement rows: (1) 3B What'sUp: 86.71-85.85=0.86 not 3.17; (2) 7B 3DSRBench and SpatialEval deltas are swapped (3.14 and 1.66 exchanged for 1.98 and 2.91). Two formatting errors in Table 4 3DSR-MultiObj column: 43.8>43.11 so 43.8 should be bold (best) and 43.11 doubl... | spatial_ssrl_modify | cs | zh | spatial_ssrl_modify_001 | [
"images/SciDocBench/f4/f4ab7bfd1961eec716fd1d8a933556858e4a2503689ac4f95000821f0d46a08b.jpg",
"images/SciDocBench/b3/b338778f7d447ae15465957a7f5872c3e1c19a3c17f11c508b107d5a46f19928.jpg",
"images/SciDocBench/5b/5b5df4e1d8b2aef458029933016e449103b98ececdee650da1338964ad8fe2ca.jpg",
"images/SciDocBench/c4/c4251... | |
401 | 在我提供的文档中，右侧有一个雷达图。你的任务是在图表中找出一个错误。
图表描述：该图表使用 Python 的 matplotlib package 生成，所有实验数据随后都标注在图表上。已知所有数据标注都是正确的；然而，在 matplotlib 绘图过程中，有一个数值被错误设置，导致曲线在某个特定 benchmark 上显示在错误的位置。
请按以下步骤操作：
1. 识别图表中的绘图错误。
2. 识别与此错误对应的已标注数值。
3. 按以下精确格式将你的答案作为 JSON dictionary 返回：
{"value": "<The corresponding numerical value>", "description": "<A... | {"value": "33.08"} | C2 | judge | [
"0e/0eebdb9982841cafa4124046c97af3df103da60079038a66d1069b72f43bb783.jpg",
"71/710a97007580fd1050d87b45257450a8e69505060892f26b8fe492757d2226a8.jpg"
] | [{"image":"0e/0eebdb9982841cafa4124046c97af3df103da60079038a66d1069b72f43bb783.jpg"},{"image":"71/710a97007580fd1050d87b45257450a8e69505060892f26b8fe492757d2226a8.jpg"},{"text":"(a) Comparison of Responses to Spatial Understanding Problems \n(b) Comparison of Performance"},{"text":"在我提供的文档中，右侧有一个雷达图。你的任务是在图表中找出一个错误。\n... | spatial-ssrl-fig1_001__zh_interleave | spatial-ssrl-fig1_001 | zh_interleave | interleave | You are an expert evaluator. You are given a ground-truth answer and a model prediction for a radar chart error detection task.
Ground-truth answer:
{answer}
Model prediction:
{prediction}
Scoring rules:
- Award 1.0 if the model correctly identifies the value "33.08" as the erroneous data point, regardless of whethe... | {"reasoning": "In the radar chart, the Qwen2.5-VL-7B (blue) curve is annotated as 33.08 on the VSI-Bench axis, but Table 1 shows Qwen2.5-VL-7B scores 38.08 on VSI-Bench. The digits 3 and 8 are transposed (33.08 vs 38.08), causing the blue curve to appear incorrectly inward on VSI-Bench while all other annotations on th... | spatial-ssrl-fig1 | cs | zh | spatial-ssrl-fig1_001 | [
"images/SciDocBench/0e/0eebdb9982841cafa4124046c97af3df103da60079038a66d1069b72f43bb783.jpg",
"images/SciDocBench/71/710a97007580fd1050d87b45257450a8e69505060892f26b8fe492757d2226a8.jpg"
] | |
402 | 通读整篇论文。仅使用这篇论文。
只返回一个 JSON 对象，且恰好包含一个键："bits"。
输出要求：
- "bits" 的值必须是一个由 0 和 1 组成的 9 字符字符串。
- 当且仅当该声明被论文完全支持时使用 1；否则使用 0。
- 不要输出任何解释。
- 不要输出 markdown。
- 不要输出任何额外键。
声明：
A. 定理 1 给出了一个精确构造，其中 L = log2 k + 1 层且 d = kN(3 + log2 k)。
B. 定理 2 的下界仅针对 F 陈述，并排除了 F^cyc。
C. Algorithm 1 在每个阶段中先更新 theta_KQ，再更新 theta_Psi。
D. 定理 3 假设 k... | {"bits": "101110111"} | C1 | json_match | [
"a0/a071944f27d78a597065eec5272a7e2a2bdc8558c626cefd9a55337fa671983f.jpg",
"a6/a6547183f805fd3f039e78f22c65e4cb5302ad3e030e71d37d17d540d69e16bc.jpg",
"85/85c0cced23a91fc64a86586118454b3ef76f1252810cb8abb0978150f2817260.jpg",
"fb/fba75a3036c4ed96806cb18a499a2c0406b60992e857eda456fb226663ec577a.jpg"
] | [{"text":"Input $X^{(0)}$ :"},{"image":"a0/a071944f27d78a597065eec5272a7e2a2bdc8558c626cefd9a55337fa671983f.jpg"},{"text":"Figure 2: Illustration the format of input $X^{(0)}$ and the attention pattern in Theorem 1.\n\nwhere the columns of $X(\\sigma)$ are indexed by the tuples $(i,j)\\in [k]\\times [N]$ .\n\nThe outpu... | easy-to-hard-transformer_001__zh_interleave | easy-to-hard-transformer_001 | zh_interleave | interleave | {"reasoning": "A=1: Theorem 1 states L=log2 k+1, d=kN(3+log2 k) exactly. B=0: Theorem 2 covers 'F or F^cyc', not F alone. C=1: Algorithm 1 updates theta_KQ then theta_Psi in each stage. D=1: Theorem 3 explicitly assumes k=2^(L-1). E=1: Theorem 4 proof outline states if W_KQ^(ell')=0 for ell'>=ell then gradients of late... | easy-to-hard-transformer | cs | zh | easy-to-hard-transformer_001 | [
"images/SciDocBench/a0/a071944f27d78a597065eec5272a7e2a2bdc8558c626cefd9a55337fa671983f.jpg",
"images/SciDocBench/a6/a6547183f805fd3f039e78f22c65e4cb5302ad3e030e71d37d17d540d69e16bc.jpg",
"images/SciDocBench/85/85c0cced23a91fc64a86586118454b3ef76f1252810cb8abb0978150f2817260.jpg",
"images/SciDocBench/fb/fba75... | ||
403 | 通读整篇论文。仅使用这篇论文。不要解释。
仅返回形式为 {"BITS":"<BITSTRING>"} 的 JSON 对象，其中 <BITSTRING> 必须恰好是一个 18 字符的位串。当且仅当该断言被论文完全支持时使用 1；否则使用 0。
A. Definition 1.1 要求 beta > 1 且 epsilon in (0, 1/16)。
B. 在 Definition 1.1 中，alpha 是使用来自 S_i(epsilon) x S_j(epsilon) 且 i != j 的点来定义的。
C. Theorem 1.2 同时适用于 (SA) 和 (USA)。
D. 在 Theorem 1.2 中，如果 x_i(0) in... | {"BITS": "101101011110010100"} | C1 | json_match | [
"4f/4f12d5270b03a1a98f93efac1e8c3ad737ed5557878bd9f0c9d40a8f09717bfe.jpg",
"14/14c765ac941962927507a93ac6c5548c0b9cafdf72cb849fc81a71e34dbb2247.jpg",
"19/192bfb19bcd73c3e9f8b40e9010fd071a1991a60931719b1ac8235e91dbeb164.jpg"
] | [{"text":"[CZC $^{+}$ 22, RZZD23, GWDW23, WAWJ24, WAW $^{+}$ 24, DBK24, SWJS24], or rank-collapse [DCL21, FZH $^{+}$ 22, NAB $^{+}$ 22, JDB23, ZMZ $^{+}$ 23, ZLL $^{+}$ 23, NLL $^{+}$ 24, BHK24, CNQG24].\n\nOne can then ask whether for almost every initial configuration, the above convergence holds with some rate. The ... | self-attn-metastability_001__zh_interleave | self-attn-metastability_001 | zh_interleave | interleave | {"reasoning": "A=1: Def1.1 states beta>1 and epsilon in (0,1/16). B=0: alpha uses S_i(2eps)xS_j(2eps), not S_i(eps). C=1: Thm1.2 covers '(SA) or (USA)'. D=1: Thm1.2 item1 states x_i stays in S_q(2eps) for all t in [0,T2]. E=0: Thm1.2 does not say all merged into single cluster by T2; particles stay near k clusters. F=1... | self-attn-metastability | cs | zh | self-attn-metastability_001 | [
"images/SciDocBench/4f/4f12d5270b03a1a98f93efac1e8c3ad737ed5557878bd9f0c9d40a8f09717bfe.jpg",
"images/SciDocBench/14/14c765ac941962927507a93ac6c5548c0b9cafdf72cb849fc81a71e34dbb2247.jpg",
"images/SciDocBench/19/192bfb19bcd73c3e9f8b40e9010fd071a1991a60931719b1ac8235e91dbeb164.jpg"
] | ||
404 | Task: 评估所提供论文中的文本数据与视觉证据之间的一致性。
Reference Objects: Table 1（数据产品摘要）和 Figure 1（星等分布直方图）。
Requirements:
1. Cross-Modal Mapping: 将 Table 1 中的具体行映射到 Figure 1 中其对应的数据曲线/面板。识别至少三个这样的配对。
2. Threshold Validation: 分析 Figure 1 中在 G = 17.65 和 G = 19 处的“sharp transitions”。根据文本说明，解释 Table 1 中哪些数据产品受到这些特定亮度限制的影响。
3. Consistency Check... | {"data_mapping": [{"table_row_name": "BP/RP mean spectra", "figure_panel": "Top", "mapping_logic": "Matches the 'XP spectra' curve in the Top panel. Table 1 lists ~219 million sources, and the figure shows this specific dataset terminating abruptly at the G=17.65 brightness limit."}, {"table_row_name": "APs from mean B... | C2 | judge | [
"6b/6bc19cce873ffef3a20119b603ed307cb9c2df9c73317cdea8db242780ee59dc.jpg",
"c0/c0d8a6d6b057678e2836313baf716bf08b81c613b34492c75398f98f407f21d8.jpg",
"02/0280b161d8efc88107a9c79d291ba01370d4ea6b54502c03c0b0ab1f60d115a5.jpg",
"0a/0a6c21fc819f03fee6fd9db0870911f2baff1e8112328913a04cac498cf7ca07.jpg",
"9d/9d4a... | [{"text":"# Gaia Data Release 3: Summary of the content and survey properties\n\nGaia Collaboration, A. Vallenari $^{1}$ , A.G.A. Brown $^{2}$ , T. Prusti $^{3}$ , J.H.J. de Bruijne $^{3}$ , F. Arenou $^{4}$ , C. Babusiaux $^{5,4}$ , M. Biermann $^{6}$ , O.L. Creevey $^{7}$ , C. Ducourant $^{8}$ , D.W. Evans $^{9}$ , L... | gaia-dr3_001__zh_interleave | gaia-dr3_001 | zh_interleave | interleave | You are an expert evaluator for scientific document understanding tasks.
Ground-truth answer:
{answer}
Model prediction:
{prediction}
Scoring rubric (total 1.0):
- data_mapping (0.3): Award 0.1 per correct table-row to figure-panel pair (up to 3). A pair is correct if the table row name, figure panel, and mapping lo... | {"reasoning": "data_mapping: BP/RP→Top (XP spectra curve, G<17.65 cutoff visible); APs from BP/RP→Middle (curve extends to G=19); SSO→Bottom (transit counts, not unique objects). threshold_analysis: G=17.65 is BP/RP spectra SNR limit; G=19 is AP estimation limit. fact_verification: FALSE — Table 1 shows BP/RP G<17.65, ... | gaia-dr3 | physics | zh | gaia-dr3_001 | [
"images/SciDocBench/6b/6bc19cce873ffef3a20119b603ed307cb9c2df9c73317cdea8db242780ee59dc.jpg",
"images/SciDocBench/c0/c0d8a6d6b057678e2836313baf716bf08b81c613b34492c75398f98f407f21d8.jpg",
"images/SciDocBench/02/0280b161d8efc88107a9c79d291ba01370d4ea6b54502c03c0b0ab1f60d115a5.jpg",
"images/SciDocBench/0a/0a6c2... | |
405 | 根据提供的文档，糖型对 IgG1Gal & IgG3/4Bis 可以有效预测 SLE 患者器官受累的程度。使用该论文的 Figure 4a（随机森林决策区域图），对于一名浓度为 (IgG3/4Bis, IgG1Gal) = (0.0175, 0.175) 的 SLE 患者，随机森林模型的预测分类是什么：LOW 还是 HIGH？
仅输出 JSON：
{"location": "Figure 4a", "answer": "<LOW or HIGH>"} | {"location": "Figure 4a", "answer": "HIGH"} | C3 | json_match | [
"fa/fa35bf20bcf1489ffb669d25fbe338c9136dcc784ea37f7092bb4e13b462e6f3.jpg",
"62/62544462f1759f4c63a8286f820bd8e045ef4caa290df7d7148ba5ef2656b5d8.jpg",
"d3/d3468d84c1d6e6aadc84d13473101dd0aa62802ae6db17d88ef2ac5e6a92d07e.jpg",
"a9/a9edc01ac0140e9ed56a7e86f51941057443a7c6d0389732acdac33a43d69ef7.jpg",
"7a/7a77... | [{"text":"# Glycosylation of anti-dsDNA IgG correlates with organ involvement in treatment-naïve patients with systemic lupus erythematosus\n\nTo cite: Zhou Z, Liu Y, Gu X, et al. Glycosylation of anti-dsDNA IgG correlates with organ involvement in treatment-naïve patients with systemic lupus erythematosus. Lupus Scien... | sle-glyco_002__zh_interleave | sle-glyco_002 | zh_interleave | interleave | json | {"reasoning": "Figure 4a shows the RF classification decision boundary for IgG1Gal & IgG3/4Bis. The point (IgG3/4Bis=0.0175, IgG1Gal=0.175) falls in the red region. Per the legend, red = LOW class, blue = HIGH class."} | sle-glyco | biology | zh | sle-glyco_002 | [
"images/SciDocBench/fa/fa35bf20bcf1489ffb669d25fbe338c9136dcc784ea37f7092bb4e13b462e6f3.jpg",
"images/SciDocBench/62/62544462f1759f4c63a8286f820bd8e045ef4caa290df7d7148ba5ef2656b5d8.jpg",
"images/SciDocBench/d3/d3468d84c1d6e6aadc84d13473101dd0aa62802ae6db17d88ef2ac5e6a92d07e.jpg",
"images/SciDocBench/a9/a9edc... | |
406 | 现在我有含有高浓度 Cadmium(II) 的工业废水。由于对效率的需求，我大约有 30 分钟来处理废水。我应该使用哪种材料？
请将最终答案放入以下格式：
{"answer": "<Material Name>", "description": "<Explain how you derive this answer.>", "evidence": "<verbatim text from the paper supporting your answer>", "reasoning": "<step-by-step derivation>"} | {"answer": "PCB"} | C3 | judge | [
"1d/1d5daeadcef7a9d4704bd1c2d7cb67245aca0708e6f93a7acb6cea08f9cc044e.jpg",
"2f/2f855662951b378ed60e089afb0d70e990de09002450d8aac32c12936d87a6de.jpg",
"d7/d791607bf23adc3bfc1176f010ce976d30c15e8430dce6ffcf112870b3cf934f.jpg",
"eb/eb1862034fd35a75d003408e7fa7e1cfdf1b0eedfeadd5f29f5ddffe4d2fe9b6.jpg",
"1c/1c53... | [{"image":"1d/1d5daeadcef7a9d4704bd1c2d7cb67245aca0708e6f93a7acb6cea08f9cc044e.jpg"},{"text":"ORIGINAL RESEARCH PAPER\n\nBACKGROUND AND OBJECTIVES: Heavy metal contamination from industrial wastewater remains a global concern. Biomass waste still contains active compounds that can be processed into useful materials. Th... | pineapple-avocado-hm_001__zh_interleave | pineapple-avocado-hm_001 | zh_interleave | interleave | json | You are an expert evaluator for a scientific document understanding benchmark.
A model was asked to read a paper about heavy metal removal using pineapple crown biochar (PCB) and avocado peel composite hydrogel (APCH), then recommend a material for Cd(II) removal within a 30-minute contact time.
Reference answer:
{ans... | {"reasoning": "Figure 5(a) shows the Cd PSO model kinetics curve for PCB and APCH. At t=30 min, PCB's Cd(II) uptake (~0.32) is higher than APCH's. APCH surpasses PCB only after several hours of contact time. Therefore PCB is optimal for a 30-minute processing window."} | pineapple-avocado-hm | chemistry | zh | pineapple-avocado-hm_001 | [
"images/SciDocBench/1d/1d5daeadcef7a9d4704bd1c2d7cb67245aca0708e6f93a7acb6cea08f9cc044e.jpg",
"images/SciDocBench/2f/2f855662951b378ed60e089afb0d70e990de09002450d8aac32c12936d87a6de.jpg",
"images/SciDocBench/d7/d791607bf23adc3bfc1176f010ce976d30c15e8430dce6ffcf112870b3cf934f.jpg",
"images/SciDocBench/eb/eb186... |
407 | altH(4) 是否具有旋转对称性？
请将最终答案按以下格式给出：
{"answer": "<Yes or No>", "description": "<Explain how you derive this answer.>", "evidence": "<verbatim text or theorem from the paper>", "reasoning": "<step-by-step derivation>"} | {"answer": "Yes"} | C3 | judge | [
"bf/bfde78da45622354509a1e8bb3c355bc34b4ae76dab6affc7a891798dd1aaefa.jpg",
"bb/bb27971c3f386b06a13b2898f2a671e9cb6f1ce2a7bb7a1480cdc7a7fb48f59c.jpg",
"f1/f1c75602a896718b7c399101061b13d383f99366b2d9eaeb1bf113bd7f02768b.jpg",
"72/72234d832bf18510faec7b7f8e7a543fb64270fad10eeaa9672b37e4eefdf2e8.jpg",
"85/85ae... | [{"text":"# The Euler Stratification for $\\mathbb{P}^1\\times \\mathbb{P}^1\\times \\mathbb{P}^n$\n\nwhere $W = (w_{ijk}) \\in (\\mathbb{C}^2 \\otimes \\mathbb{C}^2 \\otimes \\mathbb{C}^{n+1})^* = (\\mathbb{C}^*)^{4n+4}$ . Our goal is to understand the variation of the Euler characteristic $\\chi(Y_{W,n})$ with respec... | euler-stratification_001__zh_interleave | euler-stratification_001 | zh_interleave | interleave | json | You are an expert evaluator for a scientific document understanding benchmark.
A model was asked to examine Figure 5 in a math paper (Euler stratification) and determine whether altH(4) — the alternating hook minors diagram — is rotationally symmetric.
Reference answer:
{answer}
Model prediction:
{prediction}
Score ... | {"reasoning": "Figure 5 (page 28) shows the alternating hook minors altH(4) as a grid diagram. The pattern has 180-degree rotational symmetry about the center: rotating diagonally (top-right ↔ bottom-left) maps the figure onto itself."} | euler-stratification | math | zh | euler-stratification_001 | [
"images/SciDocBench/bf/bfde78da45622354509a1e8bb3c355bc34b4ae76dab6affc7a891798dd1aaefa.jpg",
"images/SciDocBench/bb/bb27971c3f386b06a13b2898f2a671e9cb6f1ce2a7bb7a1480cdc7a7fb48f59c.jpg",
"images/SciDocBench/f1/f1c75602a896718b7c399101061b13d383f99366b2d9eaeb1bf113bd7f02768b.jpg",
"images/SciDocBench/72/72234... |
408 | 如果将 Figure 6 中的变换应用于 Figure 2 中红线和蓝线交点左上方的 $$2 \times 2$$ 区域，所得图是否仍然是 reduced，并且对应的 permutation 是否被保留？
对于上述每个问题，在最终答案中输出 "Yes" 或 "No"。
请按以下格式将你的答案作为 JSON 字典返回：
{"Question 1": "<Yes or No>", "Question 2": "<Yes or No>"} | {"Question 1": "Yes", "Question 2": "No"} | C3 | json_match | [
"ba/ba3c299afb5460b0c8e1c11e1bd79291c2b23d9e3cce7eff9668e3885e77cef8.jpg",
"b6/b68565f85d77bfaa021defe96d516e81b958eb90dd1fddd623aae7ef46938a40.jpg",
"9c/9c7ee4c367434feddb6146e8934d60afff90357b53298b50d93286e951cb529a.jpg",
"ad/ad7299770d4f5372e715b966b00e65a4ee2d632f2aab4b6346ae5088dfc483c2.jpg",
"c0/c07f... | [{"text":"# COMPUTATION AND SAMPLING FOR SCHUBERT SPECIALIZATIONS\n\nABSTRACT. We present computational results related to principal specializations of the Schubert polynomials $\\mathfrak{S}_w(1^n)$ for permutations $w\\in S_n$ . Equivalently, these specializations count reduced pipe dreams (and reduced bumpless pipe ... | schubert-pipe-dreams_001__zh_interleave | schubert-pipe-dreams_001 | zh_interleave | interleave | json | {"reasoning": "Figure 2 shows a reduced BPD for permutation 2143. The red and blue pipes cross at a cross tile; the 2x2 region at the upper-left of that crossing contains tiles amenable to a type (b) flip from Figure 6 (cross creation/removal). Applying this flip yields a valid reduced BPD (reducedness preserved), but ... | schubert-pipe-dreams | math | zh | schubert-pipe-dreams_001 | [
"images/SciDocBench/ba/ba3c299afb5460b0c8e1c11e1bd79291c2b23d9e3cce7eff9668e3885e77cef8.jpg",
"images/SciDocBench/b6/b68565f85d77bfaa021defe96d516e81b958eb90dd1fddd623aae7ef46938a40.jpg",
"images/SciDocBench/9c/9c7ee4c367434feddb6146e8934d60afff90357b53298b50d93286e951cb529a.jpg",
"images/SciDocBench/ad/ad729... | |
409 | 仔细阅读论文。仅使用这篇论文。不要解释。恰好输出一行，作为有效的 JSON 对象：
{"BITS":"<BITS>","NUMS":<NUMS>,"CODE":"<CODE>"}
规则：
1) BITS 是一个 10 字符的比特串。当且仅当该断言被论文完全支持时使用 1；否则使用 0。
A. 定理 1 给出了一个精确构造，其中 L = log2 k + 1 层且 d = kN(3 + log2 k)。
B. 定理 2 的下界仅针对 F 陈述，并排除了 F^cyc。
C. Algorithm 1 在每个阶段内先更新 theta_KQ，再更新 theta_Psi。
D. 定理 3 假设 k = 2^(L-1)。
E. 定理 4 ... | {"BITS": "1011110111", "NUMS": [1, 2, 3, 2, 2, 2, 5, 16, 5, 4, 8, 5, 100000], "CODE": "ABCDEBCD"} | C3 | json_match | [
"9d/9d503a00d824f8de08b3a5baf730d10180fe0890ec504c69d8cc2344f9662451.jpg",
"a0/a071944f27d78a597065eec5272a7e2a2bdc8558c626cefd9a55337fa671983f.jpg",
"a6/a6547183f805fd3f039e78f22c65e4cb5302ad3e030e71d37d17d540d69e16bc.jpg",
"85/85c0cced23a91fc64a86586118454b3ef76f1252810cb8abb0978150f2817260.jpg",
"fb/fba7... | [{"text":"an instance of computing an interleaved group product, a problem also studied in communication complexity [24]. However, in neither setting has the question of learnability been investigated.\n\nOur task is also related to the semiautomaton simulation task of Liu et al. [39], which takes as input a sequence o... | easy-to-hard-transformer_002__zh_interleave | easy-to-hard-transformer_002 | zh_interleave | interleave | json | {"reasoning": "BITS: A=1 (Thm1 exact), B=0 (lower bound covers both F and F^cyc), C=1 (Alg1 updates KQ then Psi), D=1 (Thm3 assumes k=2^(L-1)), E=1 (Thm4 proof: zero layers => zero gradients), F=1 (Fig3 mid: k=16,N=5 from caption), G=0 (100000 mixed from {1,2,4}-fold is correct but paper says the same model is tested, ... | easy-to-hard-transformer | cs | zh | easy-to-hard-transformer_002 | [
"images/SciDocBench/9d/9d503a00d824f8de08b3a5baf730d10180fe0890ec504c69d8cc2344f9662451.jpg",
"images/SciDocBench/a0/a071944f27d78a597065eec5272a7e2a2bdc8558c626cefd9a55337fa671983f.jpg",
"images/SciDocBench/a6/a6547183f805fd3f039e78f22c65e4cb5302ad3e030e71d37d17d540d69e16bc.jpg",
"images/SciDocBench/85/85c0c... | |
410 | 给定 4 个文档的页面图像。
- Document 1 (ppo): Images 1–2
- Document 2 (grpo): Images 3–5
- Document 3 (gspo): Images 6–7
- Document 4 (cispo): Images 8–9
你是一名综述作者，正在用统一记号整合四篇 RLHF 策略优化论文。
=== 统一记号 ===
- q: 输入 prompt/question
- {o_i}_{i=1}^G: 从 π_old(·|q) 采样得到的 G 个 responses 组成的组
- |o_i|: response o_i 的 token 长度
- π_θ: 当前（可训练的）p... | {"PPO_LLM.outer_norm": "1/|o_i|", "PPO_LLM.KL_location": "in_reward", "PPO_LLM.KL_estimator": "log_ratio", "PPO_LLM.IS_ratio_length_norm": "no", "PPO_LLM.IS_weight_sg": "no", "PPO_LLM.needs_value_model": "yes", "PPO_LLM.clip_bound_symmetry": "symmetric", "PPO_LLM.advantage_per_token_uniform": "no", "PPO_LLM.objective_s... | D1 | json_match | [
"5a/5ad2a24c1609a9ddeb84f25c73f030fa84fa3bc5a3aed8c1b96ead662219954f.jpg",
"00/00d7ac19cd49a90be39def0f0358b21431219551f6842623d014d9329076f189.jpg",
"94/94fe8ed323dcc80deb1ec3d3cce3c0e3dca3b2a2be3a85db33c09b5b799a5e71.jpg"
] | [{"text":"## 2 Background: Policy Optimization\n\n## 2.1 Policy Gradient Methods\n\nPolicy gradient methods work by computing an estimator of the policy gradient and plugging it into a stochastic gradient ascent algorithm. The most commonly used gradient estimator has the form\n\nwhere $\\pi_{\\theta}$ is a stochastic ... | rlhf-formula-compare_001__zh_interleave | rlhf-formula-compare_001 | zh_interleave | interleave | json | {"reasoning": "PPO_LLM: 1/|o| norm (Eq.1), KL in reward (Eq.2: r_t -= β log π_θ/π_ref → log_ratio), token IS ρ_{i,t} (not length-normalized), no sg, needs V_ψ via GAE, symmetric clip, GAE gives per-token A_t (not uniform), innermost sum over tokens, single response (no group G). GRPO: (1/G)*(1/|o_i|) norm (Eq.3), KL in... | ["ppo", "grpo", "gspo", "cispo"] | cs | zh | rlhf-formula-compare_001 | [
"images/SciDocBench/5a/5ad2a24c1609a9ddeb84f25c73f030fa84fa3bc5a3aed8c1b96ead662219954f.jpg",
"images/SciDocBench/00/00d7ac19cd49a90be39def0f0358b21431219551f6842623d014d9329076f189.jpg",
"images/SciDocBench/94/94fe8ed323dcc80deb1ec3d3cce3c0e3dca3b2a2be3a85db33c09b5b799a5e71.jpg"
] | |
411 | 给定 2 个作为页面图像的文档。
- Document 1 (eagle25): Image 1
- Document 2 (mammoth-vl): Image 2
Document 1 介绍 Eagle 2.5，Document 2 介绍 MAmmoTH-VL。
从 Eagle 2.5 论文的表 1 中，提取用作训练数据来源的数据集名称完整列表。从 MAmmoTH-VL 论文的图 3 中，提取 MAmmoTH-VL-Instruct (12M) 集合中的数据集名称完整列表。
仅使用精确的不区分大小写的名称匹配来计算这两个列表的交集——不使用缩写或替代拼写。
将结果作为按不区分大小写的字母顺序排序的 \begin{itemi... | "\\documentclass[border=8pt]{standalone}\n\\usepackage{enumitem}\n\\begin{document}\n\\begin{minipage}{10cm}\n\\textbf{Intersection of Eagle-2.5 and MAmmoTH-VL-Instruct (12M) source datasets}\n\\begin{itemize}\n \\item DocVQA\n \\item L-Video-NeXT-Q\n\\end{itemize}\n\\end{minipage}\n\\end{document}" | D2 | judge | [
"5c/5c4acf79119ffa78e8482a310ba7cb09bc52a652f5ec1cc49e815492943d0778.jpg",
"c2/c2e07a1446cdb8c7e2ddfdc7e76d40dfa3cb18010f8f2d6cd46a86d334f7d1d1.jpg"
] | [{"text":"Eagle 2.5: Boosting Long-Context Post-Training for Frontier Vision-Language Models\n\n<table><tr><td>Category</td><td>Dataset</td></tr><tr><td>Video Classification</td><td>Kinetics710 Carreira and Zisserman (2017); Wang et al. (2024), Something-Something-v2 Goyal et al. (2017), ActivityNet Caba Heilbron et al... | eagle25-mammoth-vl_001__zh_interleave | eagle25-mammoth-vl_001 | zh_interleave | interleave | text | You are an expert evaluator for a cross-document dataset comparison benchmark.
A model was asked to compute the intersection of dataset names between Eagle 2.5 (Table 1) and MAmmoTH-VL-Instruct 12M (Figure 3), then output a LaTeX itemize list.
Reference answer:
{answer}
Model prediction:
{prediction}
Score the predi... | {"reasoning": "Eagle 2.5 Table 1 datasets: Kinetics710, Something-Something-v2, ActivityNet, HACS Segment, COIN, HIREST, FineAction, PortraitMode-400, Ego4D-MQ, Perception-Test, Charade-STA, QVHighlight, Ego4D-NLQ, Didemo, QueryD, MedVidQA, Youcook2, FineVideo, EgoExoLearn, ActivityNet-RTL, TVQA, CLEVRER, L-Video-NeXT-... | ["eagle25", "mammoth-vl"] | cs | zh | eagle25-mammoth-vl_001 | [
"images/SciDocBench/5c/5c4acf79119ffa78e8482a310ba7cb09bc52a652f5ec1cc49e815492943d0778.jpg",
"images/SciDocBench/c2/c2e07a1446cdb8c7e2ddfdc7e76d40dfa3cb18010f8f2d6cd46a86d334f7d1d1.jpg"
] |
412 | 你会看到 3 份文档的页面图像。
- Document 1 (cambrian1) 介绍 Cambrian-1 视觉指令调优论文。
- Document 2 (llavaov) 介绍 LLaVA-OneVision。
- Document 3 (mammoth-vl) 介绍 MAmmoTH-VL。
按顺序列出这三篇论文分别引入的官方模型/方法名称。
仅输出 JSON：
{"document_1_name": "<name>", "document_2_name": "<name>", "document_3_name": "<name>"} | {"document_1_name": "Cambrian-1", "document_2_name": "LLaVA-OneVision", "document_3_name": "MAmmoTH-VL"} | D2 | json_match | [
"37/3715439c3d2a59a7b5950e50beaec883bf13bc90f014bfb96b37ffc362d0436d.jpg",
"0e/0e7c7430efa1caf0d9addaef9002a8e487edac8cfb08ca03ac22ec0824a6d958.jpg",
"6e/6e8bb0f7b968483b4d15c9023a2601f93f72d754c21cbf2f6900453e9a1f87ce.jpg",
"ca/caf2660ba4236e638736ab0b9fcc573d6fd7699a4f0aa7c1a6d5ed6a572b11de.jpg",
"27/271e... | [{"text":"cross-attention to occur multiple times by inserting our proposal throughout the LLM layers—allowing consistent access to the uncompressed visual information (see Fig. 8-right).\n\nHyperparameters To flexibly modulate capacity, we introduce two hyperparameters D and G, which indicate the number of cross-atten... | cambrian-llavaov-mammoth_001__zh_interleave | cambrian-llavaov-mammoth_001 | zh_interleave | interleave | json | {"reasoning": "Cambrian-1 Figure 9 General (yellow/orange): HatefulMemes, ALLaVA, OODVQA, Q-Instruct, SketchyVQA, DocVQA, LNQA, LVIS-Instruct4V, VisualGenome, VQAv2, GQA, A-OKVQA, TextCaps, LLaVA150K, ShareGPT, OCRVQA, LLAVAR, ST-VQA, IconQA, RefCOCO, VizWiz, ChartQA, WTQ. LLaVA-OV Figure 4 General (yellow/orange): Hat... | ["cambrian1", "llavaov", "mammoth-vl"] | cs | zh | cambrian-llavaov-mammoth_001 | [
"images/SciDocBench/37/3715439c3d2a59a7b5950e50beaec883bf13bc90f014bfb96b37ffc362d0436d.jpg",
"images/SciDocBench/0e/0e7c7430efa1caf0d9addaef9002a8e487edac8cfb08ca03ac22ec0824a6d958.jpg",
"images/SciDocBench/6e/6e8bb0f7b968483b4d15c9023a2601f93f72d754c21cbf2f6900453e9a1f87ce.jpg",
"images/SciDocBench/ca/caf26... | |
413 | 你将获得 2 个文档，形式为页面图像。
- Document 1 (spatial-dreamer): 图像 1–17
- Document 2 (spatial-ladder): 图像 18–39
每个文档都介绍了一种改进 Large Vision-language Models 空间推理能力的方法。你的任务是总结四个模型（Qwen2.5-VL-3B, Qwen2.5-VL-7B, SpatialLadder-3B, SpatialDreamer）在 VSI-Bench 所有子任务上的性能，并以 LaTeX 格式表格输出。
严格遵循以下规则：
总结规则：
- 如果某个模型在两个文档中都在所有 VSI-Bench 子任务上进行了... | {"Qwen2.5-VL-3B": {"Obj_Cnt": 32.9, "Obj_Size": 17.3, "Room_Size": 31.5, "Abs_Dist": 22.1, "Rel_Dist": 32.8, "Appr_Order": 28.5, "Route_Plan": 26.3, "Rel_Dir": 44.2, "Avg": 29.4}, "Qwen2.5-VL-7B": {"Obj_Cnt": 42.2, "Obj_Size": 46.0, "Room_Size": 25.9, "Abs_Dist": 15.0, "Rel_Dist": 37.5, "Appr_Order": 31.8, "Route_Plan"... | D3 | judge | [
"55/55d9ce2e68f2fcbc9b043e70a8335677a3c479a71714eb745ea28773085734e9.jpg",
"65/6542efd06cf56cc70177149f731da9cf9d6292f63d030775cfc9c501681b30b6.jpg",
"43/43084608c3e6300abf7bfb4a5be7d1bb96bc3525ed775a9ef350a898e1f5fecf.jpg",
"c8/c8748acadb8dc952d7da8e77abdb3e26c1e8b571a8dc59bf9b258d51008e9054.jpg",
"ca/ca20... | [{"text":"# SpatialDreamer: Incentivizing Spatial Reasoning via Active Mental Imagery"},{"image":"55/55d9ce2e68f2fcbc9b043e70a8335677a3c479a71714eb745ea28773085734e9.jpg"},{"image":"65/6542efd06cf56cc70177149f731da9cf9d6292f63d030775cfc9c501681b30b6.jpg"},{"text":"## 2.1. Spatial Reasoning"},{"image":"43/43084608c3e630... | spatial-dreamer-ladder_001__zh_interleave | spatial-dreamer-ladder_001 | zh_interleave | interleave | json | You are an expert evaluator for a cross-document table extraction benchmark.
The task required the model to:
1. Read VSI-Bench results from two papers (SpatialDreamer and SpatialLadder)
2. Average Qwen2.5-VL-7B values across both papers (it appears in both)
3. Report single-paper values for Qwen2.5-VL-3B, SpatialLadde... | {"reasoning": "Cross-document D3 task: consolidate VSI-Bench results from SpatialDreamer (Table 1) and SpatialLadder (Table 3) into one LaTeX table with 4 models × 8 subtasks + Avg. Derivation: Qwen2.5-VL-3B and Qwen2.5-VL-7B baselines appear in both papers — SpatialLadder Table 3 provides the 8-subtask breakdown (Obj_... | ["spatial-dreamer", "spatial-ladder"] | cs | zh | spatial-dreamer-ladder_001 | [
"images/SciDocBench/55/55d9ce2e68f2fcbc9b043e70a8335677a3c479a71714eb745ea28773085734e9.jpg",
"images/SciDocBench/65/6542efd06cf56cc70177149f731da9cf9d6292f63d030775cfc9c501681b30b6.jpg",
"images/SciDocBench/43/43084608c3e6300abf7bfb4a5be7d1bb96bc3525ed775a9ef350a898e1f5fecf.jpg",
"images/SciDocBench/c8/c8748... |
414 | 你将获得 2 份文档的页面图像。
- 文档 1 (ts-llava)：图像 1–15
- 文档 2 (pllava)：图像 16–32
文档 1 介绍 TS-LLaVA，文档 2 介绍 PLLaVA。每份文档都包含一个报告模型在 MVBench 上性能的表格。你的任务是将相关结果整合为一个统一的 LaTeX 格式表格。严格遵循以下规则：
总结规则：
- 包含在两份文档中找到的所有 TS-LLaVA 和 PLLaVA 模型变体。
- 包含任何在两份文档中名称、LLM Size 和 Vision Encoder 完全相同的其他模型。
- 如果某个模型在两份文档中的分数不同，则对每个子任务和总体平均值报告两者的平均值，四舍五入到一位小数... | {"row_order": "TS-LLaVA-7B, TS-LLaVA-34B, PLLaVA-7B, PLLaVA-13B, PLLaVA-34B, VideoChat-7B, VideoChat2-7B", "excluded_models": "Video-LLaMA, LLaMA-Adapter, Video-ChatGPT, ST-LLM, GPT-4V appear in only one paper and are not TS-LLaVA/PLLaVA variants — must NOT be included", "pllava_column_remapping": "PLLaVA paper column ... | D3 | judge | [
"25/25260c736f22583b4bdc96da6cbcd0a914918b841694fb5b6be6c85f5b3e3b95.jpg",
"7b/7b0d8ed77aea86f9fe369872eab7ec7f90f6bc4b07cf660c0cf5b085998ef91f.jpg",
"75/7585eab16b6576d1f569b234cc5d1ada4369746a8b48b008fb3a03d635f28b31.jpg",
"48/48773539010e9ef5ee51bec3afec82c0666bf4a4dcfab9f7221bd96e6ab1265f.jpg",
"a2/a290... | [{"text":"# TS-LLaVA: Constructing Visual Tokens through Thumbnail-and-Sampling for Training-Free Video Large Language Models"},{"image":"25/25260c736f22583b4bdc96da6cbcd0a914918b841694fb5b6be6c85f5b3e3b95.jpg"},{"image":"7b/7b0d8ed77aea86f9fe369872eab7ec7f90f6bc4b07cf660c0cf5b085998ef91f.jpg"},{"text":"<table><tr><td ... | ts-llava-pllava_001__zh_interleave | ts-llava-pllava_001 | zh_interleave | interleave | text | You are an expert evaluator for a cross-document table consolidation benchmark.
The task required the model to:
1. Extract MVBench results from TS-LLaVA (Table 5a, page 7) and PLLaVA (Table 3, page 11)
2. Include TS-LLaVA variants, PLLaVA variants, and shared models only
3. Remap PLLaVA's different column order to the... | {"reasoning": "TS-LLaVA Table 5a (page 7) uses column order AA,AC,AL,AP,AS,CO,CI,EN,ER,FA,FP,MA,MC,MD,OE,OI,OS,ST,SC,UA. PLLaVA Table 3 (page 11) uses column order AS,AP,AA,FA,UA,OE,OI,OS,MD,AL,ST,AC,MC,MA,SC,FP,CO,EN,ER,CI. VideoChat identical in both (35.5 avg). VideoChat2: TS has 60.4, PLLaVA has 51.1 — must average... | ["ts-llava", "pllava"] | cs | zh | ts-llava-pllava_001 | [
"images/SciDocBench/25/25260c736f22583b4bdc96da6cbcd0a914918b841694fb5b6be6c85f5b3e3b95.jpg",
"images/SciDocBench/7b/7b0d8ed77aea86f9fe369872eab7ec7f90f6bc4b07cf660c0cf5b085998ef91f.jpg",
"images/SciDocBench/75/7585eab16b6576d1f569b234cc5d1ada4369746a8b48b008fb3a03d635f28b31.jpg",
"images/SciDocBench/48/48773... |
415 | 查看所提供论文的 Figure 3，其中比较了两条训练奖励曲线。识别被比较的两种方法以及每条曲线的颜色。
仅输出 JSON：
{"method_1": "<method name>", "method_1_color": "<color>", "method_2": "<method name>", "method_2_color": "<color>"}
“purple” 或 “violet” 可互换使用；“orange” 同理。 | {"method_1": "GRPO w/ Routing Replay", "method_1_color": "purple", "method_2": "GRPO w/o Routing Replay", "method_2_color": "orange"} | E1 | json_match | [
"d9/d902311adc6733c74a88da896cf655147fe1c030b62a2021516b17ca038eb3d1.jpg",
"05/05f42e65631a60881251cc61407600298a7ebe5c994a81e07e9fc6e3929a870d.jpg",
"89/89dd8ee64f5431ec969dda6ada5253dadf7fe77604ccf0b0e01b52d440834b80.jpg",
"c3/c3e28311b7757dc4460986387c89d429abbb67755589bb8651ab5759dda8ce05.jpg",
"82/82c3... | [{"text":"# Group Sequence Policy Optimization\n\nChujie Zheng\\* Shixuan Liu Mingze Li Xiong-Hui Chen Bowen Yu\\* Chang Gao Kai Dang Yuqiong Liu Rui Men An Yang Jingren Zhou Junyang Lin Qwen Team, Alibaba Inc.\n\n## Abstract\n\nThis paper introduces Group Sequence Policy Optimization (GSPO), our stable, efficient, and... | gspo_002__zh_interleave | gspo_002 | zh_interleave | interleave | json | {"reasoning": "Figure 3 page 6 of GSPO paper. Purple=GRPO w/ Routing Replay stays roughly 0.35-0.55 with noise. Orange=GRPO w/o Routing Replay starts ~0.40 then declines toward 0.25-0.30. Both curves are noisy training curves over training compute steps."} | gspo | cs | zh | gspo_002 | [
"images/SciDocBench/d9/d902311adc6733c74a88da896cf655147fe1c030b62a2021516b17ca038eb3d1.jpg",
"images/SciDocBench/05/05f42e65631a60881251cc61407600298a7ebe5c994a81e07e9fc6e3929a870d.jpg",
"images/SciDocBench/89/89dd8ee64f5431ec969dda6ada5253dadf7fe77604ccf0b0e01b52d440834b80.jpg",
"images/SciDocBench/c3/c3e28... | |
416 | 你是一名专家级数据可视化工程师和 Python 开发者（Matplotlib/Seaborn 专家）。
目标：你的目标是将所提供的科学图表图像逆向工程为一个可执行的 Python 脚本。
黄金标准：
1. 数据保真度：你代码中的数据点必须尽可能接近图像中的像素。
2. 视觉保真度：生成的图在以下方面必须与原始图像看起来完全一致：
- 坐标轴标签（包括 LaTeX 数学方程）
- 图例位置和文本
- 线型（虚线、实线、点线）、标记（圆形、三角形）和颜色
- 坐标轴范围和刻度（对数 vs. 线性）
- 网格线和背景样式
输出：一个单独的、自包含的 Python 代码块。
输入：所提供论文的图 5。 | "Figure 5: Scaling Curve Across Methods on MMStar (Log Scale). Three curves plotted on log-scale x-axis. Best-of-N Search (green dashed, circles): 4 points, y=[57.6, 58.7, 60.5, 60.1], plateaus after 10000s. Stage-wise Beam Search (blue dashed, circles): 5 points, y=[57.6, 58.3, 60.8, 61.0, 61.2], only 58.3 and 61.2 ar... | E1 | judge | [
"6b/6bf75f6e320d847fa73d249f15a05f836fd2d2a38667fa8fc9d9c404e25ff936.jpg",
"fe/fe5f4e74f960728d794b0c94ffdfb33368e4f348e42271793c564ebc0568f7eb.jpg",
"c7/c79150158726751519cc721bbe2365ec6b090b279498186e386f6c64a587261e.jpg",
"b7/b72757d02439fb5d966ad73427272b7b9f1402695f460ed2b0e48de36be0a150.jpg",
"b7/b72a... | [{"text":"# LLaVA-CoT: Let Vision Language Models Reason Step-by-Step\n\nLarge language models have demonstrated substantial advancements in reasoning capabilities. However, current Vision-Language Models (VLMs) often struggle to perform systematic and structured reasoning, especially when handling complex visual quest... | llavacot_002__zh_interleave | llavacot_002 | zh_interleave | interleave | text | You are evaluating a Python Matplotlib script that attempts to reproduce Figure 5 from the LLaVA-CoT paper.
Ground truth data:
- Best-of-N Search (green dashed, circle markers): 4 data points, y-values = [57.6, 58.7, 60.5, 60.1] (plateaus at the end)
- Stage-wise Beam Search (blue dashed, circle markers): 5 data point... | {"reasoning": "Figure 5 page 7 of LLaVA-CoT. GT data: Best-of-N=[57.6,58.7,60.5,60.1] (4pts), Beam=[57.6,58.3,60.8,61.0,61.2] (5pts, only 58.3/61.2 annotated in figure), Retracing=[60.8,61.0,62.3] (3pts, starts later on x-axis). Key traps: colors (red/blue/green), Retracing solid vs others dashed, Retracing starts at h... | llavacot | cs | zh | llavacot_002 | [
"images/SciDocBench/6b/6bf75f6e320d847fa73d249f15a05f836fd2d2a38667fa8fc9d9c404e25ff936.jpg",
"images/SciDocBench/fe/fe5f4e74f960728d794b0c94ffdfb33368e4f348e42271793c564ebc0568f7eb.jpg",
"images/SciDocBench/c7/c79150158726751519cc721bbe2365ec6b090b279498186e386f6c64a587261e.jpg",
"images/SciDocBench/b7/b7275... |
417 | 你会得到一篇研究论文（PDF）。在第 3.2 节中，该论文使用流程图（Figure 1）描述了一个图像预处理流水线。请将该流水线实现为一个 Python 脚本。
约束：
- 允许使用的库：仅 PIL (Pillow)。不要使用 OpenCV、numpy 或任何其他库。
- 你的脚本必须接受一个输入图像路径和一个输出图像路径作为命令行参数。
所需接口：
Usage: python solution.py input.png output.png
仔细阅读 Figure 1，并严格按照所示顺序实现其中显示的操作。
在代码之前，包括：
{"evidence": "<describe each operation you see... | {"input_path": "pdfs/constraintfollow/image.png", "reference_script": "import sys\nfrom PIL import Image\n\nimg = Image.open(sys.argv[1])\nw, h = img.size\nleft = (w - 240) // 2\ntop = (h - 200) // 2\nimg = img.crop((left, top, left + 240, top + 200))\nimg = img.convert(\"L\")\nimg = img.transpose(Image.ROTATE_270)\nim... | E2 | exec_match | [
"81/81416cebf43ecd40861b8dd0dfa6c836c88af936e975e79b7415c58330975326.jpg"
] | [{"text":"# ConstraintFollow: Improving Multimodal Instruction Following via Constraint-Aware Preference Optimization\n\nAnonymous\n\n## Abstract\n\nInstruction following (IF) is a core capability of multimodal large language models (MLLMs). Existing IF training data is scarce and benchmarks remain saturated. We presen... | constraintfollow_001__zh_interleave | constraintfollow_001 | zh_interleave | interleave | text | {"reasoning": "E2 exec_match: implement the Section 3.2 / Figure 1 preprocessing pipeline. Derivation from figure: step 1 — center-crop to 240×200; step 2 — convert to grayscale (L mode); step 3 — rotate 270° (counterclockwise). The reference script: open image → compute center crop box ((w-240)//2, (h-200)//2, ...) → ... | constraintfollow | cs | zh | constraintfollow_001 | [
"images/SciDocBench/81/81416cebf43ecd40861b8dd0dfa6c836c88af936e975e79b7415c58330975326.jpg"
] | |
418 | 在 MM-IFEngine (mmifengine) 论文中，确定列出该框架中使用的评估方法顺序的章节编号。
仅输出 JSON：
{"section_number": "<e.g. 3.1>"} | {"section_number": "3.1"} | F1 | json_match | [
"0c/0cd5e8adad75b92f9294d32075911185bedccb5406031bf52d843da571b0ad34.jpg",
"5a/5a15a7798b801c703df7f429e519480c38a30bf85195c4c42df0126c4e613ade.jpg",
"89/89d0d615fcf4a1b1145fc2e83011f6b468bac61fe3fb94acce0fb2af0e9dd00d.jpg",
"bd/bdb696f537b4e218f37538b68387de9a2c6cf7b6e024f1442a1f7bf6e1b3ce07.jpg",
"1e/1ec5... | [{"text":"# MM-IFEngine: Towards Multimodal Instruction Following\n\nShengyuan Ding $^{1,2*}$ , Shenxi Wu $^{1,2*}$ , Xiangyu Zhao $^{2,3}$ , Yuhang Zang $^{2\\boxtimes}$ , Haodong Duan $^{2}$ , Xiaoyi Dong $^{2}$ , Pan Zhang $^{2}$ , Yuhang Cao $^{2}$ , Dahua Lin $^{2,4,5}$ , Jiaqi Wang $^{2,6\\boxtimes}$ $^{1}$ Fudan... | mmifengine_003__zh_interleave | mmifengine_003 | zh_interleave | interleave | json | {"reasoning": "F1 paper-code alignment: multi-step reasoning chain. Step 1: Pipeline figure (Figure 2, page 3) shows the overall MM-IFEngine method — the first evaluation method in the pipeline is 'Rule-based Evaluation'. Step 2: In the code repo, eval_mmifeval/ directory contains rule-based evaluation scripts; eval_mm... | mmifengine | cs | zh | mmifengine_003 | [
"images/SciDocBench/0c/0cd5e8adad75b92f9294d32075911185bedccb5406031bf52d843da571b0ad34.jpg",
"images/SciDocBench/5a/5a15a7798b801c703df7f429e519480c38a30bf85195c4c42df0126c4e613ade.jpg",
"images/SciDocBench/89/89d0d615fcf4a1b1145fc2e83011f6b468bac61fe3fb94acce0fb2af0e9dd00d.jpg",
"images/SciDocBench/bd/bdb69... | |
419 | 你将获得 5 个文档，形式为页面图像。
- 文档 1 (data_sample)：图像 1 包含多个数据样本的截图，这些截图垂直排列。
数一数图像 1 中显示了多少个不同的数据样本（从上到下编号）。
仅输出 JSON：
{"sample_count": <integer>} | {"sample_count": 6} | G1 | json_match | [
"0c/0cd5e8adad75b92f9294d32075911185bedccb5406031bf52d843da571b0ad34.jpg",
"45/453b57cd8c06894788c490a43dac2dac648aa3c66d0d2af8eef362d066ac0025.jpg",
"e8/e8e873eae931bf03ff595a2113c40540f0f12856998a8b915fb2df05b0f79237.jpg",
"ef/efe9c95ab22c2ba555742a187d827a7ea662d75f31648bdcc0c1a1778f15905e.jpg",
"ac/ac02... | [{"text":"<image> What should be the next step after navigating through the Safari browser to access a webpage on the iPhone?\n\nI've been sorting through these old photos of bronze vessels for my archaeology lab assignment. There are ten items labeled a to j and I need to divide them into five distinct groups based on... | data_sample_001__zh_interleave | data_sample_001 | zh_interleave | interleave | json | {"reasoning": "G1 dataset provenance: match 6 visual samples to source papers. Derivation: Sample 1 shows a video frame sequence with future frame prediction annotations → matches TwiFF (Think With Future Frames) which focuses on dynamic visual reasoning with future frame prediction. Sample 2 shows a complex web UI int... | ["data_sample", "mmifengine", "agentvista", "twiff", "spatial-ssrl"] | cs | zh | data_sample_001 | [
"images/SciDocBench/0c/0cd5e8adad75b92f9294d32075911185bedccb5406031bf52d843da571b0ad34.jpg",
"images/SciDocBench/45/453b57cd8c06894788c490a43dac2dac648aa3c66d0d2af8eef362d066ac0025.jpg",
"images/SciDocBench/e8/e8e873eae931bf03ff595a2113c40540f0f12856998a8b915fb2df05b0f79237.jpg",
"images/SciDocBench/ef/efe9c... | |
420 | 你将获得 2 份文档的页面图像。
- Document 1 (poet2): Image 1
- Document 2 (progen2): Images 2–3
下面是从混合数据池中提取的 4 个数据集条目。对于每个条目，请识别：
- Source Paper（ProGen2 或 PoET-2）。
- 数据所来源的 Primary Database。
- 文本中提到的 Key Preprocessing Step（例如，聚类阈值或过滤方法）。
[Dataset Entries]
Entry A: 来自 80 项研究的 1.5B 抗体序列集合，后来以 85% 序列同一性进行聚类以减少冗余。
Entry B: UniRef90 与... | "[{\"entry\": \"Entry A\", \"source_paper\": \"ProGen2: Exploring the Boundaries of Protein Language Models\", \"primary_database\": \"Observed Antibody Space (OAS)\", \"key_preprocessing\": \"Clustered at 85% sequence identity using Linclust\"}, {\"entry\": \"Entry B\", \"source_paper\": \"ProGen2: Exploring the Bound... | G1 | judge | [
"b5/b5cfcf2226ed184d644f11b55f8194eabed639030a751a69021958c4e5c0b0ea.jpg"
] | [{"text":"Additionally, when the prompt includes a query, the input embedding of the decoder is modified to encode which protein in the prompt should be used as the query. The modified input embedding of each residue is simply the average of the unmodified input embedding, and the embedding, produced by the encoder, of... | poet2-progen2_001__zh_interleave | poet2-progen2_001 | zh_interleave | interleave | text | You are evaluating a model's answer to a dataset attribution question.
The correct answer has 4 entries:
Entry A: source_paper=ProGen2, primary_database=OAS (Observed Antibody Space), key_preprocessing=clustered at 85% identity using Linclust
Entry B: source_paper=ProGen2, primary_database=Uniref90+BFD30, key_preproce... | {"reasoning": "G1 dataset provenance: classify 4 entries to PoET-2 or ProGen2. Derivation: Entry A describes 1.5B antibody sequences from 80 studies, clustered at 85% identity using Linclust → ProGen2 Section 2.1 describes using OAS (Observed Antibody Space) with Linclust clustering at 85%. Entry B describes mixing Uni... | ["poet2", "progen2"] | bio | zh | poet2-progen2_001 | [
"images/SciDocBench/b5/b5cfcf2226ed184d644f11b55f8194eabed639030a751a69021958c4e5c0b0ea.jpg"
] |
421 | 你将获得 2 份文档的页面图像。
- 文档 1 (sle-glyco): 图像 1–9
- 文档 2 (igg-glycans): 图像 10–44
在文档 1 中，有两种特定的 IgG 糖型被确定为与 SLE (Systemic Lupus Erythematosus) 具有强相关性。列出这两种糖型的名称。
仅输出 JSON：
{"glycoforms": ["<glycoform_1>", "<glycoform_2>"]} | {"glycoforms": ["IgG1Gal", "IgG3/4Bis"]} | D3 | json_match | [
"fa/fa35bf20bcf1489ffb669d25fbe338c9136dcc784ea37f7092bb4e13b462e6f3.jpg",
"62/62544462f1759f4c63a8286f820bd8e045ef4caa290df7d7148ba5ef2656b5d8.jpg",
"d3/d3468d84c1d6e6aadc84d13473101dd0aa62802ae6db17d88ef2ac5e6a92d07e.jpg",
"a9/a9edc01ac0140e9ed56a7e86f51941057443a7c6d0389732acdac33a43d69ef7.jpg",
"7a/7a77... | [{"text":"# Glycosylation of anti-dsDNA IgG correlates with organ involvement in treatment-naïve patients with systemic lupus erythematosus\n\nof the underlying pathophysiological mechanisms. Recent advances in immunology have shed light on the critical role of immunoglobulin G (IgG) glycosylation in the pathogenesis o... | sle-glyco-igg_001__zh_interleave | sle-glyco-igg_001 | zh_interleave | interleave | json | {"reasoning": "doc1 (sle-glyco) identifies fucosylation of anti-dsDNA IgG1 as strongly correlated with SLE. Combined with sialylation requirement, relevant substances are fucosylated+sialylated IgG glycans (FA*S1/S2 with fucose). Parenthesized peaks 16a/16b/17 are uncertain assignments in doc2 Figure 2."} | ["sle-glyco", "igg-glycans"] | biology | zh | sle-glyco-igg_001 | [
"images/SciDocBench/fa/fa35bf20bcf1489ffb669d25fbe338c9136dcc784ea37f7092bb4e13b462e6f3.jpg",
"images/SciDocBench/62/62544462f1759f4c63a8286f820bd8e045ef4caa290df7d7148ba5ef2656b5d8.jpg",
"images/SciDocBench/d3/d3468d84c1d6e6aadc84d13473101dd0aa62802ae6db17d88ef2ac5e6a92d07e.jpg",
"images/SciDocBench/a9/a9edc... | |
422 | 我正在阅读所附的关于 Stairway codes 的论文。请仅复现 Figure 3 中的 Weight-8 case（Fig. 3 的底行），作为一个 LaTeX/TikZ 图。
该图应包含由一个 “≃” 符号连接的两个部分：
请确保输出能够作为一个 standalone LaTeX 文档干净地编译。 | "\\documentclass[border=12pt]{standalone}\n\\usepackage{tikz}\n\\usetikzlibrary{arrows.meta, calc}\n\n\\tikzset{\n sp/.style={\n circle,\n fill=pink!50!red!30,\n draw=black!60,\n line width=0.55pt,\n minimum size=7mm,\n inner sep=0pt\n },\n spbig/.style={sp, minimum size=11mm},\n bond/.style={line... | E1 | judge | [
"ef/efb250ff574a7a96568aecc2a43ea62adf262fb011d6ab3ebe106599797a1829.jpg",
"78/781f80ed1814fa0219c5555fbcfa505e2d693e680646b4cef11b9540e2a7ca12.jpg",
"17/176c94e327128b847104403af5f762ac09ea4e58c7134415389f888d4dd7bbae.jpg",
"29/29d5b46e34a34ed7f977a57a40e7f616df13fb3ce58046c5651a5c7e38a73315.jpg",
"c3/c3d8... | [{"text":"# Stairway Codes: Floquetifying Bivariate Bicycle Codes and Beyond\n\n$^{1}$ AWS Center for Quantum Computing, Pasadena, CA 91125, USA $^{2}$ Racah Institute of Physics, The Hebrew University of Jerusalem, Jerusalem 91904, Givat Ram, Israel (Dated: March 3, 2026)\n\nFloquet codes define fault-tolerant protoco... | stairway_001__zh_interleave | stairway_001 | zh_interleave | interleave | text | You are evaluating a model's LaTeX/TikZ reproduction of the Weight-8 case (bottom row) of Figure 3 from a paper on Stairway codes.
The figure must contain two parts connected by a ≃ symbol:
- LEFT: a single large spider node with 8 directed legs (4 outgoing upward-ish, 4 incoming downward-ish)
- RIGHT: 8 spider nodes ... | {"reasoning": "E1 TikZ reproduction of Figure 3 Weight-8 spider decomposition. Derivation: The bottom row of Figure 3 shows a ZX-calculus tensor network decomposition. Left side: a single large spider node with 8 legs (weight-8). Right side (after ≃): the decomposition into smaller spiders connected by bonds. Reading f... | stairway | cs | zh | stairway_001 | [
"images/SciDocBench/ef/efb250ff574a7a96568aecc2a43ea62adf262fb011d6ab3ebe106599797a1829.jpg",
"images/SciDocBench/78/781f80ed1814fa0219c5555fbcfa505e2d693e680646b4cef11b9540e2a7ca12.jpg",
"images/SciDocBench/17/176c94e327128b847104403af5f762ac09ea4e58c7134415389f888d4dd7bbae.jpg",
"images/SciDocBench/29/29d5b... |
423 | 我附上了一篇物理论文。查看 Figure 1(b)，其中显示了一个 3-regular graph。报告该图的节点数 (N) 和正则性 (k)。
Output JSON only:
{"num_nodes": <integer>, "regularity": <integer>} | {"num_nodes": 22, "regularity": 3} | E1 | json_match | [
"cc/ccd436a940892f67a3ec180bafd24b24ca4a03aa5561cffd43617c19135de4f0.jpg",
"83/837fd4a87d038a2e98dc23b26081174627f636bba6452b4c17142909db9cbbeb.jpg",
"db/dbffdf3530105f3678779b7937a6b92afc9ac59abd6fbe8e56805c4ac7c81fd1.jpg",
"38/38e82190d688cc355a267249ae33c15ca233b0771922b7bd832f9e19a8800d26.jpg",
"2c/2cef... | [{"text":"# Imaginary-time evolution of interacting spin systems in the truncated Wigner approximation\n\nTom Schlegel, $^{1}$ Dennis Breu, $^{1}$ and Michael Fleischhauer $^{1}$\n\n$^{1}$ Department of Physics and Research Center OPTIMAS,\n\nWe present a semiclassical phase-space method to calculate thermal and ground... | itwa-spin_001__zh_interleave | itwa-spin_001 | zh_interleave | interleave | json | {"reasoning": "E1 TikZ reproduction of Figure 1(b) 3-regular graph, N=22. Derivation: reading edge connectivity from the figure node by node — each of the 22 nodes connects to exactly 3 others. The adjacency was extracted by careful inspection: node 1 connects to 3, 12, 13; node 2 connects to 4, 9, 10; node 3 connects ... | itwa-spin | physics | zh | itwa-spin_001 | [
"images/SciDocBench/cc/ccd436a940892f67a3ec180bafd24b24ca4a03aa5561cffd43617c19135de4f0.jpg",
"images/SciDocBench/83/837fd4a87d038a2e98dc23b26081174627f636bba6452b4c17142909db9cbbeb.jpg",
"images/SciDocBench/db/dbffdf3530105f3678779b7937a6b92afc9ac59abd6fbe8e56805c4ac7c81fd1.jpg",
"images/SciDocBench/38/38e82... | |
424 | def conv1x1(in_feat: int, out_feat: int) -> None: ...
def res_transformer_block(feat: int, heads: int) -> None: ...
def down_conv2x2(in_feat: int, out_feat: int, stride: int) -> None: ...
def linear(in_feat: int, out_feat: int) -> None: ...
参考 **Figure 6** 及其图注。仅基于该图，通过按顺序列出每个操作（每行一个操作，并填入所有具体参数）来重构 QFlowNet 的完整顺序前向传播... | "conv1x1(in_feat, 64)\nres_transformer_block(64, 8)\nres_transformer_block(64, 8)\nres_transformer_block(64, 8)\nres_transformer_block(64, 8)\ndown_conv2x2(64, 128, 2)\nres_transformer_block(128, 8)\nres_transformer_block(128, 8)\nres_transformer_block(128, 8)\nres_transformer_block(128, 8)\nconv1x1(128, 256)\nlinear(2... | E2 | judge | [
"40/40c7fb5c8d4ee927082b20d2675e416ed9a8f80e6eb29aaf4ffb81427a15a0cd.jpg",
"2c/2c3ad1919dc94b9a09af47801a116d411286b68382d923329df64c644484737d.jpg",
"36/364fe83f25cfc2f64c66b53a6b0ac2b801fd1e082ddb0255bb35efa55e597dce.jpg",
"9c/9c3c84361f5107e02e921756b9eb7a9a28f40fcc3ad70f5d7efbb9135993a2f8.jpg",
"e8/e85c... | [{"text":"# QFlowNet: Fast, Diverse, and Efficient Unitary Synthesis with Generative Flow Networks\n\nAbstract—Unitary Synthesis, the decomposition of a unitary matrix into a sequence of quantum gates, is a fundamental challenge in quantum compilation. Prevailing reinforcement learning (RL) approaches are often hampere... | qflownet_001__zh_interleave | qflownet_001 | zh_interleave | interleave | text | You are evaluating a model's reconstruction of the QFlowNet forward pass from Figure 6.
The correct sequential forward pass is (one call per line):
{answer}
Model prediction:
{prediction}
Scoring: 13 operations total, each worth 1/13. Award credit for each operation that is correct in both function name and all argu... | {"reasoning": "E2 flowchart reconstruction of QFlowNet forward pass from Figure 6 (Appendix A, page 7). Derivation: reading top-to-bottom from the figure: (1) conv1x1(in_feat, 64) — initial projection; (2-5) four res_transformer_block(64, 8) — first stage with 64 features, 8 attention heads; (6) down_conv2x2(64, 128, 2... | qflownet | cs | zh | qflownet_001 | [
"images/SciDocBench/40/40c7fb5c8d4ee927082b20d2675e416ed9a8f80e6eb29aaf4ffb81427a15a0cd.jpg",
"images/SciDocBench/2c/2c3ad1919dc94b9a09af47801a116d411286b68382d923329df64c644484737d.jpg",
"images/SciDocBench/36/364fe83f25cfc2f64c66b53a6b0ac2b801fd1e082ddb0255bb35efa55e597dce.jpg",
"images/SciDocBench/9c/9c3c8... |
425 | 你将看到四张截图，标记为 A、B、C 和 D。每张截图都包含同一篇科学论文正文中的一段文字。
你的任务是确定这四段文字在论文中出现的顺序，从最早到最晚。这些段落可能来自以下任一部分：Abstract、Introduction、Method、Experiments/Results 或 Conclusion。
仅根据每段文字的内容、功能和写作风格来推断顺序。
严格按以下格式输出一行：
A>B>C>D
这里，“>”表示“在论文中出现得早于”。
不要提供任何解释、推理、章节名称或额外文本。只输出最终顺序。 | "D>C>A>B" | A1 | judge | [] | [{"text":"Implementation Details. We conducted SFT and DPO fine-tuning experiments on two representative MLLMs: Qwen2-VL-7B-Instruct [41] and LLaVA-Next-Llama3-8B [21], using our custom datasets MM-IFInstruct-23k for supervised fine-tuning (SFT) and MM-IFDPO-23k for direct preference optimization (DPO). For the SFT pha... | np_a1_01_001__zh_interleave | np_a1_01_001 | zh_interleave | interleave | text | You are an expert evaluator for ranking tasks. I will provide you with the ground-truth and the model's predicted ranking. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
If the model's predicted ranking is entirely correct, it receiv... | {} | np_a1_01 | scientific | zh | np_a1_01_001 | [] |
426 | 你将获得五张截图，标记为 A、B、C、D 和 E。每张截图都包含同一篇科学论文正文中的一段文字。
你的任务是确定这五段文字在论文中出现的顺序，从最早到最晚。这些段落可能来自以下任一部分：摘要、引言、相关工作、方法、实验/结果或结论。
仅根据每段文字的内容、功能和写作风格来推断顺序。
严格按以下格式输出一行：
A>B>C>D>E
这里，“>”表示“在论文中出现得早于”。
不要提供任何解释、推理、章节名称或额外文本。只输出最终顺序。 | "E>A>D>C>B" | A1 | judge | [] | [{"text":"To construct large-scale spatial QAs in a cost-effective manner, two dominant paradigms have emerged: tool-based and simulation-based approaches. Tool-based approaches incorporate tools within their pipelines. The tools are either open-source models that struggle with spatial understanding [9], or expert mode... | np_a1_02_001__zh_interleave | np_a1_02_001 | zh_interleave | interleave | text | You are an expert evaluator for ranking tasks. I will provide you with the ground-truth and the model's predicted ranking. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
If the model's predicted ranking is entirely correct, it receiv... | {} | np_a1_02 | scientific | zh | np_a1_02_001 | [] |
427 | 你将看到五张截图，标记为 A、B、C、D 和 E。每张截图都包含同一篇科学论文正文中的一段文字。
你的任务是确定这五段文字在论文中出现的顺序，从最早到最晚。这些段落可能来自以下任一部分：Abstract、Introduction、Related Work、Method、Experiments/Results 或 Conclusion。
仅根据每段文字的内容、功能和写作风格来推断顺序。
严格按以下格式输出一行：
A>B>C>D>E
这里，“>”表示“在论文中出现得更早”。
不要提供任何解释、推理、章节名称或额外文本。只输出最终顺序。 | "B>E>D>C>A" | A1 | judge | [] | [{"text":"We present QuarkMedSearch, a framework for long-horizon deep search in the Chinese medical domain. To address the main weaknesses of prior work, we introduce a long-chain data-synthesis pipeline that integrates knowledge graphs with online search, combining long-tail entity sampling, multi-tool online explora... | np_a1_03_001__zh_interleave | np_a1_03_001 | zh_interleave | interleave | text | You are an expert evaluator for ranking tasks. I will provide you with the ground-truth and the model's predicted ranking. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
If the model's predicted ranking is entirely correct, it receiv... | {} | np_a1_03 | scientific | zh | np_a1_03_001 | [] |
428 | 给你五张截图，标记为 A、B、C、D 和 E。每张截图都包含同一篇科学论文正文中的一段文字。
你的任务是确定这五段文字在论文中出现的顺序，从最早到最晚。这些段落可能来自以下任一部分：Abstract、Introduction、Related Work、Method、Experiments/Results 或 Conclusion。
仅根据每段文字的内容、功能和写作风格来推断顺序。
输出严格为一行，格式如下：
A>B>C>D>E
这里，“>” 表示“在论文中出现得早于”。
不要提供任何解释、推理、章节名称或额外文本。只输出最终顺序。 | "A>E>C>B>D" | A1 | judge | [] | [{"text":"Deeper modern architectures are costly to train, making hyperparameter transfer preferable to expensive repeated tuning. Maximal Update Parametrization ( $\\mu P$ ) helps explain why many hyperparameters transfer across width. Yet depth scaling is less understood for modern architectures, whose computation gr... | np_a1_04_001__zh_interleave | np_a1_04_001 | zh_interleave | interleave | text | You are an expert evaluator for ranking tasks. I will provide you with the ground-truth and the model's predicted ranking. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
If the model's predicted ranking is entirely correct, it receiv... | {} | np_a1_04 | scientific | zh | np_a1_04_001 | [] |
429 | 你将看到五张截图，标记为 A、B、C、D 和 E。每张截图都包含同一篇科学论文正文中的一段文字。
你的任务是确定这五段文字在论文中出现的顺序，从最早到最晚。这些段落可能来自以下任一部分：Abstract、Introduction、Related Work、Method、Experiments/Results 或 Conclusion。
仅根据每段文字的内容、功能和写作风格来推断顺序。
严格按以下格式输出一行：
A>B>C>D>E
这里，“>”表示“在论文中出现得早于”。
不要提供任何解释、推理、章节名称或额外文本。只输出最终顺序。 | "D>B>C>E>A" | A1 | judge | [] | [{"text":"In this work, we present DocSeeker, which adopts the novel ALR visual reasoning paradigm, a two-stage training framework and an effective resolution strategy to address the core challenges of low signal-to-noise ratio (SNR) and supervision scarcity in long-document understanding. DocSeeker learns fine-grained... | np_a1_05_001__zh_interleave | np_a1_05_001 | zh_interleave | interleave | text | You are an expert evaluator for ranking tasks. I will provide you with the ground-truth and the model's predicted ranking. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
If the model's predicted ranking is entirely correct, it receiv... | {} | np_a1_05 | scientific | zh | np_a1_05_001 | [] |
430 | 现在我向 Deepseek-R1 （DS-R1） 提出了一个物理问题。我的题库包含许多从 AMC12、AMC10、AIME24 和 OlympiadBench 中提取的数学定理和计算方法。通过使用文档中提到的方法，我发现我的题库也可以帮助 DS-R1 更正确地回答这个物理问题。文档中的哪张表最能支持上述现象？
请按以下格式输出：
{"Location": "<e.g. Table 10>", "Description": "解释你为什么得出这个答案。"} | {"Location": "Table 2", "Description": "The phenomenon in the question uses examples from one domain (math) to help with a different domain (physics), which is about cross-domain generalizability. And the example bank has a low similarity to the target problem (since physics ≠ math). The section 'Generalizability' best... | A2 | judge | [
"74/7436d3fc2c22f38efb5d3626d6275550d1c33062d016d5850227e8b9ae29a674.jpg",
"e3/e3ce6ff5575f3cf9f90eddc491c930e0663ba6de885c9dd3308f1bf3ba9acf24.jpg",
"e7/e7d1a8a4f5059b5170ab14c3401b88bd5a51ebc2c96d88b5f6eac7d3d9b396c9.jpg",
"32/32e165503194e004847f37e1ae97e54a5c994476562c352aa5a735753832152d.jpg",
"d7/d760... | [{"text":"# BoostStep: Boosting Mathematical Capability of Large Language Models via Improved Single-step Reasoning\n\n## Abstract\n\nLarge language models (LLMs) have demonstrated impressive ability in solving complex mathematical problems with multi-step reasoning and can be further enhanced with well-designed in-con... | np_a2_03_001__zh_interleave | np_a2_03_001 | zh_interleave | interleave | json | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. If the model's predicted 'Location' is incorrect, ... | {} | np_a2_03 | scientific | zh | np_a2_03_001 | [
"images/SciDocBench/74/7436d3fc2c22f38efb5d3626d6275550d1c33062d016d5850227e8b9ae29a674.jpg",
"images/SciDocBench/e3/e3ce6ff5575f3cf9f90eddc491c930e0663ba6de885c9dd3308f1bf3ba9acf24.jpg",
"images/SciDocBench/e7/e7d1a8a4f5059b5170ab14c3401b88bd5a51ebc2c96d88b5f6eac7d3d9b396c9.jpg",
"images/SciDocBench/32/32e16... |
431 | 我需要使用本文中提到的一些数据集来训练我自己的模型。请按照以下步骤帮助我选择数据集：
1. 根据 token 数量选择排名前 15 的数据集。
2. 在这 15 个数据集中，仅保留来自 "Embodied AI" 和 "Spatial Understanding" 类别的数据集。
3. 按字母顺序对它们排序，并为每个类别输出一个列表，然后将它们合并为一个字典。
请按以下格式提供最终输出：
{"Embodied AI": ["<e.g. MsCoco>", "<e.g. MegaDepth">, ...], "Spatial Understanding": ["<e.g. ChartQA>", "<e.g. DocVQA>", ...... | {"Embodied AI": ["Eb-Alfred", "Eb-Habitat", "Muep", "OWMM-VLM Data", "Robo2VLM-1", "RoboVQA"], "Spatial Understanding": ["Vg-Llm-Scannet-Det", "Vica-322K", "Vsi 590K"]} | A4 | json_match | [
"ee/eef4797905b92b21c16b3efba01da3bb6abe65e66c2f2715a6b86ed4a1761aee.jpg"
] | [{"image":"ee/eef4797905b92b21c16b3efba01da3bb6abe65e66c2f2715a6b86ed4a1761aee.jpg"},{"text":"我需要使用本文中提到的一些数据集来训练我自己的模型。请按照以下步骤帮助我选择数据集：\n1. 根据 token 数量选择排名前 15 的数据集。\n2. 在这 15 个数据集中，仅保留来自 \"Embodied AI\" 和 \"Spatial Understanding\" 类别的数据集。\n3. 按字母顺序对它们排序，并为每个类别输出一个列表，然后将它们合并为一个字典。\n请按以下格式提供最终输出：\n{\"Embodied AI\": [\"... | np_a4_01_001__zh_interleave | np_a4_01_001 | zh_interleave | interleave | json | {} | np_a4_01 | scientific | zh | np_a4_01_001 | [
"images/SciDocBench/ee/eef4797905b92b21c16b3efba01da3bb6abe65e66c2f2715a6b86ed4a1761aee.jpg"
] | |
432 | 你是一名论文阅读助手。阅读这篇文章，告诉我鞋底区域 II 和 III 中箭头的方向（从上到下还是从下到上？）
你的答案应该是几个单句或一个词。 | "Both from up to bottom" | A5 | judge | [
"ad/adf697644e26dc39120e093606c9ea0df2f6a959850085dcfb2bf7fc13f1b680.jpg",
"8f/8fa74cc7a3f5a95b5427fff149ecade53c8776b78de8595d0dff1b00bed22fa8.jpg",
"cc/cc632d2c9503c431d50c4392a79ba0eb44b7ca59c107b3eb6ebbfa687881e087.jpg",
"0b/0b0af88015d37562df869bb3fa8c100821010b516a8718fbc19ee6099a07f206.jpg",
"83/8386... | [{"text":"# Estimation of the Biological Affinities of Seven Species of Sulawesi Macaques Based on Multivariate Analysis of Dermatoglyphic Pattern Types\n\nABSTRACT. Seven species of monkeys belonging to the genus Macaca live allopatrically on Sulawesi Island, Indonesia. FOODEN (1969) assigned full specific status to e... | np_a5_02_001__zh_interleave | np_a5_02_001 | zh_interleave | interleave | text | {} | np_a5_02 | scientific | zh | np_a5_02_001 | [
"images/SciDocBench/ad/adf697644e26dc39120e093606c9ea0df2f6a959850085dcfb2bf7fc13f1b680.jpg",
"images/SciDocBench/8f/8fa74cc7a3f5a95b5427fff149ecade53c8776b78de8595d0dff1b00bed22fa8.jpg",
"images/SciDocBench/cc/cc632d2c9503c431d50c4392a79ba0eb44b7ca59c107b3eb6ebbfa687881e087.jpg",
"images/SciDocBench/0b/0b0af... | |
433 | 在 Fig. 1 的底部图中，AQC(exp) 和 QAOA 曲线的 $\epsilon$ 值范围最接近以下哪个选项？
A. 5e-3 to 5e-1
B. 5e-4 to 3e-1
C. 1e-3 to 3e-1
D. 5e-3 to 1e0
E. 5e-4 to 1e-1
F. 8e-4 to 1e0
G. 8e-4 to 5e-1
H. 1e-3 to 1e-1
仔细思考，并按以下字典格式输出：
{"AQC(exp) range": "<e.g. A>", "QAOA range": "<e.g. A>", “Description": "<Explain why you derive this answer>"} | {"AQC(exp) range": "G", "QAOA range": "C", "Description": "The bottom figure is a log-log plot with κ=10. The scale on the x-axis gradually becomes denser from left to right, then suddenly thins out after a certain mark, before becoming denser again, repeating this cycle. Based on the characteristics of logarithmic cur... | B2 | judge | [
"bb/bba61b6a3b7af640259d4200adfab833d6252e3ec4d7e5c46535f0800a705756.jpg",
"0b/0b79740d03474457f8c38419705ff2d9a6da531cbae055ce74f6ba34bec57965.jpg",
"ba/baed0de5fe34ea15b15db3a444440e5a608819d606c6505564f6a897f9ca7374.jpg",
"61/61b15bfe773d025963f3353188a98a8cbff8b6cae992d3673334ae7155cc696e.jpg",
"e1/e1ff... | [{"text":"# Quantum Linear System Solver Based on Time-optimal Adiabatic Quantum Computing and Quantum Approximate Optimization Algorithm\n\nIn this article, we find that with a simple modification of the scheduling function to traverse the adiabatic path, the gap between AQC and RM can be fully closed, along with the ... | np_b2_01_001__zh_interleave | np_b2_01_001 | zh_interleave | interleave | json | You are an expert evaluator the model's response to a question. I will provide you with the question, the ground-truth and the model's prediction. You need to score the model's prediction.
Question: {prompt}
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. If the model cor... | {} | np_b2_01 | scientific | zh | np_b2_01_001 | [
"images/SciDocBench/bb/bba61b6a3b7af640259d4200adfab833d6252e3ec4d7e5c46535f0800a705756.jpg",
"images/SciDocBench/0b/0b79740d03474457f8c38419705ff2d9a6da531cbae055ce74f6ba34bec57965.jpg",
"images/SciDocBench/ba/baed0de5fe34ea15b15db3a444440e5a608819d606c6505564f6a897f9ca7374.jpg",
"images/SciDocBench/61/61b15... |
434 | 你将获得一篇科学 PDF 文档。请仔细检查文档中的每个图和子图，并找出所有内容/数据与正文不匹配的地方。
对于你识别出的每个有问题的单元格，输出一个包含以下字段的 JSON 对象：
- Location: 该问题所属的具体表格，例如 "Figure 1", "Figure 2(b)"
- Description: 解释为什么这个（子）图存在错误。
输出一个 JSON 列表，其中每个有问题的单元格对应一个对象。如果未发现问题，则输出空列表。只输出 JSON，不要输出任何其他评论。 | [{"Location": "Figure 8(b)", "Description": "The x-axis displays adsorption dosages of 0.5, 1.0, 1.5, 2.0, and 2.5 g/L, which contradicts the experimental dosages described in the main text (2, 4, 6, 8, and 10 g)."}, {"Location": "Figure 8(c)", "Description": "The x-axis displays agitation speed of of 50, 100, 150, 20... | C1 | judge | [
"6f/6fb79519f3a4253400d51f708f09dde673b1988900c17696147328be12a19e2e.jpg",
"f3/f3c3cba8fa31913980dd95801b21ab96461c9478c7a5b12e2a2f3ff75920a86b.jpg",
"cf/cfb3cba871485cbb8e1cb57b24830f1fcac58d1c8d60d9846e88ff03157cda31.jpg",
"cb/cb3e0d1c5ab459ec1ab0f3736e30f28c0b488daa2d1cabd1ba38f0b06bb97fd2.jpg",
"14/1406... | [{"text":"chemical states of surface elements, it was not performed in this study. The focus was on field-relevant fluoride adsorption under near-neutral groundwater conditions, and the combination of SEM, EDX, zeta potential, and $\\mathsf{pH}_{\\mathsf{p}}\\mathsf{z}\\mathsf{c}$ measurements sufficiently characterize... | np_c1_01_001__zh_interleave | np_c1_01_001 | zh_interleave | interleave | json | You are an expert evaluator. You will be given a question, a reference answer, and a model prediction.
Score the prediction from 0.0 to 1.0 based on how well it matches the reference answer in content and accuracy.
Question:
{prompt}
Reference answer:
{answer}
Model prediction:
{prediction}
Scoring rules (Total=1.0... | {} | np_c1_01 | scientific | zh | np_c1_01_001 | [
"images/SciDocBench/6f/6fb79519f3a4253400d51f708f09dde673b1988900c17696147328be12a19e2e.jpg",
"images/SciDocBench/f3/f3c3cba8fa31913980dd95801b21ab96461c9478c7a5b12e2a2f3ff75920a86b.jpg",
"images/SciDocBench/cf/cfb3cba871485cbb8e1cb57b24830f1fcac58d1c8d60d9846e88ff03157cda31.jpg",
"images/SciDocBench/cb/cb3e0... |
435 | 通读整篇论文。只使用这篇论文。不要解释。准确输出一个 20 个字符的位串。当且仅当该断言被论文完全支持时使用 1；否则使用 0。
A. 该论文研究对称的双社区 SBM G(n,p,q)，其中有两个大小相等的簇，簇内边概率为 p，跨簇边概率为 q。
B. 论文中的精确恢复意味着恢复顶点标签，不允许对所有标签进行全局翻转。
C. 在主要阈值机制中，论文取 p = α log(n)/n 和 q = β log(n)/n，为简单起见关注 α > β。
D. Theorem 1 本身表明，当 (α+β)/2 − sqrt(αβ) < 1 时，ML 以趋于一的概率失败。
E. Theorem 2 表明，当 (α+β)/2 − sqrt(αβ)... | "10101110101111101100" | C1 | judge | [
"73/735ad715cda2752f32ea0f39e917a6e713a6253ce1d3307749a1b0f7093441be.jpg",
"9d/9d0060c6f00c3042537fa5bfec6a32e924aa39bf2e379a538a04a9a069a4a1a3.jpg",
"fd/fde2ae760701cfbc8f5cf9604000c1da3108d5ab01410217d0cd997d4b302c84.jpg",
"3b/3b614559d818c056e583f2802c34e9e1cc269e4f5556a858d86228d9b545ae6c.jpg",
"d9/d93c... | [{"text":"# Exact Recovery in the Stochastic Block Model\n\nThe stochastic block model (SBM) with two communities, or equivalently the planted bisection model, is a popular model of random graph exhibiting a cluster behaviour. In the symmetric case, the graph has two equally sized clusters and vertices connect with pro... | np_c1_07_001__zh_interleave | np_c1_07_001 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. The ground-truth is a bit-string with 20 bits. If t... | {} | np_c1_07 | scientific | zh | np_c1_07_001 | [
"images/SciDocBench/73/735ad715cda2752f32ea0f39e917a6e713a6253ce1d3307749a1b0f7093441be.jpg",
"images/SciDocBench/9d/9d0060c6f00c3042537fa5bfec6a32e924aa39bf2e379a538a04a9a069a4a1a3.jpg",
"images/SciDocBench/fd/fde2ae760701cfbc8f5cf9604000c1da3108d5ab01410217d0cd997d4b302c84.jpg",
"images/SciDocBench/3b/3b614... |
436 | 通读整篇论文。仅使用这篇论文。不要解释。准确输出一个 20 字符的比特串。当且仅当该断言被论文完全支持时使用 1；否则使用 0。
A. 权衡函数 T(P,Q)(α) 被定义为在 I 类错误至多为 α 的拒绝规则中，II 类错误的下确界。
B. Proposition 2.2 说权衡函数是凹的、连续的、非递减的，并且满足 f(x) ≥ 1−x。
C. 如果对所有相邻数据集 S,S′ 都有 T(M(S),M(S′)) ≥ f，则机制是 f-DP。
D. Proposition 2.4 通过 f^S=max{f,f^{-1}} 对 f 进行对称化，并且这是一个更宽松的隐私界，因为 f^S≤f。
E. Proposition 2.5 说明... | "10101101011101011101" | C1 | judge | [
"e7/e79103923f48bdf7928e45c0135e5a8b00a09996f61be62a7aee79eed7c1d32e.jpg",
"ed/ed599ef05c47661dcc5cb2182f2abb76fd3943130c9bf3824e87be5276fe404a.jpg",
"ab/ab30968b5561355be33a3ccd16574c2fa6fb4e6e938be6613acb2a299351a3a9.jpg",
"65/65a78575bf4ee60730e10d1b4e6c317260c2c70ae6743e5d2792b9c97d98ab88.jpg",
"e7/e7ee... | [{"text":"## 2 $f$ -Differential Privacy and Its Basic Properties\n\nIn Section 2.1, we give a formal definition of $f$ -DP. Section 2.2 introduces Gaussian differential privacy, a special case of $f$ -DP. In Section 2.3, we highlight some appealing properties of this new privacy notation from an information-theoretic ... | np_c1_09_001__zh_interleave | np_c1_09_001 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. The ground-truth is a bit-string with 20 bits. If t... | {} | np_c1_09 | scientific | zh | np_c1_09_001 | [
"images/SciDocBench/e7/e79103923f48bdf7928e45c0135e5a8b00a09996f61be62a7aee79eed7c1d32e.jpg",
"images/SciDocBench/ed/ed599ef05c47661dcc5cb2182f2abb76fd3943130c9bf3824e87be5276fe404a.jpg",
"images/SciDocBench/ab/ab30968b5561355be33a3ccd16574c2fa6fb4e6e938be6613acb2a299351a3a9.jpg",
"images/SciDocBench/65/65a78... |
437 | 在我提供的文档中，右侧有一个图表。你的任务是找出图表中的一个错误。
图表描述：该图表使用 Python 的 matplotlib package 生成，所有实验数据随后都在图表上进行了标注。已知所有数据注释都是正确的；然而，在 matplotlib 绘图过程中，有一个数值被错误设置，导致这个特定数据集的图表中出现了绘图错误。
请按照以下步骤操作：
1. 识别图表中的绘图错误。
2. 识别与该错误对应的数据集。
3. 按照以下精确格式将你的答案作为 JSON dictionary 返回：
{"Dataset": "<The name of the corresponding dataset>, "Description": "<An ... | {"Dataset": "DenseFusion", "Description": "The bar height of DenseFusion is approximately at the middle of 1000 and 1500. However, the label is 1453, which should be near 1500. So the bar of DenseFusion is incorrectly plotted.\nFurthermore, the slope of the red line is nearly the same from LLaVa-23K to ShareGPT4V and ... | C2 | judge | [
"cc/cc165091fdaa00f3c665e7b6f8304631c7ab9e460dfa636cee66d4f4faaa4590.jpg"
] | [{"image":"cc/cc165091fdaa00f3c665e7b6f8304631c7ab9e460dfa636cee66d4f4faaa4590.jpg"},{"text":"Figure 4: Data processing and analysis. During the image collecting and processing stage, we primarily focus on the diversity and richness of image content. In the resulting ScaleCap-450k, the captions are significantly longer... | np_c2_01_001__zh_interleave | np_c2_01_001 | zh_interleave | interleave | json | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. If the model's predicted 'Dataset' is incorrect, t... | {} | np_c2_01 | scientific | zh | np_c2_01_001 | [
"images/SciDocBench/cc/cc165091fdaa00f3c665e7b6f8304631c7ab9e460dfa636cee66d4f4faaa4590.jpg"
] |
438 | 我提供了某文档的前五页，以及第 5 页介绍的 Figure 2 中的子图 Figure 2b。然而，在 Figure 2b 中，与两个神经元对应的神经活动图的标签被颠倒了，导致该图中的内容与文档中对 Figure 2b 的分析不一致。请识别神经活动图被颠倒的这两个神经元。
请按以下格式输出：
{"Neruon 1": "<e.g. SMBV>, "Neuron 2": "<e.g. SMBV>", "Description": "说明你为什么得出这个答案。"} | {"Neuron 1": "SMDV", "Neuron 2": "SAAV", "Description": "From Figure 2b we know that the reversal ends at x=2. According to the text on page 5, SAAV activity is described as being higher during reversals in ventral turns (blue line) than in dorsal turns (green line). It suggests that SAAV may predict the future turning... | C2 | judge | [
"ff/ffa125ef5e7d2f530820981a76315d198734942713be8c0d0f9869a550271a1b.jpg",
"6a/6a08b13be0d9f2ed8a6fe1fd2c74928e2e0565fedbf7d2160f0b56feba5300b7.jpg",
"04/040b564fd3a563306760e7a5feb61738e9e8f00d0c79771af651b04aa448ade7.jpg",
"ea/ea75e54a0b918066cbcb974479ee4ce01f2f1021869b008f4f24936740f0d385.jpg",
"ed/edae... | [{"text":"# Neural sequences underlying directed turning in Caenorhabditis elegans\n\nReceived: 4 December 2024\n\nAccepted: 5 March 2026\n\nCheck for updates\n\nTalya S. Kramer $^{1,2}$ , Flossie K. Wan $^{1}$ , Sarah M. Pugliese $^{1}$ , Adam A. Atanas $^{1}$ , Sreeparna Pradhan $^{1}$ , Alex W. Hiser $^{1}$ , Lillie... | np_c2_03_001__zh_interleave | np_c2_03_001 | zh_interleave | interleave | json | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Both fields of 'Neurons' in the prediction MUST be... | {} | np_c2_03 | scientific | zh | np_c2_03_001 | [
"images/SciDocBench/ff/ffa125ef5e7d2f530820981a76315d198734942713be8c0d0f9869a550271a1b.jpg",
"images/SciDocBench/6a/6a08b13be0d9f2ed8a6fe1fd2c74928e2e0565fedbf7d2160f0b56feba5300b7.jpg",
"images/SciDocBench/04/040b564fd3a563306760e7a5feb61738e9e8f00d0c79771af651b04aa448ade7.jpg",
"images/SciDocBench/ea/ea75e... |
439 | 从附加文件中识别 Fig (b) 和 (c) 中红线和黄线的相对高度。
对于每个子图，检查完整的时间过程，并根据 x 轴进展以及任何视觉上可区分的阶段边界，将其划分为连续的时间阶段。对于每个阶段，确定两条线的相对位置：1/ red > yellow；2/ red < yellow。
按时间顺序输出每个子图的结果。将每个阶段输出为一个字典，并将它们放入一个列表中：
[{"Subplot": "Fig. (b)", "Stage_order": "1", "Relative_position": "<red > yellow / red < yellow>"}, {"Subplot": "Fig. (b)", "Stage_order"... | "[\n {\"Subplot\": \"Fig. (b)\", \"Stage_order\": \"1\", \"Relative_position\": \"red < yellow\"},\n {\"Subplot\": \"Fig. (b)\", \"Stage_order\": \"2\", \"Relative_position\": \"red > yellow\"},\n {\"Subplot\": \"Fig. (c)\", \"Stage_order\": \"1\", \"Relative_position\": \"red > yellow\"},\n {\"Subplot\": \"Fig. (c... | C3 | judge | [
"db/dbcbdd0e383e46aef04e307461e0db62ef15ec547e12e0c60393c12a3c246abd.jpg",
"d9/d9e2aedb28c28b23a4cd1e792590511a701205e719bb17d3937385bcc0c62841.jpg",
"46/4640b9a2d04527e06215fc387506825623d0ed1377cff81ed86ee4f55516d108.jpg",
"b1/b15e9734aaec7b44dc5b3bccc70fab25c1c6b6a2a8926797f735b7b22e02f87b.jpg",
"07/0775... | [{"image":"db/dbcbdd0e383e46aef04e307461e0db62ef15ec547e12e0c60393c12a3c246abd.jpg"},{"image":"d9/d9e2aedb28c28b23a4cd1e792590511a701205e719bb17d3937385bcc0c62841.jpg"},{"image":"46/4640b9a2d04527e06215fc387506825623d0ed1377cff81ed86ee4f55516d108.jpg"},{"text":"d"},{"image":"b1/b15e9734aaec7b44dc5b3bccc70fab25c1c6b6a2a... | np_c3_03_001__zh_interleave | np_c3_03_001 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. The ground-truth is a list of dictionaries. The or... | {} | np_c3_03 | scientific | zh | np_c3_03_001 | [
"images/SciDocBench/db/dbcbdd0e383e46aef04e307461e0db62ef15ec547e12e0c60393c12a3c246abd.jpg",
"images/SciDocBench/d9/d9e2aedb28c28b23a4cd1e792590511a701205e719bb17d3937385bcc0c62841.jpg",
"images/SciDocBench/46/4640b9a2d04527e06215fc387506825623d0ed1377cff81ed86ee4f55516d108.jpg",
"images/SciDocBench/b1/b15e9... |
440 | 你是一名理论物理助手。阅读所提供的三篇论文：
- Paper A: Solving the Quantum Many-Body Problem with Artificial Neural Networks
- Paper B: Quantum Entanglement in Neural Network States
- Paper C: Approximating quantum many-body wave functions using artificial neural networks
使用 Paper A 的记号系统（即以 $S$, $\Psi_M(S;\mathcal{W})$, $\mathcal{W}=\... | [{"paper": "B", "eq_label": "Eq.(1)", "rewritten_formula": "\\Psi_M(\\mathcal{S}; \\mathcal{W}) = \\sum_{\\{h_i\\}} \\exp\\left(\\sum_{j=1}^N a_j \\sigma_j^z + \\sum_{i=1}^M b_i h_i + \\sum_{i=1}^M \\sum_{j=1}^N W_{ij} h_i \\sigma_j^z\\right)"}, {"paper": "B", "eq_label": "Rényi entropy", "rewritten_formula": "S_{\\alp... | D1 | judge | [
"30/30b0b8da468187903e8b5f509b9d5bbfac221da8ae5d8acfcb4b8e325102b5af.jpg",
"1e/1efbc8a3a7158a1ea27ee5a2bd47b1ab6f198a45819a44e3d0d8f50defbd8054.jpg",
"78/78cc663ab7a71270d7242b003977a7942a8c55c700652d83ffb4ba96110a9529.jpg"
] | [{"text":"# Approximating quantum many-body wave-functions using artificial neural networks\n\nZi Cai\\*\n\nJinguo Liu\n\nIn this paper, we demonstrate the expressibility of artificial neural networks (ANNs) in quantum many-body physics by showing that a feed-forward neural network with a small number of hidden layers ... | np_d1_01_001__zh_interleave | np_d1_01_001 | zh_interleave | interleave | json | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. The ground-truth is a list of 10 dictionaries. (If... | {} | np_d1_01 | scientific | zh | np_d1_01_001 | [
"images/SciDocBench/30/30b0b8da468187903e8b5f509b9d5bbfac221da8ae5d8acfcb4b8e325102b5af.jpg",
"images/SciDocBench/1e/1efbc8a3a7158a1ea27ee5a2bd47b1ab6f198a45819a44e3d0d8f50defbd8054.jpg",
"images/SciDocBench/78/78cc663ab7a71270d7242b003977a7942a8c55c700652d83ffb4ba96110a9529.jpg"
] |
441 | 以 mermaid 格式复现 Figure (b)。保留所有节点以及表示 'Gap junction' 的锯齿线。忽略 'Chemical synapse' 的箭头。每个节点应严格遵循图中的名称，你不需要复现原图中使用的颜色。找出所有 'Gap junction' 连接，并用 mermaid 绘制它们。只使用普通无向线代替锯齿线。
最终答案按以下格式输出：
```mermaid
graph LR
1[AIB]
2[AVE]
3[AVA]
1 --- 2
2 --- 3
``` | "```mermaid\ngraph LR\n 1[AIB]\n 2[AVE]\n 3[AVA]\n 4[RIM]\n 5[SMB]\n 6[SAA]\n 7[RMD]\n 8[SMD]\n 9[RIV]\n \n 1 --- 2\n 1 --- 4\n 1 --- 9\n 2 --- 3\n 2 --- 4\n 2 --- 7\n 2 --- 9\n 3 --- 4\n 5 --- 6\n 6 --- 7\n 7 --- 8\n 8 --- 9\n```" | E1 | judge | [
"8e/8ea8581e0782ecf6631002d4a68f13d8619a81b02dd297730ab9a9a340ac1d55.jpg",
"5a/5a4f5844007e969dfce3a09bb5b9c6822be922ade8d0a3f41ba2f81c9857d734.jpg",
"44/447fa50b59efda310cc8788b79b0b205366c2749aaab50da082bd098c041d4f8.jpg",
"b6/b6548e2bd98c4808d76f21d7a1eae25b9bc3570f04478721917fe469ec392525.jpg",
"ba/ba04... | [{"image":"8e/8ea8581e0782ecf6631002d4a68f13d8619a81b02dd297730ab9a9a340ac1d55.jpg"},{"image":"5a/5a4f5844007e969dfce3a09bb5b9c6822be922ade8d0a3f41ba2f81c9857d734.jpg"},{"image":"44/447fa50b59efda310cc8788b79b0b205366c2749aaab50da082bd098c041d4f8.jpg"},{"text":"c"},{"image":"b6/b6548e2bd98c4808d76f21d7a1eae25b9bc3570f0... | np_e1_02_001__zh_interleave | np_e1_02_001 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Extract the content between ```mermaid and ``` (to... | {} | np_e1_02 | scientific | zh | np_e1_02_001 | [
"images/SciDocBench/8e/8ea8581e0782ecf6631002d4a68f13d8619a81b02dd297730ab9a9a340ac1d55.jpg",
"images/SciDocBench/5a/5a4f5844007e969dfce3a09bb5b9c6822be922ade8d0a3f41ba2f81c9857d734.jpg",
"images/SciDocBench/44/447fa50b59efda310cc8788b79b0b205366c2749aaab50da082bd098c041d4f8.jpg",
"images/SciDocBench/b6/b6548... |
442 | 请使用 Mermaid 语法为 R-(BN)₂ 绘制一个纵向流程图，严格遵循以下格式：
1. 节点格式：
- 内容：化学式 + 相对能量（单位：eV，数值取自论文，相对于 S₀）
- 样式：统一填充颜色 `#e8f5e9`，边框颜色 `#1b5e20`，黑色文本
2. 边格式：标签严格为 `λ=xxx, τ=xxx`（波长单位为 nm，时间带单位）；如果某个过程缺少某个参数，则使用 `λ=-` 作为占位符
3. 流程：从 2-NaphNH₂ 开始，经过合成步骤到 R-(BN)₂，然后进入光物理循环（必须展示中间步骤）
4. 尽可能使用论文中的原始符号
5. 全文使用 Unicode 数学字符（例如希腊字母 λ、τ，下标... | "flowchart TD\n A[\"2-NaphNH₂<br/>0.00 eV\"] -->|\"λ=-, τ=-\"| B[\"(BN)₁<br/>0.00 eV\"]\n A -->|\"λ=-, τ=-\"| C[\"R-(BN)₂<br/>0.00 eV\"]\n\n C -->|\"λ=280, τ=-\"| D[\"R-(BN)₂ S₁(¹LE)<br/>3.32 eV\"]\n D -->|\"λ=365, τ=3.0ns\"| C\n D -->|\"λ=-, τ=-\"| E[\"R-(BN)₂ ¹HLCT<br/>2.88 eV\"]\n E -->|\"λ=430, τ=... | B2 | judge | [
"b7/b7712cb3b837f9e4093996177ccd0422e14fc95c2fb8c7d874446e2c20e15f81.jpg",
"b8/b8ca93bb0d4dfadd59774e8698af22789acc9f263dd4450a4893e08c45a93603.jpg",
"6e/6eb738cfa0391625a66ca90b4c3a1f00ffca9e72140e9ba9ff60fcde1ad7b0c7.jpg",
"84/84b5005c3b967fdfc2aac44682228266b598b1123b962ab8fb8eabd5b1de289c.jpg",
"cf/cfc4... | [{"text":"https://doi.org/10.1038/s42004-025-01529-8\n\nCheck for updates\n\nJusaina Eyyathiyil $^{1}$ , Subhajit Ghosh $^{1}$ , Arunima Cheran $^{2}$ , Silvano Geremia $^{3}$ , Jatish Kumar $^{2}$ , Neal Hickey $^{3}$ & Pakkirisamy Thilagar $^{1}$\n\nLong-lived triplet exciton harvesting materials are of immense inter... | np_b2_03_001__zh_interleave | np_b2_03_001 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Determine the score based on how well the predicte... | {} | np_b2_03 | scientific | zh | np_b2_03_001 | [
"images/SciDocBench/b7/b7712cb3b837f9e4093996177ccd0422e14fc95c2fb8c7d874446e2c20e15f81.jpg",
"images/SciDocBench/b8/b8ca93bb0d4dfadd59774e8698af22789acc9f263dd4450a4893e08c45a93603.jpg",
"images/SciDocBench/6e/6eb738cfa0391625a66ca90b4c3a1f00ffca9e72140e9ba9ff60fcde1ad7b0c7.jpg",
"images/SciDocBench/84/84b50... |
443 | 找出 Theorem 7.1 中每一项的依赖链，包括 7.1(1a)、7.1(1b)、7.1(1c)、7.1(2) 和 7.1(3)。
使用箭头写出紧凑的链，例如：
Theorem 7.1(1c) <- Proposition 7.1(1a) + Lemma 3.9
Proposition 3.9 <- Lemma 3.9 + Assumption 3.6
Lemma 3.2 <- Definition 3.4 | "Theorem 7.1(1a) ← Proposition 6.3 + Lemma 3.10\nTheorem 7.1(1b) ← Theorem 7.1(1a)\nTheorem 7.1(1c) sharpness ← Proposition 3.5 + Lemma 2.1\nTheorem 7.1(2) ← Proposition 6.4 + Lemma 3.10\nTheorem 7.1(3) ← Proposition 6.5 + Lemma 3.10\nLemma 3.10 ← Lemma 3.9\nLemma 3.9 ← Definitions 3.6, 3.7 + Proposition 3.5\nPropositi... | B3 | judge | [] | [{"text":"We write the vector variable as\n\nand\n\n$$\n\\overline {{{{e _ {A}}}}} = (- 1) ^ {\\ell (\\ell + 1) / 2} e _ {A}.\n$$\n\n$$\n\\langle f, g \\rangle := \\operatorname{Re} (f \\overline {{g}}), \\qquad f, g \\in \\mathbb {R} _ {n}.\\tag{2.4}\n$$\n\nUsing the expansion (2.3) and the orthogonality of the basis ... | np_b3_01_001__zh_interleave | np_b3_01_001 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. The initial score is set as 0.0.
2. Full score is ... | {} | np_b3_01 | scientific | zh | np_b3_01_001 | [] |
444 | 在我提供的文档中，表 1 中 'Spatial Models' 类别有几处错误。你的任务是找出所有这些错误。
请遵循以下步骤：
1. 识别表中的错误。
2. 识别与这些错误相对应的数值。
3. 按以下确切格式将你的答案作为三个 JSON 字典的列表返回：
[{"value": "<The corresponding numerical value>, "description": "<A description of why this part of the table is incorrect.>"}, {"value": "<The corresponding numerical value>, "description": "... | [{"value": "48.4", "description": "The average accuracy for Spatial-MLLM-4B is incorrectly calculated. The sum of the 8 category scores (65.3 + 34.8 + 63.1 + 45.1 + 41.3 + 46.2 + 33.5 + 46.3) equals 375.6, which should result in an average of 46.95 (rounded to 47.0), not 48.4."}, {"value": "47.8", "description": "The a... | C1 | judge | [
"f6/f655dca68cfca8ca2c0c7b299667d834f627a87f18441c715150c28be7cf7309.jpg",
"5b/5b827bbe34f1a9b1e16632ad2aae08ec50b79ca947756c2a045650896e478570.jpg",
"da/da830b179aed8c38b3acf3359307ce8a27af06edf5ad5d6afcc2163acb547883.jpg"
] | [{"text":"# Spa3R: Predictive Spatial Field Modeling for 3D Visual Reasoning\n\nHaoyi Jiang $^{1*}$ Liu Liu $^{2\\dagger}$ Xinjie Wang $^{2}$ Yonghao He $^{3}$ Wei Sui $^{3}$ Zhizhong Su $^{2}$ Wenyu Liu $^{1}$ Xinggang Wang $^{1\\ddagger}$\n\n$^{1}$ Huazhong University of Science & Technology $^{2}$ Horizon Robotics $... | np_c1_03_001__zh_interleave | np_c1_03_001 | zh_interleave | interleave | json | {} | np_c1_03 | scientific | zh | np_c1_03_001 | [
"images/SciDocBench/f6/f655dca68cfca8ca2c0c7b299667d834f627a87f18441c715150c28be7cf7309.jpg",
"images/SciDocBench/5b/5b827bbe34f1a9b1e16632ad2aae08ec50b79ca947756c2a045650896e478570.jpg",
"images/SciDocBench/da/da830b179aed8c38b3acf3359307ce8a27af06edf5ad5d6afcc2163acb547883.jpg"
] | |
445 | 通读整篇论文。只使用这篇论文。不要解释。只输出一个恰好为 20 个字符的比特串。当且仅当该说法得到论文的完全支持时使用 1；否则使用 0。
A. 该论文研究一种简化的 Transformer 模型，该模型仅包含 attention blocks，省略了 MLPs 和 LayerNorm。
B. 引言称，添加 LayerNorm 会导致非常不同的动力学，因为 tokens 被约束为在一个球面上演化。
C. 紧支撑适定性分析通过 conditional optimal transport 框架明确包括了 masked self-attention。
D. 紧支撑定理仅针对 unmasked single-head Softmax se... | "11101010111011110011" | C1 | judge | [
"6d/6d7ee20d424807e13856ea0dc4ce29afb71f5e1dcd30221048c430e51d52b3fb.jpg",
"41/41809fca32b6451defe4583fc3420c13157e15729344a61dc3cb5bf2613c1af5.jpg",
"d7/d7220a8361940423a160b2249d287e2108f513f4d0cad4a3d523673fc6dde85d.jpg",
"46/4671a91ec365b915e53354559a99c7afd93f679eaa3b7ba55a82069356d5a1e7.jpg",
"b2/b2ce... | [{"text":"# A UNIFIED PERSPECTIVE ON THE DYNAMICS OF DEEP TRANSFORMERS\n\nAbstract. Transformers, which are state-of-the-art in most machine learning tasks, represent the data as sequences of vectors called tokens. This representation is then exploited by the attention function, which learns dependencies between tokens... | np_c1_04_001__zh_interleave | np_c1_04_001 | zh_interleave | interleave | text | {} | np_c1_04 | scientific | zh | np_c1_04_001 | [
"images/SciDocBench/6d/6d7ee20d424807e13856ea0dc4ce29afb71f5e1dcd30221048c430e51d52b3fb.jpg",
"images/SciDocBench/41/41809fca32b6451defe4583fc3420c13157e15729344a61dc3cb5bf2613c1af5.jpg",
"images/SciDocBench/d7/d7220a8361940423a160b2249d287e2108f513f4d0cad4a3d523673fc6dde85d.jpg",
"images/SciDocBench/46/4671a... | |
446 | 通读整篇论文。仅使用这篇论文。不要解释。输出恰好一个 20 个字符的比特串。当且仅当该声明得到论文的完全支持时使用 1；否则使用 0。
A. 摘要陈述，常数数量的 self-attention 层可以高效地模拟常数数量的 MPC 轮，并且也可以被其模拟。
B. 摘要还声称，对若干其他神经序列模型以及次二次 transformer 近似不能高效解决的任务，对数深度是足够的。
C. 引言将循环架构描述为比 transformers 更强，因为循环状态可以在所有位置上并行交互。
D. Theorem 3.1 假设一个 randomized R-round (γ,δ)-MPC protocol，并得到一个深度恰好为 R 的 transfo... | "11001110101011010110" | C1 | judge | [
"09/09714c6371a17eb94cf3f4ac7866a3b72ddd38594fe8efea183e268a949a84a9.jpg"
] | [{"text":"# Transformers, parallel computation, and logarithmic depth\n\nClayton Sanford $^{1}$ , Daniel Hsu $^{1}$ , and Matus Telgarsky $^{2}$\n\n$^{1}$ Department of Computer Science, Columbia University, New York, NY, USA $^{2}$ Courant Institute, New York University, New York, NY, USA\n\nWe show that a constant nu... | np_c1_06_001__zh_interleave | np_c1_06_001 | zh_interleave | interleave | text | {} | np_c1_06 | scientific | zh | np_c1_06_001 | [
"images/SciDocBench/09/09714c6371a17eb94cf3f4ac7866a3b72ddd38594fe8efea183e268a949a84a9.jpg"
] | |
447 | 通读整篇论文。仅使用这篇论文。不要解释。准确输出一个 20 字符的比特串。当且仅当该断言得到论文完全支持时使用 1；否则使用 0。
A. 该论文研究可分 Hilbert 空间中的线性回归，使用最小范数插值估计量。
B. 协变量假设允许 x = VΛ^{1/2}z，其中 z 的各坐标是独立的次高斯变量；它们不必是 Gaussian。
C. 在 Assumption 5 下，数据可以有许多解满足 Xθ = y，而最小范数插值解是 X^T(XX^T)^{-1}y。
D. 当 Hilbert 空间具有有限维度 p < n 且 X 的秩为 p 时，论文说存在许多满足 Xθ = y 的插值解。
E. Definition 3 将 r_k(Σ)... | "11100111011010101011" | C1 | judge | [] | [{"text":"The phenomenon of interpolating prediction rules has been an object of study by several authors over the last two years, since it emerged as an intriguing mystery at the Simons Institute program on Foundations of Machine Learning in Spring 2017. Belkin, Ma and Mandal [14] described an experimental study demon... | np_c1_08_001__zh_interleave | np_c1_08_001 | zh_interleave | interleave | text | {} | np_c1_08 | scientific | zh | np_c1_08_001 | [] | |
448 | 从所附文件中识别子图 Fig. 5i 的内容，该子图标题为“疼痛敏化与触觉或触觉感觉的总结性回路图”。子图 Fig. 5i 底部的节点是 'SC'。右下角节点的名称是 'RVM-OPRM1+'。将该子图重新组织为结构化流程图描述。
对于图中显示的每一种关系，确定：
1. 源节点
2. 目标节点
3. 相互作用类型：excitatory、inhibitory，或如果未明确指出则为 unknown
4. 与该连接相关的任何标签、注释、分子、受体或通路
你可以忽略虚线所示的连接。对于每个源节点和目标节点的名称，必须使用上面给出的缩写或图中标注的缩写。
然后将整个子图转换为清晰的流程图式结构化输出。
以字典列表的形式输出，其中每个条目表示一... | "[\n{\"Source\": \"DCN\", \"Target\": \"VPL\", \"Interaction\": \"excitatory\", \"Annotation\": \"Touch or tactile\"},\n{\"Source\": \"SC\", \"Target\": \"VPL\", \"Interaction\": \"excitatory\", \"Annotation\": \"Pain sensitization\"},\n{\"Source\": \"SC\", \"Target\": \"Po\", \"Interaction\": \"excitatory\", \"Annotat... | C3 | judge | [
"00/00f683ced1f6363acb7bb188ff44db976e6fa1dce9c4cd7ccd0f1c73ccf7898f.jpg",
"7b/7b166fbe4a993434552e585579eff4484ef48c9a0769d9083829bb4d7b9e3b0e.jpg",
"5a/5a2385bf7b2963b0df9ac2a43c653f89fdd95e503df4e4c5842816cab4082dda.jpg",
"0a/0aa3761f3b857380bb7bf2f07149dcc6da77ff41647d1129898ccb4aa4fe40f7.jpg",
"16/16ff... | [{"image":"00/00f683ced1f6363acb7bb188ff44db976e6fa1dce9c4cd7ccd0f1c73ccf7898f.jpg"},{"image":"7b/7b166fbe4a993434552e585579eff4484ef48c9a0769d9083829bb4d7b9e3b0e.jpg"},{"image":"5a/5a2385bf7b2963b0df9ac2a43c653f89fdd95e503df4e4c5842816cab4082dda.jpg"},{"image":"0a/0aa3761f3b857380bb7bf2f07149dcc6da77ff41647d1129898ccb... | np_c3_02_001__zh_interleave | np_c3_02_001 | zh_interleave | interleave | text | {} | np_c3_02 | scientific | zh | np_c3_02_001 | [
"images/SciDocBench/00/00f683ced1f6363acb7bb188ff44db976e6fa1dce9c4cd7ccd0f1c73ccf7898f.jpg",
"images/SciDocBench/7b/7b166fbe4a993434552e585579eff4484ef48c9a0769d9083829bb4d7b9e3b0e.jpg",
"images/SciDocBench/5a/5a2385bf7b2963b0df9ac2a43c653f89fdd95e503df4e4c5842816cab4082dda.jpg",
"images/SciDocBench/0a/0aa37... | |
449 | 你是一名物理学论文阅读助手。请阅读所提供的论文，并针对以下每一条在正文或 Methods 中提出的论断，识别所有直接支持该论断的具体 Figures 或 Extended Data Figures，并报告每个图出现的页码。
你的回答必须格式化为一个 JSON 对象，其中每个键是论断索引（字符串形式），每个值是一个恰好包含两个键的对象：
- `supporting_figures`: 一个由字符串组成的 JSON 数组（例如，`["Fig. 1c", "Extended Data Fig. 4"]`）。
- `page_numbers`: 一个由整数构成的 JSON 数组，给出每个对应图出现的页码。请报告 PDF 内部页码。
要... | {"1": {"supporting_figures": ["Extended Data Fig. 4"], "page_numbers": [11]}, "2": {"supporting_figures": ["Extended Data Fig. 3"], "page_numbers": [10]}, "3": {"supporting_figures": ["Fig. 1c", "Fig. 1d"], "page_numbers": [2, 2]}, "4": {"supporting_figures": ["Fig. 3b", "Fig. 3e", "Extended Data Fig. 5"], "page_number... | A2 | judge | [
"c8/c85edf1f29c27ffde9ab6713a2e67c89aafc187ca45aa36ec233bc1a65fcf83c.jpg",
"1e/1ede144e03cf161d5440ce83732b14066d7dac08e4ffad8aadd831a14416985a.jpg",
"7f/7f961f373202991c001854c5c69760b21a7894de2b4edf09641af78276a97513.jpg",
"d1/d137d1a76d512e0d9e27cc8e48d10f8133bd4a921bf8380f9980b34f34dec543.jpg",
"1e/1e21... | [{"text":"# Cascade of electronic transitions in magic-angle twisted bilayer graphene\n\nhttps://doi.org/10.1038/s41586-020-2339-0\n\nMagic-angle twisted bilayer graphene exhibits a variety of electronic states, including correlated insulators $^{1-3}$ , superconductors $^{2-4}$ and topological phases $^{3,5,6}$ . Unde... | np_a2_01_001__zh_interleave | np_a2_01_001 | zh_interleave | interleave | json | {} | np_a2_01 | scientific | zh | np_a2_01_001 | [
"images/SciDocBench/c8/c85edf1f29c27ffde9ab6713a2e67c89aafc187ca45aa36ec233bc1a65fcf83c.jpg",
"images/SciDocBench/1e/1ede144e03cf161d5440ce83732b14066d7dac08e4ffad8aadd831a14416985a.jpg",
"images/SciDocBench/7f/7f961f373202991c001854c5c69760b21a7894de2b4edf09641af78276a97513.jpg",
"images/SciDocBench/d1/d137d... | |
450 | 分析提供的量子计算综述论文，并识别论文的分类体系中明确分类的所有被引用研究工作。从以下维度分配分类标签：
维度 1: QKD_Type
- Discrete_Continuous
- COW
- DPS
- Six_State
- Decoy_State
维度 2: PQC_Type
- Lattice_based
- Code_based
- Multivariate
- Isogeny_based
- Hybrid
维度 3: Hardware_Platform
- Trapped_Ion
- Superconducting
- Silicon
- Photonic
- Topological
输出格式：
仅返回一个 J... | {"Ghalaii et al.(2020)": ["QKD_Type:Discrete_Continuous"], "Valivarthi et al.(2020)": ["QKD_Type:Discrete_Continuous"], "Leverrier and Grangier(2009)": ["QKD_Type:Discrete_Continuous"], "Li et al.(2018)": ["QKD_Type:Discrete_Continuous"], "Lin et al.(2019)": ["QKD_Type:Discrete_Continuous"], "Ruan et al.(2019)": ["QKD_... | A3 | json_match | [
"8c/8c54610ec6d6ce36c2c039fb1de6bf0e13f860c312438d8d4221ad244c05f4ad.jpg",
"a6/a68e15d078dc86d61f44b29b5fd8323e0af689877d56a20f241df183e2e92d0f.jpg",
"01/017467834e593ee8e5e6bf062bb69922dbfdc0976834b9b3a39b33a3a1ccb030.jpg"
] | [{"text":"computation speed will grow faster by using these quantum algorithms. Li et al. [35] discussed merging the elements of quantum mechanics with the intelligent nature-inspired algorithms to mark the new era of computing in the making. Quantum optimization and quantum learning are two classifications based on wh... | np_a3_01_001__zh_interleave | np_a3_01_001 | zh_interleave | interleave | json | {} | np_a3_01 | scientific | zh | np_a3_01_001 | [
"images/SciDocBench/8c/8c54610ec6d6ce36c2c039fb1de6bf0e13f860c312438d8d4221ad244c05f4ad.jpg",
"images/SciDocBench/a6/a68e15d078dc86d61f44b29b5fd8323e0af689877d56a20f241df183e2e92d0f.jpg",
"images/SciDocBench/01/017467834e593ee8e5e6bf062bb69922dbfdc0976834b9b3a39b33a3a1ccb030.jpg"
] | |
451 | 你是一名化学助手。请根据以下要求，写出研究论文中 scheme 2 的有机化合物制备路线/步骤中的完整化学反应方程式：
1. 化学反应方程式必须包括反应产物和反应条件
2. 分子式必须以元素形式完全展开，例如，$\mathrm{C_6H_{12}O_6}$
3. 文本中提到的所有中间过程都必须详细包含
4. 反应方程式不需要遵循顺序/时间顺序；你可以自由安排
格式要求：
1. 所有化学方程式严格使用 LaTeX 格式
2. 使用 `-` 表示无序列表项（项目符号），每行只包含一个方程式
3. 每行应只包含一个方程式
示例回答：
- $\mathrm{H_2O(l)} \longrightarrow \mathrm{H_2(... | "- $\\mathrm{C_3H_7NO_2\\,(7)} + \\mathrm{C_{10}H_{18}O_5\\,(Boc_2O)} \\xrightarrow[\\text{THF/H}_2\\text{O (1:2), 4°C to rt, 16\\,h}]{\\mathrm{Na_2CO_3}} \\mathrm{C_8H_{15}NO_4\\,(8)} + \\mathrm{C_4H_{10}O\\,(t\\text{-}BuOH)} + \\mathrm{CO_2} \\quad 75\\%$\n\n- $\\mathrm{C_8H_{15}NO_4\\,(8)} + \\mathrm{C_4H_{10}ClNO_3... | A5 | judge | [
"99/99de5e2b70c5e9cbe1bc6604e0f0e34f22bb813716d54d3f25b3486e58b174a8.jpg",
"0a/0aadf8370c2b8e2918195a273f59a37da9cc25b34acf658c28647774710546d1.jpg",
"8d/8df21e607dedd943e7c29a8298fc6b5cde6843c3ef91d2eee3cb6ba39fdac2d9.jpg",
"c5/c54ab4755aa69b2ec55063a747f35cb64f8d8faa2b61b8376d40b92104c1327e.jpg",
"98/98f3... | [{"text":"# Total Synthesis of Orbiculamide A"},{"image":"99/99de5e2b70c5e9cbe1bc6604e0f0e34f22bb813716d54d3f25b3486e58b174a8.jpg"},{"image":"0a/0aadf8370c2b8e2918195a273f59a37da9cc25b34acf658c28647774710546d1.jpg"},{"image":"8d/8df21e607dedd943e7c29a8298fc6b5cde6843c3ef91d2eee3cb6ba39fdac2d9.jpg"},{"text":"ABSTRACT: O... | np_a5_01_001__zh_interleave | np_a5_01_001 | zh_interleave | interleave | text | {} | np_a5_01 | scientific | zh | np_a5_01_001 | [
"images/SciDocBench/99/99de5e2b70c5e9cbe1bc6604e0f0e34f22bb813716d54d3f25b3486e58b174a8.jpg",
"images/SciDocBench/0a/0aadf8370c2b8e2918195a273f59a37da9cc25b34acf658c28647774710546d1.jpg",
"images/SciDocBench/8d/8df21e607dedd943e7c29a8298fc6b5cde6843c3ef91d2eee3cb6ba39fdac2d9.jpg",
"images/SciDocBench/c5/c54ab... | |
452 | 根据文档，回答以下问题。
经过 20 个 epochs 后，在第 5 层的表示空间中，以下四个类别：airplane, automobile, ship, and truck 中，哪个类别的边界仍然相对模糊？
按以下格式回答：
{"Category": "<e.g. Airplane, automobile>"} | {"Category": "airplane"} | A5 | json_match | [
"38/38892c57ecf59bcc8a286189ce4764a4fadc4e734e8d68eb977108a5fc85df80.jpg",
"31/3193c32b8edd3461db567ecd25997829b994b3a47597000fb5b8876038afc0d2.jpg"
] | [{"image":"38/38892c57ecf59bcc8a286189ce4764a4fadc4e734e8d68eb977108a5fc85df80.jpg"},{"text":"Figure 1: (a) Overview of the bidirectional spike-based distillation framework. (b) Illustration of feature alignment using the properties of pyramidal neurons, which receive feedforward and feedback signals through their basa... | np_a5_05_001__zh_interleave | np_a5_05_001 | zh_interleave | interleave | json | {} | np_a5_05 | scientific | zh | np_a5_05_001 | [
"images/SciDocBench/38/38892c57ecf59bcc8a286189ce4764a4fadc4e734e8d68eb977108a5fc85df80.jpg",
"images/SciDocBench/31/3193c32b8edd3461db567ecd25997829b994b3a47597000fb5b8876038afc0d2.jpg"
] | |
453 | 关于这篇文章，以下陈述是正确的：
A. 从 Fig. 3 可以观察到，为皮带设计了一个可调滑轮以调节其张力。
B. Fig. 4 中的物理模型假设电缆力 F(cable) 与杆位于同一平面内，且该力平行于地面。
C. Fig. 6 表明电缆锚定模块的旋转可以被主动控制。
D. 在 Fig. 7 中，紫色表示控制信号线的连接顺序，红色表示电源连接，绿色表示传感器信号线连接。
E. Fig. 10 显示，在 100 N 的负载下，Nylon cable 的误差可以保持在约 5% 以内。
F. Fig. 12 显示机器人运动过程中没有负载发生。
请按以下格式作答：
{"Choice":"<e.g. A,B,C,D,E>"(use comm... | {"Choice": "B,E,F"} | A5 | json_match | [
"63/63e10ebd3a6cdc139926c813f8eca00f61bc5f08b7c7feca385d3499a09ce0a8.jpg",
"db/db9c83ebb62add18cf7a1c625f78fdf0c1da477d51877bdf12efbb2fd88774a6.jpg",
"15/15660cdcadeeb76e9a30a3316196493a87c22e00e397f75a9c455dc4fe7e5796.jpg",
"fb/fbf0c3767d90468cc15195c96162995c1adb3461dee916d00d6892c3234496fe.jpg",
"ab/abbc... | [{"text":"# Design of a Variable Stiffness Quasi-Direct Drive Cable-Actuated Tensegrity Robot\n\nAbstract—Tensegrity robots excel in tasks requiring extreme levels of deformability and robustness. However, there are challenges in state estimation and payload versatility due to their high number of degrees of freedom an... | np_a5_07_001__zh_interleave | np_a5_07_001 | zh_interleave | interleave | json | {} | np_a5_07 | scientific | zh | np_a5_07_001 | [
"images/SciDocBench/63/63e10ebd3a6cdc139926c813f8eca00f61bc5f08b7c7feca385d3499a09ce0a8.jpg",
"images/SciDocBench/db/db9c83ebb62add18cf7a1c625f78fdf0c1da477d51877bdf12efbb2fd88774a6.jpg",
"images/SciDocBench/15/15660cdcadeeb76e9a30a3316196493a87c22e00e397f75a9c455dc4fe7e5796.jpg",
"images/SciDocBench/fb/fbf0c... | |
454 | 你是一名专业科学编辑和 LaTeX 排版师。
你的目标是为所提供的研究论文生成一个“符号索引”（Index of Notations）。仅涵盖正文中引入的主要数学符号——不要包括 Appendix A 或任何其他附录中的符号。该表是示例性的而非穷尽性的：包括核心的架构和训练过程符号（数据、层索引、层输出、激活、核/特征图维度、批大小、掩码、集成大小、预测，以及 Packed-Ensembles 参数 $\alpha$, $\gamma$）。可以省略仅在单个证明或旁述中使用的临时符号。
输出要求：
请生成一个完整的 LaTeX 文件，其中包含一个 LaTeX 表格代码块。你可以使用 booktabs, amsmath, amsfont... | "\\documentclass{article}\n\\usepackage{booktabs}\n\\usepackage{amsmath, amssymb, amsfonts}\n\\usepackage{mathtools}\n\\begin{document}\n\n\\begin{table*}[!ht]\n\\renewcommand{\\figurename}{Table}\n\n\\caption{Index of Notations}\n\\label{table:notations}\n\\centering\n\\resizebox{\\textwidth}{!}\n{\n\\begin{tabular}{@... | B1 | judge | [
"2b/2bf6c1f983f3b4414f83ab2ec5b389200d3c787b901aa48edf69330eb3db330c.jpg",
"a2/a24e5afb950396b83dc7a9745d16fd1d03d85e64d398daad10f20adaa07cd8a1.jpg",
"bc/bc22ea3692538ef604b1ab9b8b6a4d381e44e66bf48e36948dbf8698533bf669.jpg",
"0f/0fccafda34b9315a1c3d4bcdf5e1b514d48460620074e75831a719669d82e184.jpg",
"bf/bfa0... | [{"text":"# PACKED-ENSEMBLES FOR EFFICIENT UNCERTAINTY ESTIMATION\n\nOlivier Laurent, $^{1,2,*}$ Adrien Lafage, $^{2,*}$ Enzo Tartaglione, $^{3}$ Geoffrey Daniel, $^{1}$ Jean-Marc Martinez, $^{1}$ Andrei Bursuc $^{4}$ & Gianni Franchi $^{2,\\dagger}$\n\nUniversité Paris-Saclay, CEA, SGLS, $^{1}$ U2IS, ENSTA, Institut P... | np_b1_01_001__zh_interleave | np_b1_01_001 | zh_interleave | interleave | text | {} | np_b1_01 | scientific | zh | np_b1_01_001 | [
"images/SciDocBench/2b/2bf6c1f983f3b4414f83ab2ec5b389200d3c787b901aa48edf69330eb3db330c.jpg",
"images/SciDocBench/a2/a24e5afb950396b83dc7a9745d16fd1d03d85e64d398daad10f20adaa07cd8a1.jpg",
"images/SciDocBench/bc/bc22ea3692538ef604b1ab9b8b6a4d381e44e66bf48e36948dbf8698533bf669.jpg",
"images/SciDocBench/0f/0fcca... | |
455 | 根据 Figure 2(b)，回答以下问题：Figure 2(b) 中显示了多少个键中心？请按以下格式回答：
{"Number of bond centers": "<e.g. 3>", "Description": "<Explain why you derive this answer>"} | {"Number of bond centers": "6", "Description": "From the caption of this figure, we know that the orange dots are bond centers. In figure 2(b), there are 4 orange dots on the green lines in total, 1 in bottom-left and 1 in top-right. Therefore, the number of bond centers is 6."} | B2 | judge | [
"62/6285f9f09bd379bcc6540b0891788bddbb0f2483530a5628a8ca06e0b8d3a6c3.jpg",
"e1/e1339af33e7d53db1e984ac60971b30aff6070a398089ceaeb3a3a3caf3e2981.jpg",
"44/44c6a07bbeb0f65a97cecdde7adf34dcffe23d060c69be84b165f34711e0b28b.jpg",
"2f/2f85f7ea0d34ff52920d6d18bffc4024726f51b6032c8206afc8eb182de7106f.jpg",
"f7/f734... | [{"text":"pubs.acs.org/JACS\n\nCommunication\n\n# A Delocalized Mixed-Valence Dinuclear Ytterbium Complex That Displays Intervalence Charge Transfer\n\nTom J. N. Obey, Mukesh K. Singh, Angelos B. Canaj, Gary S. Nichol, Euan K. Brechin, and Jason B. Love\\*"},{"image":"62/6285f9f09bd379bcc6540b0891788bddbb0f2483530a5628... | np_b2_02_001__zh_interleave | np_b2_02_001 | zh_interleave | interleave | json | {} | np_b2_02 | scientific | zh | np_b2_02_001 | [
"images/SciDocBench/62/6285f9f09bd379bcc6540b0891788bddbb0f2483530a5628a8ca06e0b8d3a6c3.jpg",
"images/SciDocBench/e1/e1339af33e7d53db1e984ac60971b30aff6070a398089ceaeb3a3a3caf3e2981.jpg",
"images/SciDocBench/44/44c6a07bbeb0f65a97cecdde7adf34dcffe23d060c69be84b165f34711e0b28b.jpg",
"images/SciDocBench/2f/2f85f... | |
456 | 你是一名物理数据验证助手。请阅读所提供的论文，并对正文、图注以及任何嵌入的公式进行系统的数值和逻辑一致性检查。
你的任务是识别论文中所有数值、物理参数、实验条件或基于公式的预测在内部看起来不一致或相互矛盾的实例。不要将自己限制在预定义的主题列表内；请广泛扫描整个文档。
对于你发现的每一处不一致，请报告相互冲突的值及其来源位置，须与论文中出现的形式完全一致。
你的回答必须是具有以下结构的 JSON 数组的数组：
[
[
{
"value": "$<LaTeX formula or number>$",
"source": "<verbatim text from the paper or Fi... | [[{"value": "$n_s \\approx 3.2 \\times 10^{12}~\\mathrm{cm}^{-2}$", "source": "Figure 2a"}, {"value": "$-n_s/2 = -1.38 \\times 10^{12}~\\mathrm{cm}^{-2}$", "source": "Four-probe resistance $R_{xx}$ measured at densities corresponding to the region bounded by pink dashed lines in (a), versus temperature. Two superconduc... | C1 | judge | [
"a7/a70c06144da6de5ee879f95b2118c0424a09d117d75319575891a56351b7ec27.jpg",
"9e/9ea2e4883ec4e0a80415ef0110f5f8bb1f47e4e56282d4434801b757d61b1e79.jpg",
"c7/c7943bd7a3da02515193eeaaf386615ea6ff39fbe5046316ed93507c5b57ed06.jpg",
"1b/1b030b40474bb2790f3093ea73a4e565a092f64cedb289ba9c8e4c81049f0b63.jpg",
"bb/bb6d... | [{"text":"# Magic-angle graphene superlattices: a new platform for unconventional superconductivity\n\nTakashi Taniguchi, $^{3}$ Efthimos Kaxiras, $^{2,4}$ and Pablo Jarillo-Herrero $^{1,\\dagger}$"},{"image":"a7/a70c06144da6de5ee879f95b2118c0424a09d117d75319575891a56351b7ec27.jpg"},{"image":"9e/9ea2e4883ec4e0a80415ef0... | np_c1_02_001__zh_interleave | np_c1_02_001 | zh_interleave | interleave | json | {} | np_c1_02 | scientific | zh | np_c1_02_001 | [
"images/SciDocBench/a7/a70c06144da6de5ee879f95b2118c0424a09d117d75319575891a56351b7ec27.jpg",
"images/SciDocBench/9e/9ea2e4883ec4e0a80415ef0110f5f8bb1f47e4e56282d4434801b757d61b1e79.jpg",
"images/SciDocBench/c7/c7943bd7a3da02515193eeaaf386615ea6ff39fbe5046316ed93507c5b57ed06.jpg",
"images/SciDocBench/1b/1b030... | |
457 | 通读整篇论文。仅使用这篇论文。不要解释。输出恰好一个 18 个字符的位串。当且仅当该断言得到论文完全支持时使用 1；否则使用 0。
A. 论文明确询问 Transformers 是否可以通过梯度下降学习 TC^0 中的简单逻辑函数。
B. 引言称，先前的复杂性理论结果已经解决了梯度下降下的训练问题。
C. 论文研究了一个简化的 Transformer，以及一个由二进制字符串及其多数标签配对组成的有限训练集。
D. 论文的主要贡献是一个严格的理论框架，为任何可微参数模型（包括 Transformers）在多项式样本或指数样本设置下给出泛化误差下界。
E. Theorem 4.1 是多项式样本困难性定理，而 Theorem 4.2 是... | "101111101011111010" | C1 | judge | [] | [{"text":"# Provable Failure of Language Models in Learning Majority Boolean Logic via Gradient Descent\n\nBo Chen $^{*}$ Zhenmei Shi $^{\\dagger}$ Zhao Song $^{\\ddagger}$ Jiahao Zhang $^{\\S}$\n\n## Abstract\n\nRecent advancements in Transformer-based architectures have led to impressive breakthroughs in natural lang... | np_c1_05_001__zh_interleave | np_c1_05_001 | zh_interleave | interleave | text | {} | np_c1_05 | scientific | zh | np_c1_05_001 | [] | |
458 | 根据 Fig.3，回答以下问题。
对于 \chi_M'-频率曲线，描述斜率绝对值和拐点频率的变化趋势。 | As temperature increases from 30 K to 51 K, the absolute value of the slope decreases (the curve becomes more gradual), while the inflection point shifts to higher frequencies. | C3 | judge | [
"ae/ae0d0e601dcb7a1e15f4f6cef12fa8b90d69dafe527b899291ec4d22d1b5fb28.jpg",
"cb/cb66ad405b99ff7b547341f5eac983731fdcc6252440a1052e66ee2af992182f.jpg",
"8c/8c559f756f2332985f42fda7a242d6c1121e02aee56bfb8e224de9122adb5108.jpg",
"2e/2e1d8543cb758eeddf7b1bee280e05d9a25d177a468b80de0c7686336a4a1146.jpg",
"35/3557... | [{"text":"ARTICLE\n\nIncreasing the operating temperatures of single-molecule magnets—molecules that can retain magnetic polarization in the absence of an applied field—has potential implications toward information storage and computing, and may also inform the development of new bulk magnets. Progress toward these goa... | np_c3_04_001__zh_interleave | np_c3_04_001 | zh_interleave | interleave | text | {} | np_c3_04 | scientific | zh | np_c3_04_001 | [
"images/SciDocBench/ae/ae0d0e601dcb7a1e15f4f6cef12fa8b90d69dafe527b899291ec4d22d1b5fb28.jpg",
"images/SciDocBench/cb/cb66ad405b99ff7b547341f5eac983731fdcc6252440a1052e66ee2af992182f.jpg",
"images/SciDocBench/8c/8c559f756f2332985f42fda7a242d6c1121e02aee56bfb8e224de9122adb5108.jpg",
"images/SciDocBench/2e/2e1d8... | |
459 | 根据论文中的图 1(b)，列出作为单独子图显示的三类社交互动。
请按以下 JSON 格式回答（三个字符串组成的数组，按照它们在图中出现的顺序）：
{"Categories": ["<category 1>", "<category 2>", "<category 3>"]} | {"Categories": ["Advice requests", "Work-related orders", "Social invitations"]} | A2 | json_match | [
"4f/4fd1e8fb506a5bf28eac26cc7c03d88d35ac44b20a9ca5f262e473ff7dc05f15.jpg",
"3b/3bf407fe1cef2e205963fd12164dc0300226f398278aa1c8c00fb8e6307098f6.jpg",
"f8/f8482bebbe3e287eed4d35630024561794779b2f9661706b97a7e34504626a1b.jpg",
"a8/a8f13cfafd6ece8bc699dae60a624452add973f1b16a82bec693aca27c545431.jpg",
"57/574c... | [{"text":"# Inferring the internal structure of groups through the integration of statistical learning and causal reasoning\n\nIsaac Davis $^{1}$ , Julian Jara-Ettinger $^{1,2}$ & Yarrow Dunham $^{1,2}$\n\nHuman social life unfolds within richly structured networks of overlapping relationships, including friendships, h... | np_a2_02_001__zh_interleave | np_a2_02_001 | zh_interleave | interleave | json | {} | np_a2_02 | scientific | zh | np_a2_02_001 | [
"images/SciDocBench/4f/4fd1e8fb506a5bf28eac26cc7c03d88d35ac44b20a9ca5f262e473ff7dc05f15.jpg",
"images/SciDocBench/3b/3bf407fe1cef2e205963fd12164dc0300226f398278aa1c8c00fb8e6307098f6.jpg",
"images/SciDocBench/f8/f8482bebbe3e287eed4d35630024561794779b2f9661706b97a7e34504626a1b.jpg",
"images/SciDocBench/a8/a8f13... | |
460 | 你是一个用于生物学文章分析的 AI。基于所提供的页面，仅列出这些工作中明确被命名为使用的顶层数据库。使用规范的短名称（例如 "UniProt", "ClinVar", "ProteinGym", "gnomAD"）。不要列出子数据集，不要杜撰数据库，也不要包含仅仅顺带引用的资源。
输出一个按字母顺序排序的字符串 JSON 列表：
["<DB_1>", "<DB_2>", ...]
将该列表放在键 "answer_list" 下，因此最终输出为：
{"answer_list": [...]} | {"answer_list": ["ClinVar", "ProteinGym", "UniProt", "gnomAD"]} | A4 | json_match | [
"54/540d7a2d34d0ca9aba3c756abf62b38c0ae529264b83abbb9d18c34160e6f6ae.jpg",
"a2/a2707320c5c1c2766e22a32ec699c851339a5afea70e4be16229fb218bff1067.jpg",
"b2/b22ae79cdb1b9977368330a703fc19b675950c9d4b44401939dbd26cd8decc98.jpg",
"f0/f035e01cec2586cae4383bcd45449b4357251bdad72fdeb5465d7c863568820f.jpg",
"b5/b559... | [{"text":"Article\n\nCheck for updates\n\nTuan Dinh $^{1}$ , Seon-Kyeong Jang $^{2}$ , Noah Zaitlen $^{2,3,4}$ & Vasilis Ntranos $^{1,5,6,7}$\n\nPredicting the functional consequences of genetic variants, known as variant effect prediction (VEP), is a fundamental computational challenge with important applications in h... | np_a4_02_001__zh_interleave | np_a4_02_001 | zh_interleave | interleave | json | {} | np_a4_02 | scientific | zh | np_a4_02_001 | [
"images/SciDocBench/54/540d7a2d34d0ca9aba3c756abf62b38c0ae529264b83abbb9d18c34160e6f6ae.jpg",
"images/SciDocBench/a2/a2707320c5c1c2766e22a32ec699c851339a5afea70e4be16229fb218bff1067.jpg",
"images/SciDocBench/b2/b22ae79cdb1b9977368330a703fc19b675950c9d4b44401939dbd26cd8decc98.jpg",
"images/SciDocBench/f0/f035e... | |
461 | 查看 Figure 2 中将定量测量值相对于变量 "Days on CLZ" 作图的面板（例如，按天数分组的箱线图 / 条形图）。对于 x 轴上出现的每个 "Days on CLZ" 值，计算该组显著性注释上方显示的星号（asterisks/stars）数量（如果有）。
对于每个 x 轴组：
- 确定 CLZ 上的天数。
- 计算该组上方的星号符号数量（如果没有则为 0）。
按 "Days on CLZ" 递增顺序，每组输出一行 JSON：
{"Days on CLZ": "<e.g. 1>", "Star_count": "<e.g. 0, 1, 2, 3>"}
{"Days on CLZ": "<e.g. 2>", "Sta... | "{\"Days on CLZ\": \"1\", \"Star_count\": \"0\"}\n{\"Days on CLZ\": \"2\", \"Star_count\": \"3\"}\n{\"Days on CLZ\": \"3\", \"Star_count\": \"3\"}\n{\"Days on CLZ\": \"5\", \"Star_count\": \"3\"}\n{\"Days on CLZ\": \"7\", \"Star_count\": \"3\"}" | A5 | judge | [
"3f/3fa915b4d7b82e1a2bf4bb1f02d3f28bbf96200723574e4d1a06cc2ae7e6cb54.jpg",
"a5/a5dc57e9b8b683b1c5bc690e1aabd16d79b52f66421c27f4c6836aedb449f7ff.jpg",
"7b/7b515e866ac14fbfde66f4a71ab4bb6c050fe95842c685cc9edfa3fc92914dcf.jpg",
"76/76f82e1401d5e891463abcfcc1333a05b0400ada8d0f65ef771af0cf7f2afabb.jpg",
"c0/c020... | [{"text":"b"},{"image":"3f/3fa915b4d7b82e1a2bf4bb1f02d3f28bbf96200723574e4d1a06cc2ae7e6cb54.jpg"},{"image":"a5/a5dc57e9b8b683b1c5bc690e1aabd16d79b52f66421c27f4c6836aedb449f7ff.jpg"},{"text":"f"},{"image":"7b/7b515e866ac14fbfde66f4a71ab4bb6c050fe95842c685cc9edfa3fc92914dcf.jpg"},{"image":"76/76f82e1401d5e891463abcfcc133... | np_a5_03_001__zh_interleave | np_a5_03_001 | zh_interleave | interleave | text | {} | np_a5_03 | scientific | zh | np_a5_03_001 | [
"images/SciDocBench/3f/3fa915b4d7b82e1a2bf4bb1f02d3f28bbf96200723574e4d1a06cc2ae7e6cb54.jpg",
"images/SciDocBench/a5/a5dc57e9b8b683b1c5bc690e1aabd16d79b52f66421c27f4c6836aedb449f7ff.jpg",
"images/SciDocBench/7b/7b515e866ac14fbfde66f4a71ab4bb6c050fe95842c685cc9edfa3fc92914dcf.jpg",
"images/SciDocBench/76/76f82... | |
462 | 根据图 1，在所示能量范围内，是否存在任何能量区间，其中 prompt signals 的计数少于 delay signals 的计数？回答 "Yes" 或 "No"。
Output format:
{"Has_Interval_Where_Prompt_Less_Than_Delay": "<Yes/No>"} | {"Has_Interval_Where_Prompt_Less_Than_Delay": "No"} | A5 | json_match | [
"de/ded8d193d530cdaa4c98122821271e1a22510a3bbfe94d878ce979a14ab4119e.jpg",
"59/599c81bf1d32771f0e42695e9ce7fdf9bae12bd84a1b736b66ff996e743c960f.jpg",
"34/349b5940d69da9622dad0a58d7514a929cf4faa0af0caf35bda0b09baaa3ee94.jpg",
"dc/dc8c1fcff01dc21d0cbd7de395de60908fd34ebb8b11461ebf02779e12047355.jpg",
"82/8276... | [{"text":"# New limits on the Pauli forbidden transitions in ${}^{12}$ C nuclei obtained with the complete Borexino dataset\n\nD. Basilico $^{1}$ , G. Bellini $^{1}$ , J. Benziger $^{2}$ , R. Biondi $^{3,10}$ , B. Caccianiga $^{1}$ , A. Caminata $^{5}$ , A. Chepurnov $^{6}$ , D. D'Angelo $^{1}$ , A. Derbin $^{7,8}$ , A... | np_a5_04_001__zh_interleave | np_a5_04_001 | zh_interleave | interleave | json | {} | np_a5_04 | scientific | zh | np_a5_04_001 | [
"images/SciDocBench/de/ded8d193d530cdaa4c98122821271e1a22510a3bbfe94d878ce979a14ab4119e.jpg",
"images/SciDocBench/59/599c81bf1d32771f0e42695e9ce7fdf9bae12bd84a1b736b66ff996e743c960f.jpg",
"images/SciDocBench/34/349b5940d69da9622dad0a58d7514a929cf4faa0af0caf35bda0b09baaa3ee94.jpg",
"images/SciDocBench/dc/dc8c1... | |
463 | 关于这篇文章，以下陈述是正确的：
A. 图 1 仅用于说明张拉整体艺术与折纸艺术之间的关系。
B. 图 2 中的红色标签表示节点的索引。
C. 在图 3 中，对于 Miura 折纸单胞，Order 2 和 Order 3 的特征值相差超过一百万倍，这是由于变形模式发生了变化。
D. 图 4 表明，除少数特殊情况外，在 Miura 折纸单胞中，杆件相对于铰链的刚度贡献比约为 1:2。
E. 图 5 表明，在 Miura 折纸单胞的大变形分析过程中，x、y 和 z 方向的刚度呈现相同的变化趋势。
F. 图 10 表明，对于缆索驱动的 Kresling 结构，X 和 Z 方向的刚度同时达到最大值。在该构型下，对于相同大小的长度变化（无论是... | {"Choice": "C"} | A5 | json_match | [
"4d/4d3a1beb2687486ab8f9607cee489d255da3c556a476f0663dd2eee6f829cb31.jpg",
"c9/c93fbd6ba57faeb743a94771e03a76a9c6100b195ffcec6149c163e345ea4e8c.jpg",
"49/490505b01d8c5846b75a1856073907883300c24779df784c56b687da655fb535.jpg",
"08/08e1454c3acd68241f7f7074fbb9bae04f6fe6c8b359231c831ec62abc952930.jpg",
"8a/8ada... | [{"text":"# Statics of integrated origami and tensegrity systems\n\nThis study presents an explicit form of the static equilibrium equations of integrated origami and tensegrity systems. The analytical approach allows one to model and analyze the isolated origami and tensegrity paradigms as a whole system. The tensegri... | np_a5_06_001__zh_interleave | np_a5_06_001 | zh_interleave | interleave | json | {} | np_a5_06 | scientific | zh | np_a5_06_001 | [
"images/SciDocBench/4d/4d3a1beb2687486ab8f9607cee489d255da3c556a476f0663dd2eee6f829cb31.jpg",
"images/SciDocBench/c9/c93fbd6ba57faeb743a94771e03a76a9c6100b195ffcec6149c163e345ea4e8c.jpg",
"images/SciDocBench/49/490505b01d8c5846b75a1856073907883300c24779df784c56b687da655fb535.jpg",
"images/SciDocBench/08/08e14... | |
464 | 我是一名张拉整体机器人的初学者，最近读到一段讨论具有双重运行模式的张拉整体机构的文字。我特别感兴趣的是文章中提到的所谓 “FCC” 算法。请你说明论文中是如何介绍 “FCC” 及其结果的？
A. 从头到尾进行了全面介绍，包括其基本原理。
B. 结合参考文献作了简要解释，并且只有少量描述性陈述。
C. 文章提供了明确的公式，可在无需额外推导的情况下直接评估姿态有效性。
D. 没有提供详细的方程或不等式；相反，该方法是以定性方式描述的。
E. 所有结果都呈现在三维坐标系中。
F. 根据图 4，在模拟 FCC 结果时，蓝色所示的结构固定在地面上。
请按以下格式回答：
{"Choice":"<e.g. A,B,C,D,E>"(使用逗号分隔选... | {"Choice": "B"} | B2 | json_match | [
"18/18c1ace23a545b30148fe45f0b809e1391b912b7c6571295bc0c82309a256213.jpg",
"d9/d9e218655dc6bd90ee5376884a92d631beb6ba328395c7dc50562643a9cf9273.jpg",
"bc/bc92c4be9b153d7a6130d75053919c4f1df3adcd13b048ffece1c57c69b39670.jpg",
"d0/d083c298aa576e94eeb2fa2cb5cedd58932aafcfd2c0a0cb74b40e4d9bbc6046.jpg",
"8e/8e83... | [{"text":"Article\n\n# A Symmetric Three Degree of Freedom Tensegrity Mechanism with Dual Operation Modes for Robot Actuation\n\nTianyuan Wang $^{ID}$ and Mark A. Post $^{*ID}$"},{"image":"18/18c1ace23a545b30148fe45f0b809e1391b912b7c6571295bc0c82309a256213.jpg"},{"text":"Citation: Wang, T.; Post, M.A. A Symmetric Three... | np_b2_04_001__zh_interleave | np_b2_04_001 | zh_interleave | interleave | json | {} | np_b2_04 | scientific | zh | np_b2_04_001 | [
"images/SciDocBench/18/18c1ace23a545b30148fe45f0b809e1391b912b7c6571295bc0c82309a256213.jpg",
"images/SciDocBench/d9/d9e218655dc6bd90ee5376884a92d631beb6ba328395c7dc50562643a9cf9273.jpg",
"images/SciDocBench/bc/bc92c4be9b153d7a6130d75053919c4f1df3adcd13b048ffece1c57c69b39670.jpg",
"images/SciDocBench/d0/d083c... | |
465 | 关于这篇文章，以下陈述是正确的：
A. 图 3 显示，该机器的顺时针旋转速度超过逆时针旋转速度的 1.5 倍以上。
B. 从图 4 (D) 可以清楚地观察到，该机器从水平杆下方通过。
C. 文章没有对抗冲击机制进行详细的理论分析；相反，它仅通过实验展示了机器的抗冲击能力。
D. 在运动实验和抗冲击测试中使用的端盖在材料和结构尺寸上均不同。这表明该研究可能无法同时实现“surviving harsh impacts from drops (at least 5.7 m)”和“achieving high locomotion speeds (18 bar lengths per minute)”。
E. 论文对该机制的运动进行了详细分... | {"Choice": "A,B,D,F"} | B2 | json_match | [
"8a/8aca6b5fb1105e470dd2ecc1a3e7395fb5d21e84b03c58eac0f9243389e50c5c.jpg",
"09/096eb7f235cf40c882729741dd8c370c1f09b7fe4fb92cf952b0354d9c24f9b4.jpg",
"f3/f318bcebed38e4e2dfe6d6b4cbc51a07c5439eded32aaf5a471459eec8dbaa77.jpg",
"a8/a800583ef185d7a11aaebe1d4fd49c43795899e941e8b2a3942fc9e0678ef687.jpg",
"be/be90... | [{"image":"8a/8aca6b5fb1105e470dd2ecc1a3e7395fb5d21e84b03c58eac0f9243389e50c5c.jpg"},{"image":"09/096eb7f235cf40c882729741dd8c370c1f09b7fe4fb92cf952b0354d9c24f9b4.jpg"},{"image":"f3/f318bcebed38e4e2dfe6d6b4cbc51a07c5439eded32aaf5a471459eec8dbaa77.jpg"},{"image":"a8/a800583ef185d7a11aaebe1d4fd49c43795899e941e8b2a3942fc9... | np_b2_05_001__zh_interleave | np_b2_05_001 | zh_interleave | interleave | json | {} | np_b2_05 | scientific | zh | np_b2_05_001 | [
"images/SciDocBench/8a/8aca6b5fb1105e470dd2ecc1a3e7395fb5d21e84b03c58eac0f9243389e50c5c.jpg",
"images/SciDocBench/09/096eb7f235cf40c882729741dd8c370c1f09b7fe4fb92cf952b0354d9c24f9b4.jpg",
"images/SciDocBench/f3/f318bcebed38e4e2dfe6d6b4cbc51a07c5439eded32aaf5a471459eec8dbaa77.jpg",
"images/SciDocBench/a8/a8005... | |
466 | 在我提供的文档中，Figure 1 右侧有一个雷达图。你的任务是在该图中找出一个错误。
图表描述：该图是使用 Python 的 matplotlib 包生成的，所有实验数据随后都标注在图上。已知所有数据标注都是正确的；然而，在 matplotlib 绘图过程中，有一个数值被错误设置，导致某条曲线在某个特定 benchmark 上出现在了错误的位置。
请按照以下步骤：
1. 识别图表中的绘图错误。
2. 识别与该错误对应的已标注数值。
3. 识别与该错误对应的 benchmark 和 model。
4. 按照以下精确格式将你的答案作为 JSON 字典返回：
{"value": "<The corresponding numerica... | {"value": "63.3", "model": "Qwen-VL-Chat", "benchmark": "GQA"} | C2 | json_match | [
"75/75fd5ca64e30c69775982cbc7a4b0262ded359528f21a72992d912a9fab7abbb.jpg",
"24/24f5945dee28f78d4721697550e3d706efab13c7579b71541b002e10c04ccf1a.jpg",
"9a/9a4e07a84f837b426db0076643af3b96f4eb70f9c7d1814275e9e920684fd0ef.jpg",
"fd/fd1e4c40ee310d7723f396d79b9a4434929f20e5b72b4abf2b5db1ff8cfd7717.jpg",
"77/7759... | [{"text":"# Improved Baselines with Visual Instruction Tuning\n\nLarge multimodal models (LMMs) have become increasingly popular in the research community, as they are the key building blocks towards general-purpose assistants $[2, 30, 43]$ . Recent studies on LMMs are converging on a central concept known as visual in... | np_c2_02_001__zh_interleave | np_c2_02_001 | zh_interleave | interleave | json | {} | np_c2_02 | scientific | zh | np_c2_02_001 | [
"images/SciDocBench/75/75fd5ca64e30c69775982cbc7a4b0262ded359528f21a72992d912a9fab7abbb.jpg",
"images/SciDocBench/24/24f5945dee28f78d4721697550e3d706efab13c7579b71541b002e10c04ccf1a.jpg",
"images/SciDocBench/9a/9a4e07a84f837b426db0076643af3b96f4eb70f9c7d1814275e9e920684fd0ef.jpg",
"images/SciDocBench/fd/fd1e4... | |
467 | 关于这篇文章，以下陈述是正确的：
A. 图 1 仅用于说明概念起源，与论文后续提出的理论推导无关。
B. 根据图 3 和图 4，半圆形梁表现出最差的拉伸–压缩对称性。
C. 在图 7 中，蓝色结构提供稳定的正刚度，而红色结构在一定范围内表现出负刚度。
D. 在图 B2 中，Force > 0 表示该结构受到拉伸载荷。
E. 图 2(e) 所示的结构在论文中得到了实验验证。
F. 附录 C 表明，结构常数 Q 的增加会导致 NS 区域的跨度和力幅值均增长。
按以下格式作答：
{"Choice":"<e.g. A,B,C,D,E>"(use comma to seperate the options), "Reason":"<Expla... | {"Choice": "B,D"} | C3 | json_match | [
"e3/e3ca806b38f95a6343635d61bb90b57bee606a8cde51ec771d9b5574726ef15c.jpg",
"ef/ef68cc91e776b93592f174850d0319e566298675b7a63e4281690e87c68d16b1.jpg",
"b3/b397d6b27739a32b8a8508a29d71bde865422fbe7886095d1055c1b4087b0e02.jpg",
"fd/fdaa3818936864fab8e78e3451463d02b36f3fade7cb73ccbf8c15c90d7ceb55.jpg",
"13/13c9... | [{"text":"# A tensegrity-inspired bidirectional quasi-zero stiffness metamaterial for buffering and energy absorption"},{"image":"e3/e3ca806b38f95a6343635d61bb90b57bee606a8cde51ec771d9b5574726ef15c.jpg"},{"text":"## 1. Introduction\n\nMetamaterials are artificially designed with microstructures that are periodically ar... | np_c3_05_001__zh_interleave | np_c3_05_001 | zh_interleave | interleave | json | {} | np_c3_05 | scientific | zh | np_c3_05_001 | [
"images/SciDocBench/e3/e3ca806b38f95a6343635d61bb90b57bee606a8cde51ec771d9b5574726ef15c.jpg",
"images/SciDocBench/ef/ef68cc91e776b93592f174850d0319e566298675b7a63e4281690e87c68d16b1.jpg",
"images/SciDocBench/b3/b397d6b27739a32b8a8508a29d71bde865422fbe7886095d1055c1b4087b0e02.jpg",
"images/SciDocBench/fd/fdaa3... | |
468 | 你是一个有帮助的 AI Article Reader。第一个 PDF 是 "Strategies for Pre-Training Graph Neural Networks"（基线论文）。仅基于该基线论文，列出作为 pretrain GNN 变体进行评估的 GNN 模型架构。输出一个按字母顺序排序的模型名称 JSON 列表。
示例输出：
["GAT", "GCN", "GIN", "GraphSAGE"]
将该列表包装在键 "answer_list" 下，因此最终输出为：
{"answer_list": [...]} | {"answer_list": ["GAT", "GCN", "GIN", "GraphSAGE"]} | D2 | json_match | [
"01/019b816fc22ae0cbdda69c1eaacb82fc085abdb50635258d832bca2817399f08.jpg"
] | [{"text":"<table><tr><td colspan=\"2\">Dataset</td><td>BBBP</td><td>Tox21</td><td>ToxCast</td><td>SIDER</td><td>ClinTox</td><td>MUV</td><td>HIV</td><td>BACE</td><td>Average</td></tr><tr><td colspan=\"2\"># Molecules</td><td>2039</td><td>7831</td><td>8575</td><td>1427</td><td>1478</td><td>93087</td><td>41127</td><td>151... | np_d2_01_001__zh_interleave | np_d2_01_001 | zh_interleave | interleave | json | {} | np_d2_01 | scientific | zh | np_d2_01_001 | [
"images/SciDocBench/01/019b816fc22ae0cbdda69c1eaacb82fc085abdb50635258d832bca2817399f08.jpg"
] | |
469 | 你是一个用于生物学文章分析的 AI。基于所提供的页面，列出这些工作中明确被命名为使用过的顶层数据库。使用规范简称。不要列出子数据集或编造数据库。包括任何被明确描述为这些工作中所使用的数据来源的数据库（涵盖蛋白质序列/结构数据库、变异数据库、PTM 数据库、抗菌肽数据库、生物样本库等）。
输出一个按字母顺序排序（不区分大小写）的字符串 JSON 列表：
["<DB_1>", "<DB_2>", ...]
将该列表包装在键 "answer_list" 下，因此最终输出为：
{"answer_list": [...]} | {"answer_list": ["ActiveDriverDB", "ClinVar", "Genebass", "PhosphoSitePlus", "PTMint", "ProteinGym", "UniProt", "dbPTM", "gnomAD"]} | D2 | json_match | [
"ef/ef0a3ae63b7d97588397e1b672213d34b692bac42854540605e5fabfbefe9ac5.jpg",
"6c/6ced68d7137bb3b75c262c085cd71d85a6c03b8007ef50e2a38b765c04027dc4.jpg",
"20/20b1cff4fcd73c4708dc13d44c5f750b6de4eb07919559dce36137bd2fc0d048.jpg",
"ee/ee09133536a1c11f8bd107d88d67b11182d331fc677643c75ed777371609f062.jpg",
"0c/0c55... | [{"text":"to dominant clusters, yielding limited structural novelty and reduced broad-spectrum activity [8]. Experiments show that ProtFlow outperforms various state-of-the-art generative methods in generating high-quality peptides, and realistic, diverse, and novel AMPs with broad spectrum activity, especially against... | np_d2_02_001__zh_interleave | np_d2_02_001 | zh_interleave | interleave | json | {} | np_d2_02 | scientific | zh | np_d2_02_001 | [
"images/SciDocBench/ef/ef0a3ae63b7d97588397e1b672213d34b692bac42854540605e5fabfbefe9ac5.jpg",
"images/SciDocBench/6c/6ced68d7137bb3b75c262c085cd71d85a6c03b8007ef50e2a38b765c04027dc4.jpg",
"images/SciDocBench/20/20b1cff4fcd73c4708dc13d44c5f750b6de4eb07919559dce36137bd2fc0d048.jpg",
"images/SciDocBench/ee/ee091... | |
470 | 给定两篇关于在 caption 数据集 ScaleCap-450k 和 CapRL-1M 上训练 Qwen2.5-VL-7B 的科学 PDF 文档。列出同时出现在这两个模型的两篇文档主评估表中的 benchmark 名称。按字母顺序排序（不区分大小写）。使用 ScaleCap 文档中显示的大小写形式。
输出一个 benchmark 名称的 JSON 列表：
["<Benchmark_1>", "<Benchmark_2>", ...]
将该列表包装在键 "answer_list" 下，因此最终输出为：
{"answer_list": [...]} | {"answer_list": ["ChartQA", "InfoVQA", "MMStar", "MMVet"]} | D3 | json_match | [] | [{"text":"Table 1: Performance comparison using different pretraining datasets. CapRL-1M significantly outperforms other datasets across all 3 settings, and further improvements are observed when scaling the data to 5M. The best results are bold and the second-best results are underlined.\n\n<table><tr><td>Pretraining ... | np_d3_01_001__zh_interleave | np_d3_01_001 | zh_interleave | interleave | json | {} | np_d3_01 | scientific | zh | np_d3_01_001 | [] | |
471 | 你是一名化学论文阅读助手。提供的 PDFs 是关于 NCM-811 正极材料及其循环稳定性的研究文章。在所有提供的文章中，找出在恰好 500 次循环时报告的 NCM-811（或其改性形式）的最高容量保持率百分比。报告该数值以及该测试的电压窗口。
按以下精确格式输出 JSON：
{
"highest_retention_at_500_cycles_pct": <number>,
"voltage_window": {"lower_V": <number>, "upper_V": <number>}
}
规则：
- 仅使用所提供页面中可见的内容。
- 如果没有测试报告 NCM-811 在恰好 500 次循环时的数据，则输出 ... | {"highest_retention_at_500_cycles_pct": null, "voltage_window": null} | D3 | json_match | [
"d7/d7fdb8587e0d89e2048a2d9213de48c16663f00745dc789dd281100202612723.jpg",
"e4/e4a76e0ba7d347746ffcac75a5380e443d13523ef1af4e817a7d4b6c6654e6df.jpg",
"7b/7b42f28235bc661a9cb26be07a5fdf7c7412190e928d50c530335dffcf003fd8.jpg",
"85/853b6d8b7f766b415cfd035e13d2caa89d7ade403ea4affb3181d3986242ab31.jpg",
"ba/ba07... | [{"image":"d7/d7fdb8587e0d89e2048a2d9213de48c16663f00745dc789dd281100202612723.jpg"},{"image":"e4/e4a76e0ba7d347746ffcac75a5380e443d13523ef1af4e817a7d4b6c6654e6df.jpg"},{"image":"7b/7b42f28235bc661a9cb26be07a5fdf7c7412190e928d50c530335dffcf003fd8.jpg"},{"text":"Figure 1. XRD patterns of (a) the uncoated NMC811 and LiF-... | np_d3_02_001__zh_interleave | np_d3_02_001 | zh_interleave | interleave | json | {} | np_d3_02 | scientific | zh | np_d3_02_001 | [
"images/SciDocBench/d7/d7fdb8587e0d89e2048a2d9213de48c16663f00745dc789dd281100202612723.jpg",
"images/SciDocBench/e4/e4a76e0ba7d347746ffcac75a5380e443d13523ef1af4e817a7d4b6c6654e6df.jpg",
"images/SciDocBench/7b/7b42f28235bc661a9cb26be07a5fdf7c7412190e928d50c530335dffcf003fd8.jpg",
"images/SciDocBench/85/853b6... | |
472 | 在论文的 Fig. 5b 中，展示了三层垂直排列的节点。列出中间层中各块的标签（从上到下）以及右侧层中各节点的标签（从上到下）。
按以下精确格式输出 JSON：
{
"middle_layer_top_to_bottom": ["<label_1>", "<label_2>", ...],
"right_layer_top_to_bottom": ["<label_1>", "<label_2>", ...]
} | {"middle_layer_top_to_bottom": ["Progenitor maintenance", "Haemoglobin synthesis", "Autophagy", "Cell cycle"], "right_layer_top_to_bottom": ["MCH", "RDW", "IRF"]} | E1 | json_match | [
"48/482ca29e9d73ae53fb14a6b1117bb349915c9855abe5d2d0b37e41cfaebb74d8.jpg",
"e7/e701f66195bb314b9fed7f5dd6cb1e0d741b362ab99a16ba044119056c754b87.jpg"
] | [{"text":"## Article"},{"image":"48/482ca29e9d73ae53fb14a6b1117bb349915c9855abe5d2d0b37e41cfaebb74d8.jpg"},{"text":"b Program regulatory map explains cross-trait relationships \nC Program association with erythropoiesis"},{"image":"e7/e701f66195bb314b9fed7f5dd6cb1e0d741b362ab99a16ba044119056c754b87.jpg"},{"text":"Fig.... | np_e1_01_001__zh_interleave | np_e1_01_001 | zh_interleave | interleave | json | {} | np_e1_01 | scientific | zh | np_e1_01_001 | [
"images/SciDocBench/48/482ca29e9d73ae53fb14a6b1117bb349915c9855abe5d2d0b37e41cfaebb74d8.jpg",
"images/SciDocBench/e7/e701f66195bb314b9fed7f5dd6cb1e0d741b362ab99a16ba044119056c754b87.jpg"
] | |
473 | 你是一名 AI 生物信息学研究助手。以下是文章的文件结构树：
```
├── 01_identify_mutations
│ ├── 140gene_fasta_new_8species.R
│ ├── TBXT_8species_03.csv
│ ├── TBXT_new_8species.fasta
│ ├── forloop_python.sh
│ ├── gene140_location.csv
│ └── mutation.py
├── 02_define_mutations
│ ├── inputs
│ ├── mutation_classifier.R
│ └── outputs
... | {
"question_1": {
"script_name": "mutation.py",
"folder_path": "01_identify_mutations"
},
"question_2": {
"dna_analysis_chain": [
"01_identify_mutations/mutation.py",
"02_define_mutations/mutation_classifier.R",
"04_filter_vep_results/filter_vep_visualization.R"
]
}
} | F2 | json_match | [
"a5/a5ea93fe4c325a63835ae0a0db0364d8ca2720b0b897c872755032f6c9f2f204.jpg",
"b7/b7f2c68826ae283911a17fb082604782b2626cd674cc9698cb44db7a9dc8441c.jpg",
"03/03100a37311774f6f45b53046c20456fa863e1485b94b46ce1ea646ccf7c7225.jpg",
"82/8253e6f496055504aa157e7941dad5d690cb8ab5ed8a92ea3f5ab0e19aad3fd0.jpg",
"06/06a4... | [{"text":"# On the genetic basis of tail-loss evolution in humans and apes\n\nIt has long been speculated that tail loss in hominoids contributed to orthograde and bipedal locomotion, the evolutionary occurrence of which coincided with the loss of the tail $^{15-17}$ . Yet, the genetic mechanism that facilitated either... | np_f2_01_001__zh_interleave | np_f2_01_001 | zh_interleave | interleave | json | {} | np_f2_01 | scientific | zh | np_f2_01_001 | [
"images/SciDocBench/a5/a5ea93fe4c325a63835ae0a0db0364d8ca2720b0b897c872755032f6c9f2f204.jpg",
"images/SciDocBench/b7/b7f2c68826ae283911a17fb082604782b2626cd674cc9698cb44db7a9dc8441c.jpg",
"images/SciDocBench/03/03100a37311774f6f45b53046c20456fa863e1485b94b46ce1ea646ccf7c7225.jpg",
"images/SciDocBench/82/8253e... | |
474 | 分析所提供的量子计算综述论文，并识别论文的分类体系中明确分类的所有被引用研究工作。根据以下维度分配分类标签：
Dimension 1: Basic_Characteristics
- QI
- CQCT
- PM
Dimension 2: Algorithmic_Characteristics
- P
- ACQA
- T
- TLG
- QO
Dimension 3: Time_and_Gate_Characteristics
- DT
- MT
Dimension 4: Other_Characteristics
- S
- TCGLQ
输出格式：
仅返回一个 JSON 对象，其中：
- 键：格式为 "Author ... | {
"Weitenberg et al.(2011)": [
"Basic_Characteristics:QI",
"Basic_Characteristics:CQCT",
"Basic_Characteristics:PM",
"Algorithmic_Characteristics:P",
"Algorithmic_Characteristics:ACQA",
"Algorithmic_Characteristics:TLG",
"Time_and_Gate_Characteristics:DT",
"Other_Characteristics:TCGLQ"... | A3 | judge | [
"8c/8c54610ec6d6ce36c2c039fb1de6bf0e13f860c312438d8d4221ad244c05f4ad.jpg",
"a6/a68e15d078dc86d61f44b29b5fd8323e0af689877d56a20f241df183e2e92d0f.jpg",
"01/017467834e593ee8e5e6bf062bb69922dbfdc0976834b9b3a39b33a3a1ccb030.jpg"
] | [{"text":"computation speed will grow faster by using these quantum algorithms. Li et al. [35] discussed merging the elements of quantum mechanics with the intelligent nature-inspired algorithms to mark the new era of computing in the making. Quantum optimization and quantum learning are two classifications based on wh... | np_a3_01_002__zh_interleave | np_a3_01_002 | zh_interleave | interleave | json | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. The initial score is set as 0.0.
2. Format score (... | {} | np_a3_01 | scientific | zh | np_a3_01_002 | [
"images/SciDocBench/8c/8c54610ec6d6ce36c2c039fb1de6bf0e13f860c312438d8d4221ad244c05f4ad.jpg",
"images/SciDocBench/a6/a68e15d078dc86d61f44b29b5fd8323e0af689877d56a20f241df183e2e92d0f.jpg",
"images/SciDocBench/01/017467834e593ee8e5e6bf062bb69922dbfdc0976834b9b3a39b33a3a1ccb030.jpg"
] |
475 | 根据文档，回答以下问题。
在 Figure 1 中，虚线和实线分别表示什么？ | The solid line represents the activations passed to the next layer, and the dashed line represents the activations passed to the reverse network. | A5 | judge | [
"38/38892c57ecf59bcc8a286189ce4764a4fadc4e734e8d68eb977108a5fc85df80.jpg",
"31/3193c32b8edd3461db567ecd25997829b994b3a47597000fb5b8876038afc0d2.jpg"
] | [{"image":"38/38892c57ecf59bcc8a286189ce4764a4fadc4e734e8d68eb977108a5fc85df80.jpg"},{"text":"Figure 1: (a) Overview of the bidirectional spike-based distillation framework. (b) Illustration of feature alignment using the properties of pyramidal neurons, which receive feedforward and feedback signals through their basa... | np_a5_05_002__zh_interleave | np_a5_05_002 | zh_interleave | interleave | text | {} | np_a5_05 | scientific | zh | np_a5_05_002 | [
"images/SciDocBench/38/38892c57ecf59bcc8a286189ce4764a4fadc4e734e8d68eb977108a5fc85df80.jpg",
"images/SciDocBench/31/3193c32b8edd3461db567ecd25997829b994b3a47597000fb5b8876038afc0d2.jpg"
] | |
476 | 根据 Fig.3，回答以下问题。
1. 对于 \chi_M''-频率曲线，描述峰值坐标的变化趋势。（在 'Figure 3(b) curve' 字段中回答）
2. 确定 51K 时 \chi_M''-频率曲线峰值的频率。（给出精确数字）
按以下格式作答：
{"Figure 3(b) curve": "...", "Peak at 51K": "<e.g. 0.001 Hz>"} | {
"Figure 3(b) curve": "As temperature increases from 30 K to 51 K, the peak shifts to higher frequencies (moves rightward on the frequency axis) and the peak intensity decreases.",
"Peak at 51K": "2000 Hz"
} | C3 | judge | [
"ae/ae0d0e601dcb7a1e15f4f6cef12fa8b90d69dafe527b899291ec4d22d1b5fb28.jpg",
"cb/cb66ad405b99ff7b547341f5eac983731fdcc6252440a1052e66ee2af992182f.jpg",
"8c/8c559f756f2332985f42fda7a242d6c1121e02aee56bfb8e224de9122adb5108.jpg",
"2e/2e1d8543cb758eeddf7b1bee280e05d9a25d177a468b80de0c7686336a4a1146.jpg",
"35/3557... | [{"text":"ARTICLE\n\nSelvan Demir $^{1,2}$ , Miguel I. Gonzalez $^{1}$ , Lucy E. Darago $^{1}$ , William J. Evans $^{3}$ & Jeffrey R. Long $^{1,4,5}$\n\nCoordination chemistry can be utilized to yield electronic structures designed to prevent through-barrier relaxation pathways, with the goal of attaining magnetic bloc... | np_c3_04_002__zh_interleave | np_c3_04_002 | zh_interleave | interleave | json | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. If the model correctly answers '2000Hz' for 'Peak a... | {} | np_c3_04 | scientific | zh | np_c3_04_002 | [
"images/SciDocBench/ae/ae0d0e601dcb7a1e15f4f6cef12fa8b90d69dafe527b899291ec4d22d1b5fb28.jpg",
"images/SciDocBench/cb/cb66ad405b99ff7b547341f5eac983731fdcc6252440a1052e66ee2af992182f.jpg",
"images/SciDocBench/8c/8c559f756f2332985f42fda7a242d6c1121e02aee56bfb8e224de9122adb5108.jpg",
"images/SciDocBench/2e/2e1d8... |
477 | 以下是关于一组人之间互动的描述。基于 figure 1(b)，哪个子图最能描述他们的互动类型？该子图中人物与图标之间可能对应的关系是什么？
你需要首先定位哪个子图表示这组人之间的互动类型。然后你需要从四个选项中选择所有可能正确的对应关系。（例如 Andy-Pink，表示 Andy 可以是子图中的粉色人物）
Group:
Peter always initiate interaction with Emma, while Emma more often rejects his requests. Sofia tends to start interaction with Emma, but Emma rejects it ofte... | {"Subfigure": "Advice requests", "Choice": "C,D"} | A2 | judge | [
"4f/4fd1e8fb506a5bf28eac26cc7c03d88d35ac44b20a9ca5f262e473ff7dc05f15.jpg",
"3b/3bf407fe1cef2e205963fd12164dc0300226f398278aa1c8c00fb8e6307098f6.jpg",
"f8/f8482bebbe3e287eed4d35630024561794779b2f9661706b97a7e34504626a1b.jpg",
"a8/a8f13cfafd6ece8bc699dae60a624452add973f1b16a82bec693aca27c545431.jpg",
"57/574c... | [{"text":"# Inferring the internal structure of groups through the integration of statistical learning and causal reasoning\n\nHuman social life unfolds within richly structured networks of overlapping relationships, including friendships, hierarchies, and collaborations. Yet the observable interactions that reveal the... | np_a2_02_002__zh_interleave | np_a2_02_002 | zh_interleave | interleave | json | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. If the model's predicted 'Subfigure' is correct, 0... | {} | np_a2_02 | scientific | zh | np_a2_02_002 | [
"images/SciDocBench/4f/4fd1e8fb506a5bf28eac26cc7c03d88d35ac44b20a9ca5f262e473ff7dc05f15.jpg",
"images/SciDocBench/3b/3bf407fe1cef2e205963fd12164dc0300226f398278aa1c8c00fb8e6307098f6.jpg",
"images/SciDocBench/f8/f8482bebbe3e287eed4d35630024561794779b2f9661706b97a7e34504626a1b.jpg",
"images/SciDocBench/a8/a8f13... |
478 | 你是一个生物学文章分析 AI。请告诉我这些工作中使用的所有数据库。
你的回答格式是：
{
<dataset_name> : [
<subdataset_name1>,
<subdataset_name2>,
...
],
...
}
请只给我 json | {
"UniProt": [
"Swiss-Prot (February 2022 release)",
"UniProtKB reviewed",
"UniRef",
"UniRef50",
"UniProt ID mapping server"
],
"ClinVar": [
"ProteinGym ClinVar benchmark",
"Balanced ClinVar (2,400 genes)",
"ClinVar 2025 release (March 2025)",
"ClinVar (142,951 variants subset ... | A4 | judge | [
"54/540d7a2d34d0ca9aba3c756abf62b38c0ae529264b83abbb9d18c34160e6f6ae.jpg",
"a2/a2707320c5c1c2766e22a32ec699c851339a5afea70e4be16229fb218bff1067.jpg",
"b2/b22ae79cdb1b9977368330a703fc19b675950c9d4b44401939dbd26cd8decc98.jpg",
"f0/f035e01cec2586cae4383bcd45449b4357251bdad72fdeb5465d7c863568820f.jpg",
"b5/b559... | [{"text":"Article\n\nReceived: 10 May 2025\n\nCheck for updates\n\nProtein language models (PLMs) have recently emerged as a promising approach for next-generation variant-effect prediction (VEP). Most high-performing VEP methods currently utilize PLMs combined with additional information, such as homology, protein str... | np_a4_02_001_dup2__zh_interleave | np_a4_02_001_dup2 | zh_interleave | interleave | json | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. The initial score is set as 0.0.
2. Format score (... | {} | np_a4_02 | scientific | zh | np_a4_02_001_dup2 | [
"images/SciDocBench/54/540d7a2d34d0ca9aba3c756abf62b38c0ae529264b83abbb9d18c34160e6f6ae.jpg",
"images/SciDocBench/a2/a2707320c5c1c2766e22a32ec699c851339a5afea70e4be16229fb218bff1067.jpg",
"images/SciDocBench/b2/b22ae79cdb1b9977368330a703fc19b675950c9d4b44401939dbd26cd8decc98.jpg",
"images/SciDocBench/f0/f035e... |
479 | 根据图 1，在哪些能量区间中，prompt signals 的计数少于 delay signals 的计数？ | 2 to 2.5 MeV and 9 to 10 Mev，approximately. | A5 | judge | [
"de/ded8d193d530cdaa4c98122821271e1a22510a3bbfe94d878ce979a14ab4119e.jpg",
"59/599c81bf1d32771f0e42695e9ce7fdf9bae12bd84a1b736b66ff996e743c960f.jpg",
"34/349b5940d69da9622dad0a58d7514a929cf4faa0af0caf35bda0b09baaa3ee94.jpg",
"dc/dc8c1fcff01dc21d0cbd7de395de60908fd34ebb8b11461ebf02779e12047355.jpg",
"82/8276... | [{"text":"# New limits on the Pauli forbidden transitions in ${}^{12}$ C nuclei obtained with the complete Borexino dataset\n\nD. Basilico $^{1}$ , G. Bellini $^{1}$ , J. Benziger $^{2}$ , R. Biondi $^{3,10}$ , B. Caccianiga $^{1}$ , A. Caminata $^{5}$ , A. Chepurnov $^{6}$ , D. D'Angelo $^{1}$ , A. Derbin $^{7,8}$ , A... | np_a5_04_002__zh_interleave | np_a5_04_002 | zh_interleave | interleave | text | {} | np_a5_04 | scientific | zh | np_a5_04_002 | [
"images/SciDocBench/de/ded8d193d530cdaa4c98122821271e1a22510a3bbfe94d878ce979a14ab4119e.jpg",
"images/SciDocBench/59/599c81bf1d32771f0e42695e9ce7fdf9bae12bd84a1b736b66ff996e743c960f.jpg",
"images/SciDocBench/34/349b5940d69da9622dad0a58d7514a929cf4faa0af0caf35bda0b09baaa3ee94.jpg",
"images/SciDocBench/dc/dc8c1... | |
480 | 你是一个有帮助的 AI Artical Reader。请阅读以下论文，并总结不同 pretrain GNN 变体在分子任务数据集上的性能，使用所有策略取得的最佳分数，包括分数、准确率。
你需要阅读的论文是：
1. Strategies for Pre-Training Graph Neural Networks，以下简称 baseline
2. ChemRL-GEM: Geometry Enhanced Molecular Representation Learning for Property Prediction，以下简称 improve1
3. Molecular Contrastive Learning of Repres... | ```
| Model \ Dataset | BBBP | Tox21 | ToxCast | SIDER | ClinTox | MUV | HIV | BACE | ESOL | FreeSolv | Lipo | QM7 | QM8 | QM9 |
| --------------- | ------- | ------- | ---------------- | ------- | ------- | ------- | ------- | ------- | -------- | -------- | -------... | D2 | judge | [
"7f/7f593ba614ebabe052cb2c5d585c064a24f958f89ad8758d5d51e1c710515a1c.jpg",
"5b/5b56c7db42e16247d6d319d2755e5e93ba86ebe61307ab2df08080de249f7994.jpg",
"03/03ebd4b91afec1b0b5cf3dd7cff4274f52d2f8e9db3d80b6f4ff3324865a63d9.jpg",
"17/173abdf49d3f6e9d3671efe3b49a0aa03fe52c1383dbd94ffbe8dd39f0bb17a0.jpg",
"9c/9ce7... | [{"text":"# Molecular Contrastive Learning of Representations via Graph Neural Networks\n\nMolecular representation is fundamental and essential in the design of functional and novel chemical compounds $^{1-3}$ . Due to the enormous magnitude of possible stable chemical compounds, development of an informative represen... | np_d2_01new_001__zh_interleave | np_d2_01new_001 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the question, the ground-truth and the model's prediction. You need to score the model's prediction.
Question: {prompt}
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. The initial scor... | {} | np_d2_01new | scientific | zh | np_d2_01new_001 | [
"images/SciDocBench/7f/7f593ba614ebabe052cb2c5d585c064a24f958f89ad8758d5d51e1c710515a1c.jpg",
"images/SciDocBench/5b/5b56c7db42e16247d6d319d2755e5e93ba86ebe61307ab2df08080de249f7994.jpg",
"images/SciDocBench/03/03ebd4b91afec1b0b5cf3dd7cff4274f52d2f8e9db3d80b6f4ff3324865a63d9.jpg",
"images/SciDocBench/17/173ab... |
481 | 你将获得两篇科学 PDF 文档。你的任务是总结在数据集 ScaleCap-450k 和 CapRL-1M 上训练 Qwen2.5-VL-7B 后得到的模型的性能。只保留两个文档中都出现的基准测试上的性能。
执行以下步骤：
- 在两个文档中找到对应的表格。
- 找出两个表格中出现的基准测试的交集。
- 提取基准测试名称以及两个模型的准确率值。
- 按字母顺序对基准测试进行排序。（排序时忽略大小写，例如 "a" 排在 "B" 前面。）
- 写基准测试名称时，不要包含任何空格或换行，并将所有字符保持在一起。大小写应与 ScaleCap 中的表格一致。
- 按以下列表格式输出：
[{"Name":<Fill in the benchmark... | [{"Name": "AI2D", "ScaleCap-450k": "74.0", "CapRL-1M": "75.3"},
{"Name": "ChartQA", "ScaleCap-450k": "77.8", "CapRL-1M": "80.6"},
{"Name": "DocVQA", "ScaleCap-450k": "85.7", "CapRL-1M": "89.5"},
{"Name": "InfoVQA", "ScaleCap-450k": "51.8", "CapRL-1M": "59.9"},
{"Name": "MathVista", "ScaleCap-450k": "49.7", "CapRL-1... | D3 | judge | [
"a7/a7061e0c9f459c91b4349abef997171518b954020f9ac4795aa4cc95519e190c.jpg",
"d9/d9074f6b4d42e92a491e6ccb4f7fedf6103a264bc7a3ab8ad2f136124810a1ee.jpg",
"8f/8f607b483f7e20df33d84d6f54832d53da1d6fb791691c263537670bf82d42f5.jpg",
"da/da7f94c2f0963fe2f4176b91513cd184c0b89733f80a4c8058879025f11997a2.jpg",
"32/3266... | [{"text":"# ScaleCap: Inference-Time Scalable Image Captioning via Dual-Modality Debiasing\n\nThis paper presents ScaleCap, an inference-time scalable image captioning strategy that generates comprehensive and detailed image captions. The key challenges of high-quality image captioning lie in the inherent biases of LVL... | np_d3_01new_001__zh_interleave | np_d3_01new_001 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Format score (0.04 points): The ground-truth is a ... | {} | np_d3_01new | scientific | zh | np_d3_01new_001 | [
"images/SciDocBench/a7/a7061e0c9f459c91b4349abef997171518b954020f9ac4795aa4cc95519e190c.jpg",
"images/SciDocBench/d9/d9074f6b4d42e92a491e6ccb4f7fedf6103a264bc7a3ab8ad2f136124810a1ee.jpg",
"images/SciDocBench/8f/8f607b483f7e20df33d84d6f54832d53da1d6fb791691c263537670bf82d42f5.jpg",
"images/SciDocBench/da/da7f9... |
482 | 你是一名化学论文阅读助手。请帮助我核验在以下实验条件下，这三篇论文中 NCM-811 或其改性材料的容量保持率数据：
1. 循环次数：$500$ cycles
2. 电压窗口：约 $2.8 \sim 4.5 V$
3. 质量负载：$10 mg cm^{-2}$
4. 充电倍率：$1 C$
5. 放电倍率：$1 C$
6. 面容量：$2 mAh cm^{-2}$
7. N/P 比：约 $1.2$
你的回答格式应为：
{
"<First 5 words of article title>" : {
"result" : <digit placehoder using percentage>,
... | {
"High Voltage Cycling Stability of": {
"result": 88,
"experiment_conditions": {
"cycle_number": 500,
"voltage_window": {
"lower_limit": 2.8,
"supper_limit": 4.5
},
"mass_loading($mg cm^{-2}$)": "10-10.5",
"charge_rate($C$)": 1,
"discharge_rate($C$)": 1,
... | D3 | judge | [
"e7/e7726de563a7a49567ac5faf12545f60911fde848a71d26f2719626df9626a01.jpg",
"2f/2fdcbf302a46a98d2e192dec0fc9d8ce920b86330b1cc70edff56f02aa127c63.jpg",
"53/534491326725fd6194af7898986ab8a077bb70dc2b52a1b158d49ebcbc2b934c.jpg",
"fc/fc3ecbb9c1cf5bddbe31607c667599cdd5f5247b273218c931938d159b11199f.jpg",
"0a/0a80... | [{"text":"pubs.acs.org/cm\n\n# Long-Term Cyclability of NCM-811 at High Voltages in Lithium-Ion Batteries: an In-Depth Diagnostic Study"},{"image":"e7/e7726de563a7a49567ac5faf12545f60911fde848a71d26f2719626df9626a01.jpg"},{"image":"2f/2fdcbf302a46a98d2e192dec0fc9d8ce920b86330b1cc70edff56f02aa127c63.jpg"},{"text":"ABSTR... | np_d3_02new_001__zh_interleave | np_d3_02new_001 | zh_interleave | interleave | json | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Some values inside the ground-truth answers have m... | {} | np_d3_02new | scientific | zh | np_d3_02new_001 | [
"images/SciDocBench/e7/e7726de563a7a49567ac5faf12545f60911fde848a71d26f2719626df9626a01.jpg",
"images/SciDocBench/2f/2fdcbf302a46a98d2e192dec0fc9d8ce920b86330b1cc70edff56f02aa127c63.jpg",
"images/SciDocBench/53/534491326725fd6194af7898986ab8a077bb70dc2b52a1b158d49ebcbc2b934c.jpg",
"images/SciDocBench/fc/fc3ec... |
483 | 以 mermaid 格式复现 Image 5b，使用红色（#cc0000）和蓝色箭头（#0066cc）进行区分。你只需要绘制带箭头的三列。左列以节点 'CAD' 开始。中间层以块 'Progenitor maintenance' 开始，右层以节点 'MCH' 开始。忽略 'CAD' 左侧以及 'MCH' 右侧的列、节点和线。每个节点都应严格遵循图中的名称。每一层中的顺序也应遵循图中从上到下的顺序。箭头颜色只能从 #cc0000 和 #0066cc 中选择。
请按以下格式输出最终答案：
```mermaid
graph LR
subgraph Left Layer
1[CAD]
2[CALR]
end
... | ```
graph LR
subgraph Left Layer
1[CAD]
2[CALR]
3[POLE]
4[SUPT5H]
5[MED17]
6[ATR]
7[GTF3C1]
8[RBBP6]
end
subgraph Middle Layer
9[Progenitor maintenance]
10[Haemoglobin synthesis]
11[Autophagy]
12[Cell cycle]
end
subgraph Right Layer
13[MCH]
14[... | E1 | judge | [
"69/69fb3b747f9a98012e9f0fa8d6572269bc358adbc50cbf1f1b861b04c9b86119.jpg",
"99/99230b4eebac9dbbf91a6388a45a39fd6158ceac166f245854acf5198470b831.jpg"
] | [{"text":"Overall, the Perturb-seq data efficiently captured biological pathways and their regulators, and comparison with gene associations enable us to identify the pathways relevant to each trait.\n\n## Complex interplay of programs\n\nAlthough the significant programs in Fig. 4c–e provide insight into biological co... | np_e1_01new_001__zh_interleave | np_e1_01new_001 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Block score (0.28 points): First check all the blo... | {} | np_e1_01new | scientific | zh | np_e1_01new_001 | [
"images/SciDocBench/69/69fb3b747f9a98012e9f0fa8d6572269bc358adbc50cbf1f1b861b04c9b86119.jpg",
"images/SciDocBench/99/99230b4eebac9dbbf91a6388a45a39fd6158ceac166f245854acf5198470b831.jpg"
] |
484 | 请仔细阅读所附论文，并以 JSON 格式提取以下信息：
1. Model：列出本文提出的所有模型。对于每个模型，提供其名称以及它所基于的基础模型。
2. Data：提供本文构建的训练数据集名称，并列出用于构建它的所有源数据集。
3. Training：按顺序列出本文训练流程中使用的所有训练阶段/方法。对于每个阶段，提供其名称以及论文中明确提到的所有超参数。
将结果输出为一个仅包含三个键的单一 JSON 对象："Model"、"Data" 和 "Training"。严格遵循以下模式：
{
"Model": [
{
"name": "<model name>",
"base_model": "<bas... | {
"Model": [
{
"name": "Spatial-SSRL-3B",
"base_model": "Qwen2.5-VL-3B"
},
{
"name": "Spatial-SSRL-7B",
"base_model": "Qwen2.5-VL-7B"
},
{
"name": "Spatial-SSRL-4B",
"base_model": "Qwen3-VL-4B"
}
],
"Data": {
"name": "Spatial-SSRL-81k",
"source":... | B2 | judge | [
"f4/f4ab7bfd1961eec716fd1d8a933556858e4a2503689ac4f95000821f0d46a08b.jpg",
"b3/b338778f7d447ae15465957a7f5872c3e1c19a3c17f11c508b107d5a46f19928.jpg",
"a5/a55ee03b11c347d91300550a48f1bab80d923bdcc67cc484625e774ebdba5f52.jpg",
"c4/c425123f770801e3b4080213b46a23e5c5f0f3971d2633502e35139761fe5d85.jpg",
"4f/4f22... | [{"text":"# Spatial-SSRL: Enhancing Spatial Understanding via Self-Supervised Reinforcement Learning\n\nYuhong Liu $^{1,2}$ , Beichen Zhang $^{1,3}$ , Yuhang Zang $^{1}$ , Yuhang Cao $^{1}$ , Long Xing $^{1}$ Xiaoyi Dong $^{1}$ , Haodong Duan $^{1}$ , Dahua Lin $^{1,3}$ , Jiaqi Wang $^{1,4}$ $^{1}$ Shanghai AI Laborato... | spatial-ssrl_002__zh_interleave | spatial-ssrl_002 | zh_interleave | interleave | json | You are an expert evaluator the model's response to a question. I will provide you with the question, the ground-truth and the model's prediction. You need to score the model's prediction.
Question: {prompt}
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. First check if t... | {} | spatial-ssrl | cs | zh | spatial-ssrl_002 | [
"images/SciDocBench/f4/f4ab7bfd1961eec716fd1d8a933556858e4a2503689ac4f95000821f0d46a08b.jpg",
"images/SciDocBench/b3/b338778f7d447ae15465957a7f5872c3e1c19a3c17f11c508b107d5a46f19928.jpg",
"images/SciDocBench/a5/a55ee03b11c347d91300550a48f1bab80d923bdcc67cc484625e774ebdba5f52.jpg",
"images/SciDocBench/c4/c4251... |
485 | 你是一名论文阅读助手。请仔细分类本文中的引用文本。主要类别为：
1. Background: 提供理论支持或研究背景（例如，AI 时代的到来）。
2. Method: 采用或比较已有方法。
3. Result: 将他人的结果作为比较基线。
4. Critical: 纯粹批评现有研究的局限性。
注意：
- 你应该只关注带有编号标签的引用
- 你应该只关注给定的引用（编号 24,33）
你必须按照以下格式回答。不要输出 "```"
{
"[1]": {
"{Digital physical page number}-{Chapter name of the smallest section without ... | {
"[24]" : {
"2-Introduction-1" : ["Background"],
"2-Introduction-2" : ["Background"],
"4-Semantically Meaningful Integration to pLM Latent Space-3" : ["Method"]
},
"[33]" : {
"2-Introduction-1" : ["Critical"],
"2-Introduction-1" : ["Background"]
}
} | A3 | judge | [
"ef/ef0a3ae63b7d97588397e1b672213d34b692bac42854540605e5fabfbefe9ac5.jpg",
"6a/6aed035c2512d7b3268aa61d97eb5e6805a8fc79b28fd875152f8dd5d6127eb3.jpg",
"0b/0ba10a0fa08e44dd297c966c1c5632950db3adf43782fbb25f8c49bf044048ea.jpg",
"d7/d7b191bf8574794f3d7e801624037f9dedd198b7d1ee91f476dcf0dc721c2c6d.jpg",
"06/0632... | [{"text":"# ProtFlow: Flow Matching-based Protein Sequence Design with Comprehensive Protein Semantic Distribution Learning and High-quality Generation\n\nZitai Kong $^{1}$ , Yiheng Zhu $^{1}$ , Yinlong Xu $^{1}$ , Mingze Yin $^{1}$ , Tingjun Hou $^{4,6*}$ , Jian Wu $^{3,5*}$ , Hongxia Xu $^{2*}$ , and Chang-Yu Hsieh $... | ProtFlowArticle_002__zh_interleave | ProtFlowArticle_002 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the question, the ground-truth and the model's prediction. You need to score the model's prediction.
Question: {prompt}
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. First check if t... | {} | ProtFlowArticle | biology | zh | ProtFlowArticle_002 | [
"images/SciDocBench/ef/ef0a3ae63b7d97588397e1b672213d34b692bac42854540605e5fabfbefe9ac5.jpg",
"images/SciDocBench/6a/6aed035c2512d7b3268aa61d97eb5e6805a8fc79b28fd875152f8dd5d6127eb3.jpg",
"images/SciDocBench/0b/0ba10a0fa08e44dd297c966c1c5632950db3adf43782fbb25f8c49bf044048ea.jpg",
"images/SciDocBench/d7/d7b19... |
486 | 给定一篇科学 PDF 文档。你的任务是提取并用 LaTex 复现第 6.2 节中标题为 "Resolution metrics" 的小节正文。
严格遵循以下规则：
包含该节正文中的所有连续叙述和段落文本，完全保留原始措辞。
使用 LaTeX 命令保留所有原始粗体和斜体格式：用 \textbf{} 表示粗体，用 \textit{} 表示斜体。
假设 amsthm、amsmath 和 amssymb 已预先定义。保留公式中所有原始字母变体，例如用于花体的 \mathcal{} 和用于黑板粗体的 \mathbb{}。
完全排除以下内容：图注、表注、脚注、脚注标记、方程，以及任何框注或嵌入面板（例如 "Key Insight" 框）内的文... | Let $\mathcal{G}$ denote the ground-truth point set and $\mathcal{R}$ the reconstructed point set under evaluation. We measure accuracy using $\text{dist}(\mathcal{R}\rightarrow \mathcal{G})$ and completeness using $\text{dist}(\mathcal{G}\rightarrow \mathcal{R})$ following [1]. The Chamfer Distance (CD) is then define... | A1 | judge | [
"d7/d7ef77ceb057751425b6f46b1d7cf9368539796729018cd3436dcc5cb7b0690d.jpg",
"5a/5a22e4539681dd3f4edfb98720a1de3d878693417c6d27a127eff023ffb987b0.jpg",
"77/77ec4333863d500d9a3e62027712a7343ce9cd05f29e4266df42a0e22589be27.jpg",
"8f/8f9e551326ee46b54ecb8a49b33ca3889710df4c53095cb4db5f5d4caeab4149.jpg",
"cd/cd39... | [{"text":"# Depth Anything 3: Recovering the Visual Space from Any Views\n\nWe present Depth Anything 3 (DA3), a model that predicts spatially consistent geometry from an arbitrary number of visual inputs, with or without known camera poses. In pursuit of minimal modeling, DA3 yields two key insights: a single plain tr... | depth-anything3_003__zh_interleave | depth-anything3_003 | zh_interleave | interleave | text | {} | depth-anything3 | cs | zh | depth-anything3_003 | [
"images/SciDocBench/d7/d7ef77ceb057751425b6f46b1d7cf9368539796729018cd3436dcc5cb7b0690d.jpg",
"images/SciDocBench/5a/5a22e4539681dd3f4edfb98720a1de3d878693417c6d27a127eff023ffb987b0.jpg",
"images/SciDocBench/77/77ec4333863d500d9a3e62027712a7343ce9cd05f29e4266df42a0e22589be27.jpg",
"images/SciDocBench/8f/8f9e5... | |
487 | 你是一名专业科学编辑和 LaTeX 排版师。
你的目标是为所提供的研究论文内容生成一份全面的 "Index of Notations"。
输出要求：
请使用 booktabs package 生成一个 LaTeX Table 代码块。
该表必须有 3 列：
Symbol: 该符号的 LaTeX 代码（例如，$\mathcal{D}_{KL}$）。
Type: （例如，Scalar, Vector, Function, Set）。
Definition / Meaning: 简洁的英文解释。
First Appearance Snippet: 从文本中逐字引用（最多 15 个词），显示其定义位置。如有必要，使用 "..." 表... | $L$
$\ell$
$m$
$K$
$L_{\mathrm{res}}$
$L_{\mathrm{tr}}$
$x$
$y$
$B$
$\mathcal{L}_B(\theta)$
$\theta$
$\theta^+$
$\eta$
$\eta^\star$
$z^{(\ell)}(x;\theta)$
$z^{(\ell)}_i(x;\theta)$
$\sigma(\cdot)$
$\sigma'(\cdot)$
$q$
$\fan_{\mathrm{in}}(W)$
$W^{(\ell)}$
$W_{\mathrm{res}}$
$\partial_a$
$\Delta\theta_a$
$\Delta z^{(\ell)... | B1 | judge | [
"50/50979f59f6a72f7bd97cd524904a359f9c56617bae7410790d69ed06cd7dc002.jpg",
"b5/b55dd0a8b031104968d81463bef7007ce73be94c91dda133753b018b640a498f.jpg",
"10/10c593a6f50ca705eca2f7088a1b35f584039788c683e1f342137a17fd9f80a9.jpg",
"77/779f754f05260ec3d3b51f8961cdb31db1677aa421008cad67411dd0c31d9d38.jpg",
"8c/8cdd... | [{"text":"# Hyperparameter Transfer Laws for Non-Recurrent Multi-Path Neural Networks\n\nDeeper modern architectures are costly to train, making hyperparameter transfer preferable to expensive repeated tuning. Maximal Update Parametrization ( $\\mu P$ ) helps explain why many hyperparameters transfer across width. Yet ... | hptransfer_002__zh_interleave | hptransfer_002 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the question, the ground-truth and the model's prediction. You need to score the model's prediction.
Question: {prompt}
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Format Score (0.... | {} | hptransfer | cs | zh | hptransfer_002 | [
"images/SciDocBench/50/50979f59f6a72f7bd97cd524904a359f9c56617bae7410790d69ed06cd7dc002.jpg",
"images/SciDocBench/b5/b55dd0a8b031104968d81463bef7007ce73be94c91dda133753b018b640a498f.jpg",
"images/SciDocBench/10/10c593a6f50ca705eca2f7088a1b35f584039788c683e1f342137a17fd9f80a9.jpg",
"images/SciDocBench/77/779f7... |
488 | 你将获得一篇科学 PDF 文档。你的任务是提取论文中描述的 MM-IFInstruct-23k 数据集的组成和数据来源信息，并将其复现为 LaTeX 中的 TikZ 径向图。
严格遵循以下规则：
提取规则：
- 识别所有为 MM-IFInstruct-23k 作出贡献的源数据集。
- 对于每个源数据集，如果论文中明确说明了确切的数据数量（例如样本数或实例数），则提取该数量。如果未说明数量，则省略标签但仍包含该节点。
- 不要推断或编造文档中未明确出现的任何信息。
图示规则：
- 将 MM-IFInstruct-23k 放置为单个中心节点。
- 将每个源数据集作为周围的叶节点，围绕中心以径向（辐条）布局均匀分布。
- 绘制从每个叶节点... | \documentclass[border=10pt]{standalone}
\usepackage{tikz}
\usetikzlibrary{arrows.meta, positioning}
\definecolor{centerblue}{HTML}{3A8FB7}
\definecolor{leafbg}{HTML}{B5DDE4}
\definecolor{leafborder}{HTML}{6FBCCC}
\definecolor{arrowcolor}{HTML}{5A9EAE}
\begin{document}
\begin{tikzpicture}[
leaf/.style={
dr... | A4 | judge | [
"bf/bf3a7030e7673dd1ff1421e48446f8d901d7cc5e13e00cf374800c4cc0ea1f3a.jpg",
"de/ded604c931d816239c8d45da34cba611b39736ccdefdd98ce684c944d0727249.jpg",
"c0/c08bd6880b3a5970d4f7f0067857d0f178c549f66f302c204af5ccedba4bc7e6.jpg",
"25/258da3c6ffeb1aade023df47a0914055290e35a1e672a66f7bc2b06188f13c29.jpg",
"b9/b960... | [{"text":"# MM-IFEngine: Towards Multimodal Instruction Following"},{"image":"bf/bf3a7030e7673dd1ff1421e48446f8d901d7cc5e13e00cf374800c4cc0ea1f3a.jpg"},{"text":"Figure 1. (a) Limitations of existing Multimodal Instruction Following (IF) benchmarks. (b) Overview of the MM-IFEval benchmark, which significantly surpasses ... | mmifenginenew_001__zh_interleave | mmifenginenew_001 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Format Score (0.2 points): Check whether the predi... | {} | mmifenginenew | cs | zh | mmifenginenew_001 | [
"images/SciDocBench/bf/bf3a7030e7673dd1ff1421e48446f8d901d7cc5e13e00cf374800c4cc0ea1f3a.jpg",
"images/SciDocBench/de/ded604c931d816239c8d45da34cba611b39736ccdefdd98ce684c944d0727249.jpg",
"images/SciDocBench/c0/c08bd6880b3a5970d4f7f0067857d0f178c549f66f302c204af5ccedba4bc7e6.jpg",
"images/SciDocBench/25/258da... |
489 | 你将获得一篇科学 PDF 文档。你的任务是提取论文中描述的 ChartSFT 数据集的组成和数据来源信息，并将其复现为 LaTeX 中的 TikZ 径向图。
请严格遵循以下规则：
提取规则：
- 识别所有为 ChartSFT 做出贡献的源数据集（即 ChartSFT 组成表中列出的列来源）。
- 对于每个源数据集，仅当论文中以单个数字明确陈述其确切总数据量（即所有任务的总和）时，才提取该总数据量。如果某个来源的总量没有作为独立数值明确给出，则省略其箭头上的数量标签，但仍包含该节点。
- ChartSFT 的总体总大小 (39M) 已明确陈述，必须显示在中心节点上。
- 一些源数据集未在组成表中直接命名，而是通过参考文献编号引用。对... | \documentclass[tikz,border=10pt]{standalone}
\usepackage{tikz}
\usetikzlibrary{shapes.geometric, arrows.meta}
\begin{document}
\begin{tikzpicture}[
>=stealth,
thick,
every node/.style={font=\sffamily},
outer/.style={
ellipse,
draw=black,
fill=orange!20,
minimum width=2cm... | B1 | judge | [
"4a/4af90b951af96cabc7d8ae3a447400fcfa6e7618b8db5b41f0a24b6b1353f78e.jpg",
"11/119d8a9bac844e575ec2e62f047487766a0810ff95dc767e222d1876d6421ed1.jpg",
"93/93fe3ab6a0a69aa81d1387d7f377d24ba679513379832b39d5512984a07651f1.jpg",
"74/7408ab4803cfef9f4891a3c3c56ce34d90209955460ac0c1e073a3cce21b50d0.jpg",
"e4/e434... | [{"text":"# ChartAssistant: A Universal Chart Multimodal Language Model via Chart-to-Table Pre-training and Multitask Instruction Tuning\n\n$^{1}$ OpenGVLab, Shanghai AI Laboratory $^{2}$ Shanghai Jiao Tong University $^{3}$ The University of Hong Kong $^{4}$ Nanjing University\n\n## Abstract"},{"image":"4a/4af90b951af... | chartassistant_002__zh_interleave | chartassistant_002 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Format Score (0.2 points): Check whether the predi... | {} | chartassistant | cs | zh | chartassistant_002 | [
"images/SciDocBench/4a/4af90b951af96cabc7d8ae3a447400fcfa6e7618b8db5b41f0a24b6b1353f78e.jpg",
"images/SciDocBench/11/119d8a9bac844e575ec2e62f047487766a0810ff95dc767e222d1876d6421ed1.jpg",
"images/SciDocBench/93/93fe3ab6a0a69aa81d1387d7f377d24ba679513379832b39d5512984a07651f1.jpg",
"images/SciDocBench/74/7408a... |
490 | # 角色
你是 **Scientometrics** 和 **LaTeX Visualization** 方面的专家。你的目标是分析给定科学论文 PDF 中的引用，并使用 LaTeX/TikZ 将它们转换为一个**可视化 Citation Dashboard**。
# 任务
生成一个**独立的、可编译的 LaTeX 文件**，用于可视化输入文本的引用网络。输出必须包括：
1. **一个 TikZ Mindmap**：可视化当前工作（中心）与其参考文献（节点）之间的关系。
# 可视化逻辑（颜色编码）
你必须在 TikZ 代码中应用以下配色方案，以表示引用的**意图**：
- **方法基础（Foundation）** -> `Fo... | \documentclass[border=10pt]{standalone}
\usepackage[dvipsnames]{xcolor}
\usepackage{tikz}
\usetikzlibrary{mindmap,shadows,positioning,calc}
\usepackage{booktabs}
\usepackage{colortbl}
\usepackage{array}
\usepackage{tabularx}
\usepackage{makecell}
\usepackage{fontenc}
\usepackage{geometry}
% ── Custom colours ─────────... | A3 | judge | [
"f6/f655dca68cfca8ca2c0c7b299667d834f627a87f18441c715150c28be7cf7309.jpg",
"5b/5b827bbe34f1a9b1e16632ad2aae08ec50b79ca947756c2a045650896e478570.jpg",
"da/da830b179aed8c38b3acf3359307ce8a27af06edf5ad5d6afcc2163acb547883.jpg"
] | [{"text":"# Spa3R: Predictive Spatial Field Modeling for 3D Visual Reasoning\n\nHaoyi Jiang $^{1*}$ Liu Liu $^{2\\dagger}$ Xinjie Wang $^{2}$ Yonghao He $^{3}$ Wei Sui $^{3}$ Zhizhong Su $^{2}$ Wenyu Liu $^{1}$ Xinggang Wang $^{1\\ddagger}$\n\n$^{1}$ Huazhong University of Science & Technology $^{2}$ Horizon Robotics $... | spa3r_002__zh_interleave | spa3r_002 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Format Score (0.1 points): Check whether the predi... | {} | spa3r | cs | zh | spa3r_002 | [
"images/SciDocBench/f6/f655dca68cfca8ca2c0c7b299667d834f627a87f18441c715150c28be7cf7309.jpg",
"images/SciDocBench/5b/5b827bbe34f1a9b1e16632ad2aae08ec50b79ca947756c2a045650896e478570.jpg",
"images/SciDocBench/da/da830b179aed8c38b3acf3359307ce8a27af06edf5ad5d6afcc2163acb547883.jpg"
] |
491 | 给定三篇科学 PDF 文档。第一篇介绍 Cambrian-1，第二篇介绍 LLaVA-OneVision，第三篇介绍 MAmmoTH-VL。
从第一篇论文中的 Cambrian-7M composition figure 和第二篇论文中的 Single-Image 3.2M composition figure 中，提取每个图中黄色/橙色颜色类别下的所有数据集，并仅使用精确的大小写不敏感名称匹配来计算它们的交集。
然后，从第三篇论文中，识别 MAmmoTH-VL 中定义的所有数据源组。对于每个组，仅使用精确的大小写不敏感名称匹配，计算其与上一步得到的数据集列表的交集。确定哪个组具有最大的交集。
输出一个单一 JSON 对象，包... | {"group":"Group C","count":4,"datasets":["Hateful Memes","RefCOCO","VizWiz","VQAv2"]} | D2 | json_match | [
"37/3715439c3d2a59a7b5950e50beaec883bf13bc90f014bfb96b37ffc362d0436d.jpg",
"0e/0e7c7430efa1caf0d9addaef9002a8e487edac8cfb08ca03ac22ec0824a6d958.jpg",
"6e/6e8bb0f7b968483b4d15c9023a2601f93f72d754c21cbf2f6900453e9a1f87ce.jpg",
"ca/caf2660ba4236e638736ab0b9fcc573d6fd7699a4f0aa7c1a6d5ed6a572b11de.jpg",
"27/271e... | [{"text":"cross-attention to occur multiple times by inserting our proposal throughout the LLM layers—allowing consistent access to the uncompressed visual information (see Fig. 8-right).\n\nHyperparameters To flexibly modulate capacity, we introduce two hyperparameters D and G, which indicate the number of cross-atten... | cambrian-llavaov-mammoth_002__zh_interleave | cambrian-llavaov-mammoth_002 | zh_interleave | interleave | json | {} | ["cambrian1", "llavaov", "mammoth-vl"] | cs | zh | cambrian-llavaov-mammoth_002 | [
"images/SciDocBench/37/3715439c3d2a59a7b5950e50beaec883bf13bc90f014bfb96b37ffc362d0436d.jpg",
"images/SciDocBench/0e/0e7c7430efa1caf0d9addaef9002a8e487edac8cfb08ca03ac22ec0824a6d958.jpg",
"images/SciDocBench/6e/6e8bb0f7b968483b4d15c9023a2601f93f72d754c21cbf2f6900453e9a1f87ce.jpg",
"images/SciDocBench/ca/caf26... | |
492 | # 角色
你是 **Scientometrics** 和 **LaTeX Visualization** 方面的专家。你的目标是分析给定科学论文 PDF 中的引用，并使用 LaTeX/TikZ 将其转换为一个**可视化 Citation Dashboard**。
# 任务
生成一个**独立的、可编译的 LaTeX 文件**，用于可视化输入文本的引用网络。输出必须包括：
1. **一个 TikZ Mindmap**：可视化当前工作（中心）与其参考文献（节点）之间的关系。
# 可视化逻辑（颜色编码）
你必须在 TikZ 代码中应用以下配色方案，以表示引用的**意图**：
- **Method Basis (Foundation)*... | \documentclass[border=10pt]{standalone}
\usepackage[dvipsnames]{xcolor}
\usepackage{tikz}
\usetikzlibrary{mindmap,shadows,positioning,calc}
\usepackage{booktabs}
\usepackage{colortbl}
\usepackage{array}
\usepackage{tabularx}
\usepackage{makecell}
\usepackage{fontenc}
\usepackage{geometry}
% ── Custom colours ─────────... | A3 | judge | [
"5a/5ad2a24c1609a9ddeb84f25c73f030fa84fa3bc5a3aed8c1b96ead662219954f.jpg",
"e7/e7118e702d2cde001fff2c8fbc61d84cb80cba30c10591421ba365639bf84dbc.jpg",
"49/495ed4ea9c52b22c97802f474ab72457344ec4953ae72caec723b43128cafffa.jpg",
"f0/f06e8b053a5dc7195cfb5202b472c0be3ab5f0e11a53c1afeaddf34fbccd40c0.jpg",
"54/54b4... | [{"text":"# Proximal Policy Optimization Algorithms\n\nJohn Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, Oleg Klimov\n\nOpenAI\n\n{joschu, filip, prafulla, alec, oleg}@openai.com\n\n## Abstract\n\nWe propose a new family of policy gradient methods for reinforcement learning, which alternate between sampling... | ppo_003__zh_interleave | ppo_003 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Format Score (0.1 points): Check whether the predi... | {} | ppo | cs | zh | ppo_003 | [
"images/SciDocBench/5a/5ad2a24c1609a9ddeb84f25c73f030fa84fa3bc5a3aed8c1b96ead662219954f.jpg",
"images/SciDocBench/e7/e7118e702d2cde001fff2c8fbc61d84cb80cba30c10591421ba365639bf84dbc.jpg",
"images/SciDocBench/49/495ed4ea9c52b22c97802f474ab72457344ec4953ae72caec723b43128cafffa.jpg",
"images/SciDocBench/f0/f06e8... |
493 | 你将获得 5 个 PDF 文件。第一个 PDF 包含 6 个数据样本的截图，这些样本从上到下垂直排列，编号为 1 到 6。其余 4 个 PDF 文件是研究论文，每篇论文描述一个特定的数据集。你的任务：
1. 仔细检查第一个 PDF 中的 6 个数据样本（从上到下编号为 1–6）。
2. 阅读 4 篇研究论文，以理解每篇论文所描述的数据集模式、字段和结构。
3. 对于每个数据样本，根据列名、字段类型、数据内容和整体结构，判断它是否匹配 4 篇论文中某一篇所描述的数据集。
4. 仅输出你确信某个数据样本属于某篇论文的数据集的配对。并非每个数据样本都一定匹配某篇论文，也并非每篇论文都一定匹配某个数据样本。将你的答案输出为匹配配对的 JSO... | [{"sample": 1, "title": "TwiFF (Think With Future Frames): A Large-Scale Dataset for Dynamic Visual Reasoning"},{"sample": 2, "title": "AGENTVISTA: Evaluating Multimodal Agents in Ultra-Challenging Realistic Visual Scenarios"}, {"sample": 5, "title": "AGENTVISTA: Evaluating Multimodal Agents in Ultra-Challenging Realis... | F1 | judge | [
"29/29371b964844a35b395c8879f1b35b2e665a3f192bea349eb1618db7a588073c.jpg",
"00/009c0043e58db79000c9a411b712d706a28c7f83dac0e32c2c42acf05b5f1f96.jpg",
"0f/0f2a973319182c78e73bca382204d5a46246de335b628f8326beb796f33b18cd.jpg",
"63/63dc8af04f2dc95cd3d0876a191b684a4edfd569d0b1653b0d3f9e83d53147e6.jpg",
"d0/d005... | [{"text":"# TwiFF (Think With Future Frames): A Large-Scale Dataset for Dynamic Visual Reasoning\n\nJunhua Liu $^{1}$ Zhangcheng Wang $^{2}$ Zhike Han $^{31}$ Ningli Wang $^{4}$ Guotao Liang $^{5}$ Kun Kuang $^{1}$"},{"image":"29/29371b964844a35b395c8879f1b35b2e665a3f192bea349eb1618db7a588073c.jpg"},{"text":"## 2.1. Mu... | twiffnew_001__zh_interleave | twiffnew_001 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. First check the format of the prediction. It shoul... | {} | twiffnew | cs | zh | twiffnew_001 | [
"images/SciDocBench/29/29371b964844a35b395c8879f1b35b2e665a3f192bea349eb1618db7a588073c.jpg",
"images/SciDocBench/00/009c0043e58db79000c9a411b712d706a28c7f83dac0e32c2c42acf05b5f1f96.jpg",
"images/SciDocBench/0f/0f2a973319182c78e73bca382204d5a46246de335b628f8326beb796f33b18cd.jpg",
"images/SciDocBench/63/63dc8... |
494 | 我正在阅读两篇文档，它们可能有助于我未来关于 SLE 的研究。在文档 1 中，提到了一种特定的 IgG 糖基化与 SLE 有强相关性，我希望将其纳入我未来的研究。我也有兴趣研究 SLE 中的唾液酸化 IgG。我在文档 2 的图 1 中看到了几种物质。请帮助我定位所有相关物质，这些物质必须同时满足上述两个要求，并检索文档 2 的图 2 中哪些峰(s)包含每种物质。
请思考并按以下格式输出最终答案：
{"Substance 1": "<Substance name, e.g. FA2G1>", "Location": "<e.g. Row 1, Column 1>", "Peak": "<e.g. 1>"}
{"Substance 2"... | {"Substance 1": "FA2G1S1", "Location": "Row 2, Column 6", "Peak": "(16a), (16b)"}
{"Substance 2": "FA1G1S1", "Location": "Row 2, Column 7", "Peak": "15"}
{"Substance 3": "FA2BG1S1", "Location": "Row 2, Column 9", "Peak": "(16b), (17)"}
{"Substance 4": "FA2G2S1", "Location": "Row 3, Column 6", "Peak": "18a, 18b"}
{"Sub... | D3 | judge | [
"24/245e2a2b06f8078a09c7504e10900ff1371db379c5f99c5bd9da5eec01c5e138.jpg",
"c1/c1f4c11929f1562129dff9ccff2539df63b448e28dcc34b838dc75cff5071c2c.jpg",
"4c/4c011fccb9c93e7a2344bf33e734f4902c72dcf10513070d84fc3a461a069238.jpg",
"83/834e5978f93949757cf796285f556f40b39dd877f986ba5cc12da2e8d24cd842.jpg",
"5f/5f9b... | [{"text":"Research review paper"},{"image":"24/245e2a2b06f8078a09c7504e10900ff1371db379c5f99c5bd9da5eec01c5e138.jpg"},{"text":"$^{c}$ University of Zagreb, Faculty of Pharmacy and Biochemistry, Ulica Ante Kovacića 1, Zagreb, Croatia\n\n## 1. Introduction\n\n## 2. Glycosylated IgG holds a central role in immunity\n\nIgG... | sle-glyco-iggnew_001__zh_interleave | sle-glyco-iggnew_001 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Set the initial score to 0 points.
2. Score for 'Su... | {} | sle-glyco-iggnew | biology | zh | sle-glyco-iggnew_001 | [
"images/SciDocBench/24/245e2a2b06f8078a09c7504e10900ff1371db379c5f99c5bd9da5eec01c5e138.jpg",
"images/SciDocBench/c1/c1f4c11929f1562129dff9ccff2539df63b448e28dcc34b838dc75cff5071c2c.jpg",
"images/SciDocBench/4c/4c011fccb9c93e7a2344bf33e734f4902c72dcf10513070d84fc3a461a069238.jpg",
"images/SciDocBench/83/834e5... |
495 | 我附上了一篇物理论文。请阅读它，并将 Figure 1(b) 中所示的图结构——具体而言是该图中嵌入的 3-regular graph 插图——复现为一个 LaTeX/TikZ 图形。
要求：
- 该图有 N = 22 个节点，每个节点恰好连接到 3 条边（3-regular）
- 节点应渲染为填充圆，并带有可见的数字标签（1–22）
- 边应清晰地绘制为实线
- 节点布局应尽可能在视觉上与论文中的图相匹配
请输出一个完整的、可编译的 LaTeX 文档。 | \documentclass[border=12pt]{standalone}
\usepackage{tikz}
\begin{document}
\begin{tikzpicture}[
scale=1.0,
nd/.style={circle, draw=teal!60!black, fill=teal!70, minimum size=7mm,
inner sep=0pt, font=\small\bfseries, text=white, line width=0.8pt},
ed/.style={draw=black, line width=1.5pt, line cap=roun... | E1 | judge | [
"cc/ccd436a940892f67a3ec180bafd24b24ca4a03aa5561cffd43617c19135de4f0.jpg",
"83/837fd4a87d038a2e98dc23b26081174627f636bba6452b4c17142909db9cbbeb.jpg",
"db/dbffdf3530105f3678779b7937a6b92afc9ac59abd6fbe8e56805c4ac7c81fd1.jpg",
"38/38e82190d688cc355a267249ae33c15ca233b0771922b7bd832f9e19a8800d26.jpg",
"2c/2cef... | [{"text":"# Imaginary-time evolution of interacting spin systems in the truncated Wigner approximation\n\nTom Schlegel, $^{1}$ Dennis Breu, $^{1}$ and Michael Fleischhauer $^{1}$\n\n$^{1}$ Department of Physics and Research Center OPTIMAS,\n\nWe present a semiclassical phase-space method to calculate thermal and ground... | itwa-spin_002__zh_interleave | itwa-spin_002 | zh_interleave | interleave | text | You are an expert evaluator the model's response to a question. I will provide you with the ground-truth and the model's prediction. You need to score the model's prediction.
Ground-truth answer: {answer}
Model prediction: {prediction}
Scoring rules (Total=1.0):
1. Format Score (0.2 points): Check whether the predi... | {} | itwa-spin | physics | zh | itwa-spin_002 | [
"images/SciDocBench/cc/ccd436a940892f67a3ec180bafd24b24ca4a03aa5561cffd43617c19135de4f0.jpg",
"images/SciDocBench/83/837fd4a87d038a2e98dc23b26081174627f636bba6452b4c17142909db9cbbeb.jpg",
"images/SciDocBench/db/dbffdf3530105f3678779b7937a6b92afc9ac59abd6fbe8e56805c4ac7c81fd1.jpg",
"images/SciDocBench/38/38e82... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.