撰写 研究论文简历学位论文基金申请书研究论文 少些繁琐杂务。
Oleafly 是一个免费的桌面工作区,可用 LaTeX、Typst 和 Markdown 撰写论文。在同一处查找文献、管理引用,看着论文逐步成形。让 AI 助手修改某一节、制作插图或修复失败的构建,所有改动都可以先审阅。你的项目保存为你自己拥有的普通文件,编辑器可以搭配 AI 使用,也可以完全不用。
\[ E = h\nu \]\[ E = h\nu \]\[ \int_{-\infty}^{\infty} e^{-x^2}\,dx = \sqrt{\pi} \]\[ \Delta G = \Delta H - T\Delta S \]\[ \Delta G = \Delta H - T\Delta S \]\[ a^2 + b^2 = c^2 \]



+2
重新编译
results.tex 里有 3 位共同作者


+2
编译
in mammary epithelial cells
Figure 1: Growth factors reach ERK through one cascade but leave different temporal signatures. (a) Receptor tyrosine kinases converge on the RAS–RAF–MEK–ERK module. (b) Single-cell ERK activity from the kinase translocation reporter after stimulation at t = 0.
Cells read growth factors through a small number of shared kinases, yet they respond to EGF, HGF and IGF-1 with different fates. We imaged ERK activity in 2,184 MCF10A cells with a kinase translocation reporter and found that each ligand produces pulses at a characteristic frequency rather than a characteristic amplitude. Frequency alone identifies the ligand in 91 percent of cells.
Epidermal growth factor, hepatocyte growth factor and insulin-like growth factor all reach the nucleus through the same RAS–RAF–MEK–ERK module [1], yet epithelial cells answer them with migration, scattering or growth. How a single kinase carries that distinction has been a long-standing question. Early work in PC-12 cells tied fate to whether ERK activity was transient or sustained [2, 3].
Live-cell reporters changed the picture. In single cells ERK does not rise and settle; it fires in discrete pulses, and the pulse frequency tracks the growth factor dose [4, 5]. Here we ask whether frequency also encodes which growth factor a cell has seen.
We find that it does. Each ligand sets its own pulse rate, the rate survives changes in dose, and cells that keep pulsing are the ones that go on to divide. The same code is rescaled, not erased, in cells carrying an activating RAS mutation. Blocking the pulses with a MEK inhibitor at 60 minutes stopped division under all three ligands. Section 2 sets out these results, Section 3 discusses what they mean for signalling by timing, and Section 4 gives the imaging and decoding methods.
We stimulated MCF10A cells expressing the ERK kinase translocation reporter [6] with saturating doses of EGF, HGF or IGF-1 and tracked nuclear and cytoplasmic signal every two minutes in 2,184 cells for four hours. All three ligands produced the same first spike within ten minutes (Figure 1b). After that the responses diverged: EGF drove pulses about every 19 minutes, HGF about every 41, and IGF-1 rarely more than once an hour (Table 1).
Table 1: Pulse statistics between 15 and 120 minutes after stimulation. Medians across cells.
| Ligand | Dose | Pulses/h | Width | Cells |
| EGF | 20 | 3.1 | 7.8 | 812 |
| HGF | 50 | 1.5 | 8.4 | 694 |
| IGF-1 | 100 | 0.8 | 9.6 | 521 |
| Vehicle | – | 0.2 | 11.2 | 157 |
Pulse amplitude overlapped almost completely between ligands, so we trained a decoder on inter-pulse intervalsthe gaps between pulses alone. Given the first two hours of a single cell's trace, it named the ligand correctly in 91 percent of held-out cells, compared with 47 percent for a decoder that saw only amplitudes. Shuffling the intervals within each cell removed the signal, and so did pooling cells across doses.
A decoder trained on amplitudes alone did no better than chance for HGF against IGF-1, while interval-based decoding separated all three ligands after the first hour of imaging.
Cells that kept pulsing past the second hour entered S phase within 18 hours in 74 percent of cases, against 21 percent for cells whose pulses stopped early (Figure 2). Pulse count over the first four hours predicted entry better than mean ERK activity over the same window.
Frequency coding lets one kinase carry several messages without crosstalk at the level of amplitude.
Mutant RAS rescales this code rather than erasing it [7], which suggests the frequency channel survives oncogenic signalling and could be read by downstream targets that integrate over time.
MCF10A cells were imaged every two minutes for four hours on a widefield microscope with environmental control. Nuclear and cytoplasmic reporter signal was segmented automatically and checked by hand for a random subset of cells.
Lucas Moreau
编辑者
正在编辑 sections/methods.tex
Daniel Hart
编辑者
正在编辑 sections/results.tex
- 语言分析已分析当前项目版本
- 正在编译正在生成并验证 PDF 输出
- 正在渲染 PDF正在等待已验证的编译输出
Single-Fault LATEX Compilation Repair
Independent Researcher
San Jose, CA, USA
Abstract—Compilation repair for LaTeX is easy to score badly. A candidate may compile after deleting or hiding content, while exact source equality rejects valid alternative repairs. We present TeXFix-Bench, a controlled benchmark for repairing complete, broken LaTeX sources under a pinned engine. The generator applies 12 deterministic single-fault operators to 13 hand-authored seed documents. It retains 627 engine-validated rows, comprising 300 unique hard-track inputs and 67 unique warning-only inputs after hash deduplication.
Index Terms—automated program repair, LaTeX, benchmark, compilation, large language models, reproducibility
LaTeX compilation failures are a practical target for automated repair. A missing delimiter or misspelled control sequence can prevent a complete document from building, and the first reported diagnostic may be distant from the underlying fault. LLM-assisted writing systems now operate inside LaTeX editors , but their publications do not provide a public, engine-pinned evaluation of compilation-failure repair.
Program repair benchmarks pair a defect with an executable oracle and a fixed harness . TeXFix-Bench studies a narrower contract: a broken source document in, a repaired full source out, under a specified engine.
TexOCR-Bench evaluates page-image-to-LaTeX reconstruction , and EqFix repairs errors in individual equations by synthesizing rules from examples . Overfitting to a weak oracle is a known hazard .
A system receives one broken source file and returns one repaired source file. The engine, its version, and the build command are fixed for every attempt.
Every generated row passes an engine gate before it is retained, and duplicate broken sources are removed by SHA-256 so that no input is scored twice .
Compile Recovery requires a successful Tectonic 0.17.0 run and at least 0.80 token-level LCS similarity between text extracted from the candidate and reference PDFs. Normalized exact match sits beside it, because compilation alone cannot separate a faithful repair from a deletion .
Version 0.2 results on 86 distinct hard-track inputs. CR is intention-to-treat.
| System | CR | CR % | NEM | Fail |
| Identity | 0/86 | 0.0 | 0/86 | 0 |
| Rule | 34/86 | 39.5 | 28/86 | 0 |
| Qwen-7B + fb | 57/86 | 66.3 | 17/86 | 0 |
| MiniMax M3 | 75/86 | 87.2 | 50/86 | 2 |
| DeepSeek V4 | 81/86 | 94.2 | 38/86 | 2 |
| Gemini 3.5 | 82/86 | 95.3 | 26/86 | 0 |
| Claude S5 | 85/86 | 98.8 | 52/86 | 0 |
| GPT-5.6 Sol | 85/86 | 98.8 | 66/86 | 1 |
| Grok 4.5 | 85/86 | 98.8 | 59/86 | 0 |
Claude Sonnet 5, GPT-5.6 Sol, and Grok 4.5 each recover 85 of 86 inputs, yet their normalized exact-match counts are 52, 66, and 59 . Compilation and edit precision must be reported together.
The contribution is an auditable task definition, an executable oracle, a fixed cohort, and a replayable candidate ledger.
[1]R. Just, D. Jalali, and M. D. Ernst, “Defects4J: A database of existing faults to enable controlled testing studies for Java programs,” in Proc. ISSTA, 2014, pp. 437–440.
[2]C. Le Goues, N. Holtschulte, E. K. Smith, Y. Brun, P. Devanbu, S. Forrest, and W. Weimer, “The ManyBugs and IntroClass benchmarks for automated repair of C programs,” IEEE Trans. Softw. Eng., vol. 41, no. 12, pp. 1236–1256, 2015.
[3]C. E. Jimenez, J. Yang, A. Wettig, S. Yao, K. Pei, O. Press, and K. Narasimhan, “SWE-bench: Can language models resolve real-world GitHub issues?” in Proc. ICLR, 2024.
[4]M. Chen et al., “Evaluating large language models trained on code,” arXiv:2107.03374, 2021.
[5]E. K. Smith, E. T. Barr, C. Le Goues, and Y. Brun, “Is the cure worse than the disease? Overfitting in automated program repair,” in Proc. ESEC/FSE, 2015, pp. 532–543.
[6]R. A. DeMillo, R. J. Lipton, and F. G. Sayward, “Hints on test data selection: Help for the practicing programmer,” Computer, vol. 11, no. 4, pp. 34–41, 1978.
[7]Y. Jia and M. Harman, “An analysis and survey of the development of mutation testing,” IEEE Trans. Softw. Eng., vol. 37, no. 5, pp. 649–678, 2011.
[8]R. Gupta, S. Pal, A. Kanade, and S. Shevade, “DeepFix: Fixing common C language errors by deep learning,” in Proc. AAAI, vol. 31, no. 1, pp. 1345–1351, 2017.
[9]M. Yasunaga and P. Liang, “Graph-based, self-supervised program repair from diagnostic feedback,” in Proc. ICML, PMLR, vol. 119, pp. 10799–10808, 2020.
[10]F. Zhu, Z. Liu, G. Yang, J. Li, and S. Cheung, “EqFix: Fixing LaTeX equation errors by examples,” in SETTA 2022, LNCS 13649, pp. 89–109.
[11]K. Wang et al., “TexOCR-Bench: Benchmarking page-image-to-LaTeX reconstruction,” in Proc. 64th ACL, Vol. 1, 2026.
[12]H. Wen, Z. Zhang, and Y. Li, “OverleafCopilot: Empowering academic writing in Overleaf with large language models,” arXiv:2403.09733, 2024.
[13]S. Hou et al., “PaperDebugger: Assisting scientific writing, review, and editing,” in Companion Proc. ACM Web Conf., 2026.
[14]J. Tan, K. Zhang, and X. Chen, “Inconsistencies in TeX-produced documents,” in Proc. ISSTA, 2024.
[15]L. Zheng et al., “Judging LLM-as-a-judge with MT-Bench and Chatbot Arena,” Adv. Neural Inf. Process. Syst., vol. 36, pp. 46595–46623, 2023.
[16]A. Panickssery, S. R. Bowman, and S. Feng, “LLM evaluators recognize and favor their own generations,” Adv. Neural Inf. Process. Syst., vol. 37, 2024.
[17]B. Hui et al., “Qwen2.5-Coder technical report,” arXiv:2409.12186, 2024.
[18]Meta AI, “Llama 3.2 model card,” 2024.
[19]E. Kassis et al., “Agent skills for reproducible research agents,” arXiv:2609.00065, 2026.
ibitem{legoues2015manybugs}
main.tex:956
- 步数
- 1
- 输入
- 5,144
- 输出
- 467
- 运行次数
- 1
- 步数
- 3
- Token 数
- 39,035
费用是按公开模型定价得出的估算,并非账单金额。
使用你自己的 API 密钥(OpenAI、Anthropic、Groq 等),或通过 Ollama 在本地运行模型。助手可以读取和编辑文件、编译项目并校验 PDF。
连接提供商使用 Ollama 运行本地模型
你今天想做什么?正在处理“TeXFix-Bench: LaTeX Repair Bench...”
Comprehensive citation management for academic research. Search OpenAlex, PubMed, and Google Scholar for papers, extract accurate metadata, validate citations, and generate properly formatted BibTeX entries. This skill should be used when you need to find papers, verify citation information, convert DOIs to BibTeX, or ensure reference accuracy in scientific writing.
Perform bounded, local exploratory analysis of explicitly supported scientific files. Use for redacted CSV/TSV/JSON profiles; optional NumPy, HDF5, FASTA/FASTQ, and basic image metadata inspection; missingness/leakage audits; outlier and transformation sensitivity; and rigorous EDA report scaffolds. Other domain formats are reference-only and unknown formats fail closed.
Formulate evidence-bounded scientific questions, candidate hypotheses, rival explanations, causal or associational claims, discriminating predictions, measurements, and preregistration-ready analysis plans. Use when turning observations or preliminary findings into transparent, testable research plans without treating hypotheses as facts.
正在处理“TeXFix-Bench: LaTeX Repair Bench...”
Claude Code
能耗对比需要一个支撑来源。我在引言中找到了这条论断,但它引用的文献讲的是检测准确率。你的文件没有改动。
- 卫星火情探测
- 低功耗传感器网络
- TinyML 与量化推理
我只保留带有可核验 DOI 元数据的结果。
write_filereferences.bib第 1 行
references.bib第一处更改 · 第 1 行@article{open_science2015, title = {Estimating the Reproducibility of Psychological Science},@article{giglio2003, author = {Giglio, Louis and Descloitres, Jacques and Justice, Christopher O.}, title = {An Enhanced Contextual Fire Detection Algorithm for {MODIS}}, journal = {Remote Sensing of Environment}, year = {2003}, doi = {10.1016/S0034-4257(03)00184-6}}replace_in_filemain.tex第 21 行
main.tex第一处更改 · 第 21 行Program repair benchmarks pair a defect with an executable oracle and afixed harness~\cite{just2014defects4j,legoues2015manybug}. SWE-bench extendsfixed harness~\cite{just2014defects4j,legoues2015manybugs}. SWE-bench extendsthe contract to repository-scale issues~\cite{jimenez2024swebench}.TeXFix-Bench studies a narrower contract: a system receives an existingTeXFix-Bench studies a narrower contract~\cite{zhu2022eqfix}: a system receivesSingle-Fault LATEX Compilation Repair
Independent Researcher
San Jose, CA, USA
Abstract—Compilation repair for LaTeX is easy to score badly. A candidate may compile after deleting or hiding content, while exact source equality rejects valid alternative repairs. We present TeXFix-Bench, a controlled benchmark for repairing complete, broken LaTeX sources under a pinned engine. The generator applies 12 deterministic single-fault operators to 13 hand-authored seed documents. It retains 627 engine-validated rows, comprising 300 unique hard-track inputs and 67 unique warning-only inputs after hash deduplication.
Index Terms—automated program repair, LaTeX, benchmark, compilation, large language models, reproducibility
LaTeX compilation failures are a practical target for automated repair. A missing delimiter or misspelled control sequence can prevent a complete document from building, and the first reported diagnostic may be distant from the underlying fault. LLM-assisted writing systems now operate inside LaTeX editors , but their publications do not provide a public, engine-pinned evaluation of compilation-failure repair.
Program repair benchmarks pair a defect with an executable oracle and a fixed harness . TeXFix-Bench studies a narrower contract: a broken source document in, a repaired full source out, under a specified engine.
TexOCR-Bench evaluates page-image-to-LaTeX reconstruction , and EqFix repairs errors in individual equations by synthesizing rules from examples . Overfitting to a weak oracle is a known hazard .
A system receives one broken source file and returns one repaired source file. The engine, its version, and the build command are fixed for every attempt.
Every generated row passes an engine gate before it is retained, and duplicate broken sources are removed by SHA-256 so that no input is scored twice .
Compile Recovery requires a successful Tectonic 0.17.0 run and at least 0.80 token-level LCS similarity between text extracted from the candidate and reference PDFs. Normalized exact match sits beside it, because compilation alone cannot separate a faithful repair from a deletion .
Version 0.2 results on 86 distinct hard-track inputs. CR is intention-to-treat.
| System | CR | CR % | NEM | Fail |
| Identity | 0/86 | 0.0 | 0/86 | 0 |
| Rule | 34/86 | 39.5 | 28/86 | 0 |
| Qwen-7B + fb | 57/86 | 66.3 | 17/86 | 0 |
| MiniMax M3 | 75/86 | 87.2 | 50/86 | 2 |
| DeepSeek V4 | 81/86 | 94.2 | 38/86 | 2 |
| Gemini 3.5 | 82/86 | 95.3 | 26/86 | 0 |
| Claude S5 | 85/86 | 98.8 | 52/86 | 0 |
| GPT-5.6 Sol | 85/86 | 98.8 | 66/86 | 1 |
| Grok 4.5 | 85/86 | 98.8 | 59/86 | 0 |
Claude Sonnet 5, GPT-5.6 Sol, and Grok 4.5 each recover 85 of 86 inputs, yet their normalized exact-match counts are 52, 66, and 59 . Compilation and edit precision must be reported together.
The contribution is an auditable task definition, an executable oracle, a fixed cohort, and a replayable candidate ledger.
[1]R. Just, D. Jalali, and M. D. Ernst, “Defects4J: A database of existing faults to enable controlled testing studies for Java programs,” in Proc. ISSTA, 2014, pp. 437–440.
[2]C. Le Goues, N. Holtschulte, E. K. Smith, Y. Brun, P. Devanbu, S. Forrest, and W. Weimer, “The ManyBugs and IntroClass benchmarks for automated repair of C programs,” IEEE Trans. Softw. Eng., vol. 41, no. 12, pp. 1236–1256, 2015.
[3]C. E. Jimenez, J. Yang, A. Wettig, S. Yao, K. Pei, O. Press, and K. Narasimhan, “SWE-bench: Can language models resolve real-world GitHub issues?” in Proc. ICLR, 2024.
[4]M. Chen et al., “Evaluating large language models trained on code,” arXiv:2107.03374, 2021.
[5]E. K. Smith, E. T. Barr, C. Le Goues, and Y. Brun, “Is the cure worse than the disease? Overfitting in automated program repair,” in Proc. ESEC/FSE, 2015, pp. 532–543.
[6]R. A. DeMillo, R. J. Lipton, and F. G. Sayward, “Hints on test data selection: Help for the practicing programmer,” Computer, vol. 11, no. 4, pp. 34–41, 1978.
[7]Y. Jia and M. Harman, “An analysis and survey of the development of mutation testing,” IEEE Trans. Softw. Eng., vol. 37, no. 5, pp. 649–678, 2011.
[8]R. Gupta, S. Pal, A. Kanade, and S. Shevade, “DeepFix: Fixing common C language errors by deep learning,” in Proc. AAAI, vol. 31, no. 1, pp. 1345–1351, 2017.
[9]M. Yasunaga and P. Liang, “Graph-based, self-supervised program repair from diagnostic feedback,” in Proc. ICML, PMLR, vol. 119, pp. 10799–10808, 2020.
[10]F. Zhu, Z. Liu, G. Yang, J. Li, and S. Cheung, “EqFix: Fixing LaTeX equation errors by examples,” in SETTA 2022, LNCS 13649, pp. 89–109.
[11]K. Wang et al., “TexOCR-Bench: Benchmarking page-image-to-LaTeX reconstruction,” in Proc. 64th ACL, Vol. 1, 2026.
[12]H. Wen, Z. Zhang, and Y. Li, “OverleafCopilot: Empowering academic writing in Overleaf with large language models,” arXiv:2403.09733, 2024.
[13]S. Hou et al., “PaperDebugger: Assisting scientific writing, review, and editing,” in Companion Proc. ACM Web Conf., 2026.
[14]J. Tan, K. Zhang, and X. Chen, “Inconsistencies in TeX-produced documents,” in Proc. ISSTA, 2024.
[15]L. Zheng et al., “Judging LLM-as-a-judge with MT-Bench and Chatbot Arena,” Adv. Neural Inf. Process. Syst., vol. 36, pp. 46595–46623, 2023.
[16]A. Panickssery, S. R. Bowman, and S. Feng, “LLM evaluators recognize and favor their own generations,” Adv. Neural Inf. Process. Syst., vol. 37, 2024.
[17]B. Hui et al., “Qwen2.5-Coder technical report,” arXiv:2409.12186, 2024.
[18]Meta AI, “Llama 3.2 model card,” 2024.
[19]E. Kassis et al., “Agent skills for reproducible research agents,” arXiv:2609.00065, 2026.
@article{giglio2003, title = {An Enhanced Contextual Fire Detection Algorithm for MODIS}, year = {2003},}@article{schroeder2014, title = {The New VIIRS 375 m Active Fire Detection Data Product}, year = {2014},}@article{dampage2022, title = {Forest Fire Detection System Using Wireless Sensor Networks}, year = {2022},}@article{adumanu2018, title = {Energy-Harvesting Wireless Sensor Networks: A Review}, year = {2018},}@article{adelantado2017, title = {Understanding the Limits of LoRaWAN}, year = {2017},}@article{bai2018, title = {An Empirical Evaluation of Generic Convolutional Networks}, year = {2018},}@article{jacob2018, title = {Quantization and Training of Neural Networks}, year = {2018},}@article{ray2022, title = {A Review on TinyML}, year = {2022},}@article{geifman2017, title = {Selective Classification for Deep Neural Networks}, year = {2017},}在下方连接你使用的提供商。密钥仅存储在本地。保存后即设为默认。
你可以通过聊天面板的下拉菜单切换已配置的提供商和模型。
每当你开始新聊天时都会使用该模型。你仍可在聊天面板中为单个对话切换模型。
会作为你的个人风格和偏好添加到每个 AI 请求中。助手会在内置行为的基础上遵循这些内容,它们无法覆盖助手的工具或安全规则。
允许 AI 捕获 PDF 页面允许代理栅格化编译后的页面,以进行视觉版式检查。如果你不希望向提供商发送页面图像,请关闭此选项。
- 语言分析已分析当前项目版本
- 正在编译正在生成并验证 PDF 输出
- 正在渲染 PDF正在等待已验证的编译输出
Single-Fault LATEX Compilation Repair
Independent Researcher
San Jose, CA, USA
Abstract—Compilation repair for LaTeX is easy to score badly. A candidate may compile after deleting or hiding content, while exact source equality rejects valid alternative repairs. We present TeXFix-Bench, a controlled benchmark for repairing complete, broken LaTeX sources under a pinned engine. The generator applies 12 deterministic single-fault operators to 13 hand-authored seed documents. It retains 627 engine-validated rows, comprising 300 unique hard-track inputs and 67 unique warning-only inputs after hash deduplication.
Index Terms—automated program repair, LaTeX, benchmark, compilation, large language models, reproducibility
LaTeX compilation failures are a practical target for automated repair. A missing delimiter or misspelled control sequence can prevent a complete document from building, and the first reported diagnostic may be distant from the underlying fault. LLM-assisted writing systems now operate inside LaTeX editors , but their publications do not provide a public, engine-pinned evaluation of compilation-failure repair.
Program repair benchmarks pair a defect with an executable oracle and a fixed harness . TeXFix-Bench studies a narrower contract: a broken source document in, a repaired full source out, under a specified engine.
TexOCR-Bench evaluates page-image-to-LaTeX reconstruction , and EqFix repairs errors in individual equations by synthesizing rules from examples . Overfitting to a weak oracle is a known hazard .
A system receives one broken source file and returns one repaired source file. The engine, its version, and the build command are fixed for every attempt.
Every generated row passes an engine gate before it is retained, and duplicate broken sources are removed by SHA-256 so that no input is scored twice .
Compile Recovery requires a successful Tectonic 0.17.0 run and at least 0.80 token-level LCS similarity between text extracted from the candidate and reference PDFs. Normalized exact match sits beside it, because compilation alone cannot separate a faithful repair from a deletion .
Version 0.2 results on 86 distinct hard-track inputs. CR is intention-to-treat.
| System | CR | CR % | NEM | Fail |
| Identity | 0/86 | 0.0 | 0/86 | 0 |
| Rule | 34/86 | 39.5 | 28/86 | 0 |
| Qwen-7B + fb | 57/86 | 66.3 | 17/86 | 0 |
| MiniMax M3 | 75/86 | 87.2 | 50/86 | 2 |
| DeepSeek V4 | 81/86 | 94.2 | 38/86 | 2 |
| Gemini 3.5 | 82/86 | 95.3 | 26/86 | 0 |
| Claude S5 | 85/86 | 98.8 | 52/86 | 0 |
| GPT-5.6 Sol | 85/86 | 98.8 | 66/86 | 1 |
| Grok 4.5 | 85/86 | 98.8 | 59/86 | 0 |
Claude Sonnet 5, GPT-5.6 Sol, and Grok 4.5 each recover 85 of 86 inputs, yet their normalized exact-match counts are 52, 66, and 59 . Compilation and edit precision must be reported together.
The contribution is an auditable task definition, an executable oracle, a fixed cohort, and a replayable candidate ledger.
[1]R. Just, D. Jalali, and M. D. Ernst, “Defects4J: A database of existing faults to enable controlled testing studies for Java programs,” in Proc. ISSTA, 2014, pp. 437–440.
[2]C. Le Goues, N. Holtschulte, E. K. Smith, Y. Brun, P. Devanbu, S. Forrest, and W. Weimer, “The ManyBugs and IntroClass benchmarks for automated repair of C programs,” IEEE Trans. Softw. Eng., vol. 41, no. 12, pp. 1236–1256, 2015.
[3]C. E. Jimenez, J. Yang, A. Wettig, S. Yao, K. Pei, O. Press, and K. Narasimhan, “SWE-bench: Can language models resolve real-world GitHub issues?” in Proc. ICLR, 2024.
[4]M. Chen et al., “Evaluating large language models trained on code,” arXiv:2107.03374, 2021.
[5]E. K. Smith, E. T. Barr, C. Le Goues, and Y. Brun, “Is the cure worse than the disease? Overfitting in automated program repair,” in Proc. ESEC/FSE, 2015, pp. 532–543.
[6]R. A. DeMillo, R. J. Lipton, and F. G. Sayward, “Hints on test data selection: Help for the practicing programmer,” Computer, vol. 11, no. 4, pp. 34–41, 1978.
[7]Y. Jia and M. Harman, “An analysis and survey of the development of mutation testing,” IEEE Trans. Softw. Eng., vol. 37, no. 5, pp. 649–678, 2011.
[8]R. Gupta, S. Pal, A. Kanade, and S. Shevade, “DeepFix: Fixing common C language errors by deep learning,” in Proc. AAAI, vol. 31, no. 1, pp. 1345–1351, 2017.
[9]M. Yasunaga and P. Liang, “Graph-based, self-supervised program repair from diagnostic feedback,” in Proc. ICML, PMLR, vol. 119, pp. 10799–10808, 2020.
[10]F. Zhu, Z. Liu, G. Yang, J. Li, and S. Cheung, “EqFix: Fixing LaTeX equation errors by examples,” in SETTA 2022, LNCS 13649, pp. 89–109.
[11]K. Wang et al., “TexOCR-Bench: Benchmarking page-image-to-LaTeX reconstruction,” in Proc. 64th ACL, Vol. 1, 2026.
[12]H. Wen, Z. Zhang, and Y. Li, “OverleafCopilot: Empowering academic writing in Overleaf with large language models,” arXiv:2403.09733, 2024.
[13]S. Hou et al., “PaperDebugger: Assisting scientific writing, review, and editing,” in Companion Proc. ACM Web Conf., 2026.
[14]J. Tan, K. Zhang, and X. Chen, “Inconsistencies in TeX-produced documents,” in Proc. ISSTA, 2024.
[15]L. Zheng et al., “Judging LLM-as-a-judge with MT-Bench and Chatbot Arena,” Adv. Neural Inf. Process. Syst., vol. 36, pp. 46595–46623, 2023.
[16]A. Panickssery, S. R. Bowman, and S. Feng, “LLM evaluators recognize and favor their own generations,” Adv. Neural Inf. Process. Syst., vol. 37, 2024.
[17]B. Hui et al., “Qwen2.5-Coder technical report,” arXiv:2409.12186, 2024.
[18]Meta AI, “Llama 3.2 model card,” 2024.
[19]E. Kassis et al., “Agent skills for reproducible research agents,” arXiv:2609.00065, 2026.
成果先留在独立工作区,待你审阅并应用。
查找关于低功耗野火感知的论文。保存一份附有已验证 DOI 的阅读清单。
在你应用已审阅的文件之前,原始项目保持不变。
应用前预览每个选中的文件。
research/reading-list.md新增预览+ # Wildfire sensing: reading list + + ## Satellite fire detection + Giglio et al. (2003). An Enhanced Contextual Fire + Detection Algorithm for MODIS. + DOI: 10.1016/S0034-4257(03)00184-6 + + Use: background on satellite detection methods. + Check separately: claims about ground-sensor energy use.
Attention
\definecolor{c0F172A}{HTML}{0F172A}\definecolor{c1E293B}{HTML}{1E293B}\definecolor{cC7C7CC}{HTML}{C7C7CC}\definecolor{cCDEDD0}{HTML}{CDEDD0}\definecolor{cCFE8F8}{HTML}{CFE8F8}\definecolor{cDAD2F0}{HTML}{DAD2F0}\definecolor{cE9E9EC}{HTML}{E9E9EC}\definecolor{cEEF2C3}{HTML}{EEF2C3}\definecolor{cF9D7D9}{HTML}{F9D7D9}\definecolor{cFDE3C7}{HTML}{FDE3C7}\definecolor{cFFFFFF}{HTML}{FFFFFF}\begin{tikzpicture}[>={Triangle[length=0.313cm,width=0.313cm]}] \node (n1_mds37) at (6.25,-21.75) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1cm] {Inputs}; \node (n2_hnw3j) at (6.25,-20.075) [draw=c1E293B, line width=0.025cm, fill=cF9D7D9, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Input Embedding}; \node (n3_x9vjf) at (4.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {}; \node (n4_aa0b6) at (6.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {+}; \node (n5_lbeum) at (2.625,-19.375) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=3.5cm, minimum height=1cm] {Positional Encoding}; \node (n6_gr85r) at (6.25,-16.325) [draw=c1E293B, line width=0.025cm, fill=cFDE3C7, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Multi-Head Attention}; \node (n7_it3d0) at (6.25,-14.575) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm}; \node (n8_0ae9p) at (6.25,-12.575) [draw=c1E293B, line width=0.025cm, fill=cCFE8F8, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Feed Forward}; \node (n9_31jhg) at (6.25,-10.825) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm}; \node (n11_3mv5b) at (1.875,-13.575) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=1.25cm, minimum height=0.75cm] {N×}; \node (n12_y6hil) at (16.25,-21.75) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1cm] {Outputs (shifted right)}; \node (n13_oukbf) at (16.25,-20.075) [draw=c1E293B, line width=0.025cm, fill=cF9D7D9, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Output Embedding}; \node (n14_e6m10) at (16.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {+}; \node (n15_te7fn) at (18.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {}; \node (n16_otz7n) at (19.875,-19.375) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=3.5cm, minimum height=1cm] {Positional Encoding}; \node (n17_nw4kz) at (16.25,-16.325) [draw=c1E293B, line width=0.025cm, fill=cFDE3C7, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Masked Multi-Head Attention}; \node (n18_feep8) at (16.25,-14.575) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm}; \node (n19_9lw0g) at (16.25,-12.575) [draw=c1E293B, line width=0.025cm, fill=cFDE3C7, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Multi-Head Attention}; \node (n20_lz6h6) at (16.25,-10.825) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm}; \node (n21_iyi8d) at (16.25,-8.825) [draw=c1E293B, line width=0.025cm, fill=cCFE8F8, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Feed Forward}; \node (n22_hapki) at (16.25,-7.075) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm}; \node (n24_qwbl3) at (19.375,-11.825) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=1.25cm, minimum height=0.75cm] {N×}; \node (n25_isnp1) at (16.25,-4.8) [draw=c1E293B, line width=0.025cm, fill=cDAD2F0, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.1cm] {Linear}; \node (n26_ggail) at (16.25,-2.8) [draw=c1E293B, line width=0.025cm, fill=cCDEDD0, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.1cm] {Softmax}; \node (n27_76g5v) at (16.25,-0.875) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.5cm, minimum height=1.25cm] {Output Probabilities}; \begin{scope}[on background layer] \node (n10_ma9el) at (6.25,-13.575) [draw=cC7C7CC, line width=0.025cm, fill=cE9E9EC, rounded corners=0.4cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=5.5cm, minimum height=7.65cm] {}; \node (n23_egiii) at (16.25,-11.7) [draw=cC7C7CC, line width=0.025cm, fill=cE9E9EC, rounded corners=0.4cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=5.5cm, minimum height=11.4cm] {}; \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-21.25) -- (6.25,-20.75) -- (6.25,-21.15) -- (6.25,-20.65); \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-19.5) -- (6.25,-19) -- (6.25,-19.1) -- (6.25,-18.6); \draw[->, line width=0.05cm] (n3_x9vjf.east) to[out=0, in=180] (n4_aa0b6.west); \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-17.75) -- (6.25,-17.25) -- (6.25,-17.4) -- (6.25,-16.9); \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-15.75) -- (6.25,-15.25) -- (6.25,-15.65) -- (6.25,-15.15); \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-14) -- (6.25,-13.5) -- (6.25,-13.65) -- (6.25,-13.15); \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-12) -- (6.25,-11.5) -- (6.25,-11.9) -- (6.25,-11.4); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-21.25) -- (16.25,-20.75) -- (16.25,-21.15) -- (16.25,-20.65); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-19.5) -- (16.25,-19) -- (16.25,-19.1) -- (16.25,-18.6); \draw[->, line width=0.05cm] (n15_te7fn.west) to[out=180, in=0] (n14_e6m10.east); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-17.75) -- (16.25,-17.25) -- (16.25,-17.4) -- (16.25,-16.9); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-15.75) -- (16.25,-15.25) -- (16.25,-15.65) -- (16.25,-15.15); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-14) -- (16.25,-13.5) -- (16.25,-13.65) -- (16.25,-13.15); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-12) -- (16.25,-11.5) -- (16.25,-11.9) -- (16.25,-11.4); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-10.25) -- (16.25,-9.75) -- (16.25,-9.9) -- (16.25,-9.4); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-8.25) -- (16.25,-7.75) -- (16.25,-8.15) -- (16.25,-7.65); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-6.5) -- (16.25,-6) -- (16.25,-5.925) -- (16.25,-5.85) -- (16.25,-5.35); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-4.25) -- (16.25,-3.75) -- (16.25,-3.85) -- (16.25,-3.35); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-2.25) -- (16.25,-1.75) -- (16.25,-2) -- (16.25,-1.5); \draw[->, line width=0.05cm, rounded corners=0.125cm] (8.375,-10.825) -- (8.875,-10.825) -- (11.25,-10.825) -- (11.25,-12.575) -- (13.625,-12.575) -- (14.125,-12.575); \draw[->, line width=0.05cm, rounded corners=0.125cm] (5.825,-18.175) -- (5.325,-18.175) -- (3.625,-18.175) -- (3.625,-14.575) -- (4.125,-14.575); \draw[->, line width=0.05cm, rounded corners=0.125cm] (4.125,-14.575) -- (4.1,-14.575) -- (3.625,-14.575) -- (3.625,-10.825) -- (4.125,-10.825); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.675,-18.175) -- (17.175,-18.175) -- (18.875,-18.175) -- (18.875,-14.575) -- (18.375,-14.575); \draw[->, line width=0.05cm, rounded corners=0.125cm] (18.375,-14.575) -- (18.875,-14.575) -- (18.875,-10.825) -- (18.4,-10.825) -- (18.375,-10.825); \draw[->, line width=0.05cm, rounded corners=0.125cm] (18.375,-10.825) -- (18.875,-10.825) -- (18.875,-7.075) -- (18.4,-7.075) -- (18.375,-7.075); \end{scope}\end{tikzpicture}Attention
\definecolor{c0F172A}{HTML}{0F172A}\definecolor{c1E293B}{HTML}{1E293B}\definecolor{cC7C7CC}{HTML}{C7C7CC}\definecolor{cCDEDD0}{HTML}{CDEDD0}\definecolor{cCFE8F8}{HTML}{CFE8F8}\definecolor{cDAD2F0}{HTML}{DAD2F0}\definecolor{cE9E9EC}{HTML}{E9E9EC}\definecolor{cEEF2C3}{HTML}{EEF2C3}\definecolor{cF9D7D9}{HTML}{F9D7D9}\definecolor{cFDE3C7}{HTML}{FDE3C7}\definecolor{cFFFFFF}{HTML}{FFFFFF}\begin{tikzpicture}[>={Triangle[length=0.313cm,width=0.313cm]}] \node (n1_mds37) at (6.25,-21.75) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1cm] {Inputs}; \node (n2_hnw3j) at (6.25,-20.075) [draw=c1E293B, line width=0.025cm, fill=cF9D7D9, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Input Embedding}; \node (n3_x9vjf) at (4.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {}; \node (n4_aa0b6) at (6.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {+}; \node (n5_lbeum) at (2.625,-19.375) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=3.5cm, minimum height=1cm] {Positional Encoding}; \node (n6_gr85r) at (6.25,-16.325) [draw=c1E293B, line width=0.025cm, fill=cFDE3C7, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Multi-Head Attention}; \node (n7_it3d0) at (6.25,-14.575) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm}; \node (n8_0ae9p) at (6.25,-12.575) [draw=c1E293B, line width=0.025cm, fill=cCFE8F8, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Feed Forward}; \node (n9_31jhg) at (6.25,-10.825) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm}; \node (n11_3mv5b) at (1.875,-13.575) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=1.25cm, minimum height=0.75cm] {N×}; \node (n12_y6hil) at (16.25,-21.75) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1cm] {Outputs (shifted right)}; \node (n13_oukbf) at (16.25,-20.075) [draw=c1E293B, line width=0.025cm, fill=cF9D7D9, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Output Embedding}; \node (n14_e6m10) at (16.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {+}; \node (n15_te7fn) at (18.25,-18.175) [circle, draw=c1E293B, line width=0.025cm, fill=cFFFFFF, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=0.85cm, minimum height=0.85cm] {}; \node (n16_otz7n) at (19.875,-19.375) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=3.5cm, minimum height=1cm] {Positional Encoding}; \node (n17_nw4kz) at (16.25,-16.325) [draw=c1E293B, line width=0.025cm, fill=cFDE3C7, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Masked Multi-Head Attention}; \node (n18_feep8) at (16.25,-14.575) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm}; \node (n19_9lw0g) at (16.25,-12.575) [draw=c1E293B, line width=0.025cm, fill=cFDE3C7, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Multi-Head Attention}; \node (n20_lz6h6) at (16.25,-10.825) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm}; \node (n21_iyi8d) at (16.25,-8.825) [draw=c1E293B, line width=0.025cm, fill=cCFE8F8, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Feed Forward}; \node (n22_hapki) at (16.25,-7.075) [draw=c1E293B, line width=0.025cm, fill=cEEF2C3, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.15cm] {Add \& Norm}; \node (n24_qwbl3) at (19.375,-11.825) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=1.25cm, minimum height=0.75cm] {N×}; \node (n25_isnp1) at (16.25,-4.8) [draw=c1E293B, line width=0.025cm, fill=cDAD2F0, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.1cm] {Linear}; \node (n26_ggail) at (16.25,-2.8) [draw=c1E293B, line width=0.025cm, fill=cCDEDD0, rounded corners=0.1cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.25cm, minimum height=1.1cm] {Softmax}; \node (n27_76g5v) at (16.25,-0.875) [text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=4.5cm, minimum height=1.25cm] {Output Probabilities}; \begin{scope}[on background layer] \node (n10_ma9el) at (6.25,-13.575) [draw=cC7C7CC, line width=0.025cm, fill=cE9E9EC, rounded corners=0.4cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=5.5cm, minimum height=7.65cm] {}; \node (n23_egiii) at (16.25,-11.7) [draw=cC7C7CC, line width=0.025cm, fill=cE9E9EC, rounded corners=0.4cm, text=c0F172A, font=\rmfamily\fontsize{10}{12}\selectfont, inner sep=0pt, outer sep=0pt, minimum width=5.5cm, minimum height=11.4cm] {}; \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-21.25) -- (6.25,-20.75) -- (6.25,-21.15) -- (6.25,-20.65); \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-19.5) -- (6.25,-19) -- (6.25,-19.1) -- (6.25,-18.6); \draw[->, line width=0.05cm] (n3_x9vjf.east) to[out=0, in=180] (n4_aa0b6.west); \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-17.75) -- (6.25,-17.25) -- (6.25,-17.4) -- (6.25,-16.9); \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-15.75) -- (6.25,-15.25) -- (6.25,-15.65) -- (6.25,-15.15); \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-14) -- (6.25,-13.5) -- (6.25,-13.65) -- (6.25,-13.15); \draw[->, line width=0.05cm, rounded corners=0.125cm] (6.25,-12) -- (6.25,-11.5) -- (6.25,-11.9) -- (6.25,-11.4); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-21.25) -- (16.25,-20.75) -- (16.25,-21.15) -- (16.25,-20.65); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-19.5) -- (16.25,-19) -- (16.25,-19.1) -- (16.25,-18.6); \draw[->, line width=0.05cm] (n15_te7fn.west) to[out=180, in=0] (n14_e6m10.east); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-17.75) -- (16.25,-17.25) -- (16.25,-17.4) -- (16.25,-16.9); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-15.75) -- (16.25,-15.25) -- (16.25,-15.65) -- (16.25,-15.15); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-14) -- (16.25,-13.5) -- (16.25,-13.65) -- (16.25,-13.15); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-12) -- (16.25,-11.5) -- (16.25,-11.9) -- (16.25,-11.4); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-10.25) -- (16.25,-9.75) -- (16.25,-9.9) -- (16.25,-9.4); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-8.25) -- (16.25,-7.75) -- (16.25,-8.15) -- (16.25,-7.65); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-6.5) -- (16.25,-6) -- (16.25,-5.925) -- (16.25,-5.85) -- (16.25,-5.35); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-4.25) -- (16.25,-3.75) -- (16.25,-3.85) -- (16.25,-3.35); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.25,-2.25) -- (16.25,-1.75) -- (16.25,-2) -- (16.25,-1.5); \draw[->, line width=0.05cm, rounded corners=0.125cm] (8.375,-10.825) -- (8.875,-10.825) -- (11.25,-10.825) -- (11.25,-12.575) -- (13.625,-12.575) -- (14.125,-12.575); \draw[->, line width=0.05cm, rounded corners=0.125cm] (5.825,-18.175) -- (5.325,-18.175) -- (3.625,-18.175) -- (3.625,-14.575) -- (4.125,-14.575); \draw[->, line width=0.05cm, rounded corners=0.125cm] (4.125,-14.575) -- (4.1,-14.575) -- (3.625,-14.575) -- (3.625,-10.825) -- (4.125,-10.825); \draw[->, line width=0.05cm, rounded corners=0.125cm] (16.675,-18.175) -- (17.175,-18.175) -- (18.875,-18.175) -- (18.875,-14.575) -- (18.375,-14.575); \draw[->, line width=0.05cm, rounded corners=0.125cm] (18.375,-14.575) -- (18.875,-14.575) -- (18.875,-10.825) -- (18.4,-10.825) -- (18.375,-10.825); \draw[->, line width=0.05cm, rounded corners=0.125cm] (18.375,-10.825) -- (18.875,-10.825) -- (18.875,-7.075) -- (18.4,-7.075) -- (18.375,-7.075); \end{scope}\end{tikzpicture}Attention
Attention
@article{venkateshmurthy2026etb, author = {Venkateshmurthy, Prajwal S.}, title = {Engine-Transfer-Bench: An Evidence-Based Benchmark for Document Compilation Engine Selection}, journal = {arXiv preprint}, year = 2026, eprint = {2608.18329}, }
LaTeX (latexmk) · main.tex
9/13/2026, 4:57:36 PM
7 个文件 · 183.8 KB · ded1c2ae
LaTeX (latexmk) · main.tex
9/13/2026, 3:52:03 PM
7 个文件 · 183.8 KB · 157c1ffb
LaTeX (latexmk) · main.tex
9/6/2026, 4:22:40 PM
7 个文件 · 183.7 KB · 891b7ee7
筛选条件会立即应用到项目书架
Senior Software Engineer (L5) with 9+ years designing and operating large-scale distributed systems. Deep experience in backend infrastructure, performance, and reliability.
- Tech lead for a Search serving component handling 2M+ queries/second; cut p99 latency 38%.
- Designed a globally replicated feature store on Spanner backing 40+ ML models.
- Led migration of a 300-service fleet to a new RPC framework; reduced on-call pages 45%.
- Mentored six engineers and ran the team’s design-review program.
- Built a real-time aggregation pipeline processing 8B events/day.
- Cut batch-job cost 30% with sharding and incremental computation.
- Shipped ledger and reconciliation services processing $60B+/year.
- Reduced a reconciliation run from six hours to 25 minutes.
Go, Raft, gRPC — distributed rate limiter doing 4M+ decisions/sec/node.
Rust — typed query builder for Postgres with compile-time-checked SQL.
Languages: C++, Go, Rust, Python, Java, SQL
Systems: Spanner, Bigtable, Kubernetes, Kafka, gRPC / Protobuf, Redis
选择这份手稿从何开始。每个部分都由你掌控,随时可改。

一个研究文件夹,含章节、资料与审阅笔记,以及一个框定研究问题的首个任务。

继续你已有的手稿:压缩包、Word 或 Markdown 草稿,或 GitHub 仓库。

从现成版式开始:期刊论文、学位论文、报告、海报或演讲。
Data scientist focused on trustworthy machine learning, experimentation, and decision systems used by product and research teams.
- Built evaluation systems for large language models across safety and product quality.
- Designed experiments used to guide high-impact model launches.
- Developed ranking models and reliable offline evaluation pipelines.
- Partnered with product teams to improve marketplace search quality.
- Created retention models and executive experimentation dashboards.
Open evaluation templates for production ML systems.
Python package for reproducible causal inference workflows.
Languages: Python, R, SQL, PyTorch, JAX
Systems: Spark, dbt, BigQuery, Airflow, MLflow, Kubernetes
Data scientist focused on trustworthy machine learning, experimentation, and decision systems used by product and research teams.
- Built evaluation systems for large language models across safety and product quality.
- Designed experiments used to guide high-impact model launches.
- Developed ranking models and reliable offline evaluation pipelines.
- Partnered with product teams to improve marketplace search quality.
- Created retention models and executive experimentation dashboards.
Open evaluation templates for production ML systems.
Python package for reproducible causal inference workflows.
Languages: Python, R, SQL, PyTorch, JAX
Systems: Spark, dbt, BigQuery, Airflow, MLflow, Kubernetes
Senior Software Engineer (L5) with 9+ years designing and operating large-scale distributed systems. Deep experience in backend infrastructure, performance, and reliability.
- Tech lead for a Search serving component handling 2M+ queries/second; cut p99 latency 38%.
- Designed a globally replicated feature store on Spanner backing 40+ ML models.
- Led migration of a 300-service fleet to a new RPC framework; reduced on-call pages 45%.
- Mentored six engineers and ran the team’s design-review program.
- Built a real-time aggregation pipeline processing 8B events/day.
- Cut batch-job cost 30% with sharding and incremental computation.
- Shipped ledger and reconciliation services processing $60B+/year.
- Reduced a reconciliation run from six hours to 25 minutes.
Go, Raft, gRPC — distributed rate limiter doing 4M+ decisions/sec/node.
Rust — typed query builder for Postgres with compile-time-checked SQL.
Languages: C++, Go, Rust, Python, Java, SQL
Systems: Spanner, Bigtable, Kubernetes, Kafka, gRPC / Protobuf, Redis
Product engineer bridging design systems and frontend infrastructure to build accessible, high-craft tools for creative teams.
- Led architecture for collaborative prototyping workflows used by millions.
- Created an accessible component platform shared across six product teams.
- Shipped core editor interactions and improved canvas rendering performance 42%.
- Partnered with design to establish product-wide interaction standards.
- Prototyped service experiences for health and education clients.
Open-source accessible color and typography toolkit.
Patterns for resilient collaborative interfaces.
Languages: TypeScript, React, Rust, Swift, Figma
Systems: WebGL, Canvas, design systems, accessibility, prototyping
Product engineer bridging design systems and frontend infrastructure to build accessible, high-craft tools for creative teams.
- Led architecture for collaborative prototyping workflows used by millions.
- Created an accessible component platform shared across six product teams.
- Shipped core editor interactions and improved canvas rendering performance 42%.
- Partnered with design to establish product-wide interaction standards.
- Prototyped service experiences for health and education clients.
Open-source accessible color and typography toolkit.
Patterns for resilient collaborative interfaces.
Languages: TypeScript, React, Rust, Swift, Figma
Systems: WebGL, Canvas, design systems, accessibility, prototyping
Research engineer building fast, expressive systems for scientific communication and reproducible technical publishing.
- Built incremental layout and diagnostics for a modern typesetting engine.
- Improved large-document render latency and editor feedback loops.
- Worked on text shaping, browser layout, and developer tooling.
- Published work on interactive document systems.
A small language server for technical documents.
Experiments in high-performance text layout.
Languages: Rust, C++, TypeScript, Python
Systems: Compilers, text layout, language servers, WebAssembly
\documentclass[11pt,letterpaper]{article}\usepackage[T1]{fontenc}\usepackage[margin=0.5in]{geometry}\usepackage{titlesec}\usepackage{enumitem}\usepackage{hyperref}% ATS-friendly: single column, real selectable text, no tables-for-layout,% embedded subset fonts. Linear reading order so parsers extract cleanly.\setlength{\parindent}{0pt}\pagenumbering{gobble}\hypersetup{colorlinks=true, urlcolor=black, linkcolor=black}\titleformat{\section}{\large\bfseries\uppercase}{}{0em}{}[\vspace{2pt}\titlerule]\titlespacing*{\section}{0pt}{8pt}{4pt}\setlist[itemize]{leftmargin=14pt, itemsep=1pt, topsep=2pt, parsep=0pt}% role{Title}{Company}{Location}{Dates}\newcommand{\role}[4]{% \textbf{#1} \hfill #2 \\ \textit{#3} \hfill \textit{#4}%}\begin{document}\begin{center} {\Huge \textbf{Alex Chen}}\\[3pt] \small Senior Software Engineer\\[2pt] \href{mailto:alex.chen@gmail.com}{alex.chen@gmail.com}\,$\cdot$\, (650) 555-0142\,$\cdot$\, Mountain View, CA\,$\cdot$\, \href{https://alexchen.dev}{alexchen.dev}\,$\cdot$\, \href{https://github.com/alexchen}{github.com/alexchen}\,$\cdot$\, \href{https://linkedin.com/in/alexchen}{linkedin.com/in/alexchen}\end{center}\vspace{2pt}\section*{Summary}Senior Software Engineer (L5) with 9+ years designing and operating large-scaledistributed systems. Deep experience in backend infrastructure, performance, andreliability, with a track record of shipping products used by hundreds ofmillions of users and mentoring engineers toward senior roles.\section*{Experience}\role{Senior Software Engineer (L5)}{Google}{Mountain View, CA}{Mar 2020 -- Present}\begin{itemize} \item Tech lead for a Search serving component handling 2M+ queries/second; drove a redesign that cut p99 latency 38\% and saved an estimated \$14M/year in compute. \item Designed and launched a globally-replicated feature store on Spanner backing 40+ ML models, adopted by 12 teams across Search and Ads. \item Led migration of a 300-service fleet to a new RPC framework, improving tail latency and reducing on-call pages by 45\%. \item Mentored 6 engineers (2 promoted to senior); ran the team's design-review and readability programs.\end{itemize}\vspace{3pt}\role{Software Engineer (L4)}{Google}{Mountain View, CA}{Jul 2017 -- Mar 2020}\begin{itemize} \item Built a real-time aggregation pipeline (C++, Flume) processing 8B events/day for a Search-quality dashboard used org-wide. \item Cut batch-job cost 30\% by reworking sharding and introducing incremental recomputation.\end{itemize}\vspace{3pt}\role{Software Engineer}{Stripe}{San Francisco, CA}{Aug 2015 -- Jun 2017}\begin{itemize} \item Shipped core ledger and reconciliation services for a payments platform processing \$60B+/year. \item Reduced a reconciliation run from 6 hours to 25 minutes via parallelization and a columnar store.\end{itemize}\section*{Selected Projects}\textbf{Ratel} \hfill \href{https://github.com/alexchen/ratel}{github.com/alexchen/ratel} \\\textit{Go, Raft, gRPC} --- open-source distributed rate limiter doing 4M+ decisions/sec/node; 1.5k GitHub stars.\par\vspace{4pt}\textbf{tql} \hfill \href{https://github.com/alexchen/tql}{github.com/alexchen/tql} \\\textit{Rust} --- a typed query builder for Postgres with compile-time-checked SQL.\section*{Education}\textbf{M.S. Computer Science} \hfill Stanford University \\\textit{2013 -- 2015} \hfill \textit{Focus: Distributed Systems}\par\vspace{4pt}\textbf{B.S. Computer Science} \hfill University of Illinois Urbana-Champaign \\\textit{2009 -- 2013} \hfill \textit{GPA: 3.9 / 4.0}\section*{Skills}\textbf{Languages:} C++, Go, Rust, Python, Java, SQL \\\textbf{Systems:} Spanner, Bigtable, Kubernetes, Kafka, gRPC / Protobuf, Redis \\\textbf{Focus:} Distributed systems, performance, reliability (SLOs), system design, mentoring\end{document}Senior Software Engineer (L5) with 9+ years designing and operating large-scale distributed systems. Deep experience in backend infrastructure, performance, and reliability.
- Tech lead for a Search serving component handling 2M+ queries/second; cut p99 latency 38%.
- Designed a globally replicated feature store on Spanner backing 40+ ML models.
- Led migration of a 300-service fleet to a new RPC framework; reduced on-call pages 45%.
- Mentored six engineers and ran the team’s design-review program.
- Built a real-time aggregation pipeline processing 8B events/day.
- Cut batch-job cost 30% with sharding and incremental computation.
- Shipped ledger and reconciliation services processing $60B+/year.
- Reduced a reconciliation run from six hours to 25 minutes.
Go, Raft, gRPC — distributed rate limiter doing 4M+ decisions/sec/node.
Rust — typed query builder for Postgres with compile-time-checked SQL.
Languages: C++, Go, Rust, Python, Java, SQL
Systems: Spanner, Bigtable, Kubernetes, Kafka, gRPC / Protobuf, Redis
Results
Claude Sonnet 5, GPT-5.6 Sol, and Grok 4.5 each recover 85 of 86 inputs, yet their normalized exact-match counts are 52, 66, and 59 just2014defects4j.
Version 0.2 results on 86 distinct hard-track inputs. CR is intention-to-treat.
| System | CR | CR % | NEM | Fail |
| Identity | 0/86 | 0.0 | 0/86 | 0 |
| Rule | 34/86 | 39.5 | 28/86 | 0 |
| Qwen-7B + fb | 57/86 | 66.3 | 17/86 | 0 |
| MiniMax M3 | 75/86 | 87.2 | 50/86 | 2 |
| DeepSeek V4 | 81/86 | 94.2 | 38/86 | 2 |
| Gemini 3.5 | 82/86 | 95.3 | 26/86 | 0 |
| Claude S5 | 85/86 | 98.8 | 52/86 | 0 |
| GPT-5.6 Sol | 85/86 | 98.8 | 66/86 | 1 |
| Grok 4.5 | 85/86 | 98.8 | 59/86 | 0 |
Evaluation method
Compile Recovery requires a successful Tectonic 0.17.0 run and at least 0.80 token-level LCS similarity between text extracted from the candidate and reference PDFs. Normalized exact match sits beside it, because compilation alone cannot separate a faithful repair from a deletion.
Discussion and limitations
These results show why compilation and edit precision should be reported together, but they do not establish a universal model ranking jimenez2024swebench.
Every generated row passes an engine gate before it is retained, and duplicate broken sources are removed by SHA-256 so that no input is scored twice.
Conclusion
The contribution is an auditable task definition, an executable oracle, a fixed cohort, and a replayable candidate ledger.
- 语言分析已分析当前项目版本
- 正在编译正在生成并验证 PDF 输出
- 正在渲染 PDF正在等待已验证的编译输出
Single-Fault LATEX Compilation Repair
Independent Researcher
San Jose, CA, USA
Abstract—Compilation repair for LaTeX is easy to score badly. A candidate may compile after deleting or hiding content, while exact source equality rejects valid alternative repairs. We present TeXFix-Bench, a controlled benchmark for repairing complete, broken LaTeX sources under a pinned engine. The generator applies 12 deterministic single-fault operators to 13 hand-authored seed documents. It retains 627 engine-validated rows, comprising 300 unique hard-track inputs and 67 unique warning-only inputs after hash deduplication.
Index Terms—automated program repair, LaTeX, benchmark, compilation, large language models, reproducibility
LaTeX compilation failures are a practical target for automated repair. A missing delimiter or misspelled control sequence can prevent a complete document from building, and the first reported diagnostic may be distant from the underlying fault. LLM-assisted writing systems now operate inside LaTeX editors , but their publications do not provide a public, engine-pinned evaluation of compilation-failure repair.
Program repair benchmarks pair a defect with an executable oracle and a fixed harness . TeXFix-Bench studies a narrower contract: a broken source document in, a repaired full source out, under a specified engine.
TexOCR-Bench evaluates page-image-to-LaTeX reconstruction , and EqFix repairs errors in individual equations by synthesizing rules from examples . Overfitting to a weak oracle is a known hazard .
A system receives one broken source file and returns one repaired source file. The engine, its version, and the build command are fixed for every attempt.
Every generated row passes an engine gate before it is retained, and duplicate broken sources are removed by SHA-256 so that no input is scored twice .
Compile Recovery requires a successful Tectonic 0.17.0 run and at least 0.80 token-level LCS similarity between text extracted from the candidate and reference PDFs. Normalized exact match sits beside it, because compilation alone cannot separate a faithful repair from a deletion .
Version 0.2 results on 86 distinct hard-track inputs. CR is intention-to-treat.
| System | CR | CR % | NEM | Fail |
| Identity | 0/86 | 0.0 | 0/86 | 0 |
| Rule | 34/86 | 39.5 | 28/86 | 0 |
| Qwen-7B + fb | 57/86 | 66.3 | 17/86 | 0 |
| MiniMax M3 | 75/86 | 87.2 | 50/86 | 2 |
| DeepSeek V4 | 81/86 | 94.2 | 38/86 | 2 |
| Gemini 3.5 | 82/86 | 95.3 | 26/86 | 0 |
| Claude S5 | 85/86 | 98.8 | 52/86 | 0 |
| GPT-5.6 Sol | 85/86 | 98.8 | 66/86 | 1 |
| Grok 4.5 | 85/86 | 98.8 | 59/86 | 0 |
Claude Sonnet 5, GPT-5.6 Sol, and Grok 4.5 each recover 85 of 86 inputs, yet their normalized exact-match counts are 52, 66, and 59 . Compilation and edit precision must be reported together.
The contribution is an auditable task definition, an executable oracle, a fixed cohort, and a replayable candidate ledger.
[1]R. Just, D. Jalali, and M. D. Ernst, “Defects4J: A database of existing faults to enable controlled testing studies for Java programs,” in Proc. ISSTA, 2014, pp. 437–440.
[2]C. Le Goues, N. Holtschulte, E. K. Smith, Y. Brun, P. Devanbu, S. Forrest, and W. Weimer, “The ManyBugs and IntroClass benchmarks for automated repair of C programs,” IEEE Trans. Softw. Eng., vol. 41, no. 12, pp. 1236–1256, 2015.
[3]C. E. Jimenez, J. Yang, A. Wettig, S. Yao, K. Pei, O. Press, and K. Narasimhan, “SWE-bench: Can language models resolve real-world GitHub issues?” in Proc. ICLR, 2024.
[4]M. Chen et al., “Evaluating large language models trained on code,” arXiv:2107.03374, 2021.
[5]E. K. Smith, E. T. Barr, C. Le Goues, and Y. Brun, “Is the cure worse than the disease? Overfitting in automated program repair,” in Proc. ESEC/FSE, 2015, pp. 532–543.
[6]R. A. DeMillo, R. J. Lipton, and F. G. Sayward, “Hints on test data selection: Help for the practicing programmer,” Computer, vol. 11, no. 4, pp. 34–41, 1978.
[7]Y. Jia and M. Harman, “An analysis and survey of the development of mutation testing,” IEEE Trans. Softw. Eng., vol. 37, no. 5, pp. 649–678, 2011.
[8]R. Gupta, S. Pal, A. Kanade, and S. Shevade, “DeepFix: Fixing common C language errors by deep learning,” in Proc. AAAI, vol. 31, no. 1, pp. 1345–1351, 2017.
[9]M. Yasunaga and P. Liang, “Graph-based, self-supervised program repair from diagnostic feedback,” in Proc. ICML, PMLR, vol. 119, pp. 10799–10808, 2020.
[10]F. Zhu, Z. Liu, G. Yang, J. Li, and S. Cheung, “EqFix: Fixing LaTeX equation errors by examples,” in SETTA 2022, LNCS 13649, pp. 89–109.
[11]K. Wang et al., “TexOCR-Bench: Benchmarking page-image-to-LaTeX reconstruction,” in Proc. 64th ACL, Vol. 1, 2026.
[12]H. Wen, Z. Zhang, and Y. Li, “OverleafCopilot: Empowering academic writing in Overleaf with large language models,” arXiv:2403.09733, 2024.
[13]S. Hou et al., “PaperDebugger: Assisting scientific writing, review, and editing,” in Companion Proc. ACM Web Conf., 2026.
[14]J. Tan, K. Zhang, and X. Chen, “Inconsistencies in TeX-produced documents,” in Proc. ISSTA, 2024.
[15]L. Zheng et al., “Judging LLM-as-a-judge with MT-Bench and Chatbot Arena,” Adv. Neural Inf. Process. Syst., vol. 36, pp. 46595–46623, 2023.
[16]A. Panickssery, S. R. Bowman, and S. Feng, “LLM evaluators recognize and favor their own generations,” Adv. Neural Inf. Process. Syst., vol. 37, 2024.
[17]B. Hui et al., “Qwen2.5-Coder technical report,” arXiv:2409.12186, 2024.
[18]Meta AI, “Llama 3.2 model card,” 2024.
[19]E. Kassis et al., “Agent skills for reproducible research agents,” arXiv:2609.00065, 2026.
本版本新增
在同一个文件里,一起写论文。
与共同作者共享项目后,你会看到他们输入时光标在移动,每个光标都标着姓名。评论就在所讨论的那一行旁边。火车上 Wi-Fi 断了也能继续写,重新连接后你的编辑会同步过去。
如何搭建服务器


+2
重新编译
带姓名的光标
一眼就能看出谁在写方法部分,谁在修改表 1。
评论贴在行旁
针对某句话提问,改好后标记为已解决。
看看谁在线
头像显示谁在项目中,以及他们正在编辑哪个文件。
运行自己的服务器
只需一条命令,就能在你的团队掌控的机器上安装 Oleafly Server。
研究空间
一个窗口,容纳整篇论文。
源文件、插图、参考文献和编译出的 PDF 都在同一个窗口里。大纲和结构图会跟随你打开的文件,点一下标题或标签,就能直接跳到它所在的那一行。
- 资源管理器、大纲和结构共用一个侧边栏
- 点击标签或引用,跳到它的定义处
in mammary epithelial cells
Cells read growth factors through a small number of shared kinases, yet they respond to EGF, HGF and IGF-1 with different fates. We imaged ERK activity in 2,184 MCF10A cells with a kinase translocation reporter and found that each ligand produces pulses at a characteristic frequency rather than a characteristic amplitude. Frequency alone identifies the ligand in 91 percent of cells.
1 IntroductionEpidermal growth factor, hepatocyte growth factor and insulin-like growth factor all reach the nucleus through the same RAS–RAF–MEK–ERK module, yet epithelial cells migrate, scatter or grow. How a single kinase carries that distinction has been a long-standing question. Early work in PC-12 cells tied fate to whether ERK activity was transient or sustained.
Live-cell reporters changed the picture. In single cells ERK does not rise and settle; it fires in discrete pulses, and the pulse frequency tracks the dose. Here we ask whether frequency also encodes which growth factor a cell has seen.
We make three contributions. A decoder identifies the ligand from inter-pulse intervals alone. A perturbation screen shows that the code survives mutant RAS. And pulse count, not pulse height, predicts whether a cell enters S phase.
Section 2 describes the reporter and the decoder, and Section 3 discusses what a frequency code means for drugs that target the pathway. Traces, segmentation masks and analysis code are available on request.
We stimulated MCF10A cells with saturating doses of EGF, HGF or IGF-1 and tracked ERK every two minutes (Table 1). After the first spike the responses diverged: EGF drove pulses about every 19 minutes, HGF about every 41, and IGF-1 rarely more than once an hour.
Pulse amplitude overlapped almost completely between ligands, so we trained a decoder on inter-pulse intervals alone. Given the first two hours of a single cell’s trace, it named the ligand correctly in 91 percent of held-out cells.
Mutant RAS rescales this code rather than erasing it, so the frequency channel survives oncogenic signalling.
2.3 Sustained pulsing predicts S-phase entryCells that kept pulsing through hour eight were far more likely to enter S phase. Entry tracked how many pulses a cell fired rather than how tall they were, and blocking the receptor early removed the effect.
The same cells, stimulated with a constant dose, fired fewer pulses and entered S phase less often.
Because identity sits in the timing of pulses and not in their height, one cascade can serve many ligands at once. A cell does not need a separate kinase for each growth factor; it needs a way to read how often ERK fires.
The code is also robust. Mutant RAS shifts the pulse rates but keeps the ligands apart, which suggests that a downstream reader of frequency, rather than of level, decides what the cell does next.
We expect the same logic to hold in other receptor tyrosine kinase pathways, and we have begun to test it in cells that express other reporters.
Two limits are worth stating. We worked in one epithelial cell line, and the decoder was trained on the first two hours of each trace, so longer recordings may separate the ligands even better.
We thank the imaging facility for help with the microscopes and our reviewers for pressing us on the decoder. This work was supported by the Northbridge Institute.
References[1] Regot, S., Hughey, J. J., Bajar, B. T., Carrasco, S. and Covert, M. W. High-sensitivity measurements of multiple kinase activities in live single cells. Cell 157, 1724–1734 (2014).
[2] Gillies, T. E., Pargett, M., Silva, J. M., Teragawa, C. K., McCann, C. and Albeck, J. G. Oncogenic mutant RAS signaling activity is rescaled by the ERK/MAPK pathway. Mol. Syst. Biol. 16, e9518 (2020).
[3] Albeck, J. G., Mills, G. B. and Brugge, J. S. Frequency-modulated pulses of ERK activity transmit quantitative proliferation signals. Mol. Cell 49, 249–261 (2013).
[4] Aoki, K., Kumagai, Y., Sakurai, A., Komatsu, N., Fujita, Y., Shionyu, C. and Matsuda, M. Stochastic ERK activation induced by noise and cell-to-cell propagation regulates cell density-dependent proliferation. Mol. Cell 52, 529–540 (2013).
实时评论
在句子旁边,直接提问。
选中几个词,选择“添加评论”。输入 @ 可以把共同作者拉进来。回复都留在文字旁边的同一个讨论里,问题有了答案,点“解决”就能把它收起。
- 周围的文字移动时,讨论仍然跟着原文走
- 已解决的讨论,需要时点一下就能找回来

重新编译
建议:你的编辑会变成建议,由所有者或编辑者接受
建议模式
从“编辑”切换到“建议”后,你的改动会显示为带修订标记的插入和删除。共同作者可以逐条接受或拒绝,助手的编辑也可以作为建议送达。

重新编译
项目聊天
每个共享项目都在论文旁边配有自己的聊天,不用离开草稿,就能向共同作者问一问某张图。
该文件会为本项目中的所有人恢复到 2026年11月7日 11:20 时的状态。它当前的文本会保留在历史记录中,所以你可以再回到它。
共享历史记录
在共享项目中,可以看到谁在什么时候改了什么,为重要的版本命名,比较任意两个版本,并恢复单个文件或整个项目。在你自己的项目中,每次编译成功后保存的检查点也能做到这些。

Epidermal growth factor, hepatocyte growth factor and insulin-like growth factor all reach the nucleus through the same RAS–RAF–MEK–ERK module, yet epithelial cells migrate, scatter or grow. How a single kinase carries that distinction has been a long-standing question [1]. Early work in PC-12 cells tied fate to whether ERK activity was transient or sustained [2].
Live-cell reporters changed the picture. In single cells ERK does not rise and settle; it fires in discrete pulses, and the pulses differ from ligand to ligand [4, 5].
Here we ask whether the timing of those pulses carries the identity of the growth factor. We imaged ERK activity in 2,184 MCF10A cells stimulated with EGF, HGF or IGF-1 and compared the amplitude, width and spacing of every pulse. Each ligand settles into its own rhythm after the first spike and keeps it at every dose we tested, while amplitudes overlap. A decoder that sees only the gaps between pulses names the ligand in most cells.
We stimulated MCF10A cells expressing the ERK kinase translocation reporter [3] with saturating doses of EGF, HGF or IGF-1 and tracked nuclear and cytoplasmic signal every two minutes for four hours. All three ligands produced the same first spike within ten minutes (Figure 1b). After that the responses diverged: EGF drove pulses about every 19 minutes, HGF about every 41, and IGF-1 rarely more than once an hour (Table 1).
2.2 Frequency, not amplitude, identifies the ligandPulse amplitude overlapped almost completely between ligands, so we trained a decoder on the gaps between pulses alone. Given the first two hours of a single cell’s trace, it named the ligand correctly in 91 percent of held-out cells, compared with 47 percent for a decoder that saw only amplitudes.
浏览器中的共同作者
只要你的服务器允许,共同作者就能在浏览器中打开项目,进行编辑、评论、聊天,并阅读最新的 PDF。他们的角色决定了能修改哪些内容。
Oleafly Server
自己运行协作服务器。
服务器是一个单独的程序,由你的实验室或院系在自己的机器上运行,项目保存在磁盘或 S3 中。成员可以用 GitHub、你们的 OIDC 提供方或电子邮件登录,管理控制台则涵盖谁能加入、有哪些项目、电子邮件和安全日志。
- 人员、访问规则、服务器设置和安全日志,都在一个控制台里
- 不想自己托管?一键添加托管服务器
项目所有者将这些人邀请为访客。访客只能看到受邀的项目,且不能共享这些项目。
-
A
编辑者 等待批准已邀请 批准拒绝撤回amara.okafor@riverton.edu访客Nina Takeda邀请他们加入ERK Pulse Frequency Encodes Growth Factor Identity · 到期日:Nov 21, 2026
所有登录过此服务器的人。
要让新成员加入,请在允许的人员中添加。
-
Daniel Hartdaniel@northbridge.org已启用 成员 设备停用移除 -
Elena Petrovaelena@northbridge.org已启用 成员 设备停用移除 -
Lucas Moreaulucas@northbridge.org已启用 成员维护者 设备停用移除 -
Nina Takedanina@northbridge.org已启用 成员 设备停用移除 -
已启用 管理员 设备Prajwal你prajwal@northbridge.org
这些设置适用于此服务器上的所有人。
此服务器上的登录、设备变更和管理员操作。
- Prajwal → 更改了服务器设置允许链接共享:开启Nov 7, 2026, 10:42 AM ·开始日期203.0.113.24
- Prajwal → 更改了角色从成员改为维护者Nov 7, 2026, 10:41 AM ·开启Lucas Moreau ·开始日期203.0.113.24
- Prajwal → 批准了一个访客邀请项目:ERK Pulse Frequency Encodes Growth Factor Identity ·角色:编辑者Nov 7, 2026, 10:41 AM ·开启amara.okafor@riverton.edu ·开始日期203.0.113.24
- Nina Takeda → 邀请了一位访客项目:ERK Pulse Frequency Encodes Growth Factor Identity ·角色:编辑者Nov 7, 2026, 9:58 AM ·开启amara.okafor@riverton.edu ·开始日期198.51.100.7
- Daniel Hart → 已登录通过 GitHub · 设备:Firefox on LinuxNov 7, 2026, 9:12 AM ·开始日期198.51.100.31
- Sofia Marín → 已登录通过 GitHub · 设备:Chrome on WindowsNov 7, 2026, 8:47 AM ·开始日期198.51.100.58
云同步
你写作用的每台电脑上,都有你的项目库。
备份到 Dropbox、Google Drive 或任意 S3 存储,每个项目都会在你自己拥有的位置留一份副本。在另一台电脑上打开 Oleafly,项目库已经在那里了。API 密钥和 AI 对话只留在创建它们的那台电脑上。
- 每台电脑可以单独选择同步哪些项目
- 从云存储中删除内容前,一定会先询问你
-
ERK pulse frequency encodes growth factor identity尚未在 Dropbox 中刚刚已同步正在上传第 9 个文件,共 38 个正在上传第 24 个文件,共 38 个正在上传第 38 个文件,共 38 个2.4 MB
-
TeXFix-Bench: Engine-Grounded Evaluation of Single-Fault LaTeX Compilation Repair尚未在 Dropbox 中刚刚已同步正在上传第 18 个文件,共 61 个正在上传第 42 个文件,共 61 个正在上传第 61 个文件,共 61 个5.1 MB
-
Larkspur: Block-Sparse Attention for Long-Context Inference尚未在 Dropbox 中刚刚已同步正在上传第 31 个文件,共 120 个正在上传第 77 个文件,共 120 个正在上传第 120 个文件,共 120 个18 MB
-
Grant Proposal 2027仍在从 Lab iMac 上传仅在 Dropbox 中412 KB so far3.2 MB
LaTeX、Typst 和 Markdown
投稿要求什么格式,就用什么格式写。
LaTeX 可以用内置的 Tectonic 编译,也可以用你自己安装的 TeX。Typst 自带内置编译器,Markdown 则通过 Pandoc 处理。大纲、结构图和实时预览在三种格式中用法完全一致。
- 编译器随应用附带,不用再折腾 TeX 安装
- 在工具库中转换 LaTeX、Typst 和 Word
可视化模式
编辑时,标题、公式、插图和引用都会就地渲染,表格则有网格,可以处理行、列和合并单元格。.tex 文件始终是唯一的副本。
Typst 表单
模板的参数会显示为带标签的字段,填写标题页或简历条目时,不必去改代码。
Quarto 与 Calepin
在论文里直接运行分析。
在正文旁边写 R、Python 或 Julia,结果会直接进入 PDF:表格、数字和图表都会随数据更新。Quarto 可以把 .qmd 和 .Rmd 文件构建成 PDF、网页或幻灯片。Calepin 则在 Typst 文档中做同样的事。
- 只有在你为该项目允许后,代码才会运行
- 模板库中有 Quarto 和 Calepin 模板
此项目的代码已关闭,因此此文档中的代码没有运行。仅在你信任项目中每个文件时才允许代码。
在此项目中允许代码Each ligand settles into its own pulse rate after the first spike.
EGF pulses fastest and IGF-1 slowest.
Pulse amplitude overlapped almost completely between ligands, so we trained a decoder on the gaps between pulses alone. Given two hours of a single cell's trace, it named the ligand in 91 percent of held-out cells.
内置于 Oleafly。快速、离线、零配置。支持纯 LaTeX 和大多数常用宏包。
仅适用于受信任的项目。运行本地 TeX 发行版,可读取你的账户有权访问的文件。支持完整宏包集和多轮编译工作流。宿主机命令执行仍是一项单独的按项目权限。
未在这台电脑上检测到 TeX 发行版。
无需管理员权限的精简版 TeX Live。核心下载量约为 250 MB。期刊模板之后可能会添加宏包,总计最高约 1 GB。提前安装后,系统 LaTeX 便可在你明确选择时使用。
如果 TeX 装在 Oleafly 不会查找的位置,请在这里选择它的文件夹。可以选包含 latexmk 的文件夹,也可以选它上方的 TeX Live 年份文件夹,例如 texlive/2026。
Oleafly 内置,无需任何配置。只有当文档首次导入 @preview 包时才需要联网。
一键安装 TeX
无需管理员权限即可安装 TinyTeX,也可以让 Oleafly 使用你已有的 TeX,或者为每个项目固定一个 Typst 版本,让旧论文依然按原样构建。
导出与转换
从同一份 LaTeX 源码导出 Word、带 MathML 的 HTML、Markdown、Typst 和纯文本,书籍还能导出 EPUB,Beamer 幻灯片还能导出 PowerPoint。你也可以把 LaTeX 项目转换成 Typst,并阅读一份说明改动内容的报告。
校对
拼写和语法都在你的电脑上检查。LaTeX 命令、数学公式和引用键会被跳过。
引用
同时搜索 arXiv、Semantic Scholar、Crossref、PubMed 和 OpenAlex,直接从结果列表中引用。
参考文献
引用、标签和 bib 条目集中在一个面板里。输入时,未定义和重复的键会被标出来。
- albeck2013frequencyAlbeck et al. 2013 · Frequency-modulated pulses of ERK activity transmit qua…
- aoki2013stochasticAoki et al. 2013 · Stochastic ERK activation induced by noise and cell-to-…
- purvis2013encodingPurvis & Lahav 2013 · Encoding and decoding cellular information through sign…
Git、GitHub 与 Zotero
在侧边栏提交并推送,把项目发布到 GitHub,还能从 Zotero、EndNote 或任意 .bib 文件导入参考文献。
- albeck2013frequencyAlbeck et al. 2013 · Zotero · Frequency-modulated pulses of ERK activity transmit qua…
- albeck2008modelingAlbeck et al. 2008 · Zotero · Modeling a snap-action, variable-delay switch controlli…
- gillies2020oncogenicGillies et al. 2020 · Zotero · Oncogenic mutant RAS signaling activity is rescaled by…
从 Zotero 引用
写作时输入 @ 即可搜索你的 Zotero 文献库。只有你引用的论文才会写入 .bib,并使用你的 Better BibTeX 引用键。
Growth Factor Identity
Growth Factor Identity
- Decoder on pulse intervals: 91% of cells
- Decoder on amplitude: 47%
- Same first spike for every ligand
Thank the Hart lab first. Then the question: how does one kinase tell EGF from IGF-1?
Point at the EGF bar. Every ligand gives the same first spike; after that the rates split.
放映你的幻灯片
直接从 PDF 查看器开始放映。演示者视图会显示你的备注、下一张幻灯片和计时器。
画出来,生成 TikZ
在画布上放好方框和箭头,再切换到 TikZ 源码继续编辑。
/image-to-latex
将笔记、公式或表格的图片转换为可编辑的 LaTeX。
图片设备端模型无需项目
/pdf-to-latex
将 PDF 转换为 LaTeX,并保留数学公式、图片和结构。
数学公式提取图片导出本地运行
47 个免费工具
PDF 转 LaTeX、arXiv 转 Typst、电子表格转表格、BibTeX 检查等。无需项目即可使用。
你的所有工作,集中一处
论文、学位论文、幻灯片和基金申请书草稿放在同一个书架上,封面一眼就能分辨。搜索既能按名称找到项目,也能按项目里的某一行内容找到,书签则把本周要用的项目放在最前面。
AAAI Style Paper
XeLaTeXCC0-1.0
ACL Style Paper
XeLaTeXCC0-1.0
arXiv Preprint
XeLaTeXCC0-1.0
CVPR Style Paper
XeLaTeXCC0-1.0
ICLR Style Preprint
XeLaTeXCC0-1.0
ICML Style Preprint
XeLaTeXCC0-1.0
- 引擎
- XeLaTeX
- 许可证
- CC0-1.0
- 语言
- 英语
模板包
下载会议、期刊或学位论文模板包。下载到本机后,每个模板都能离线编译。
导入你的 Overleaf 项目
在 Oleafly 中登录 Overleaf,导入你的项目并保持同步。.zip、Word 文件、arXiv 论文或 GitHub 仓库同样可以导入。
预检
投稿前检查版式、无障碍、参考文献,以及任何会破坏盲审的内容。
AI 助手
寻求帮助,查看差异,再做决定。
助手可以读取你的项目、运行编译、修复发现的错误,或起草 TikZ 图形。每处编辑都以差异形式呈现,你可以审阅和撤销。可以使用托管模型或自己的端点,也可以运行 Ollama,让一切都留在你的电脑上。
- 计划模式先读取项目,再写入任何内容
- 关闭 AI,应用的其余功能照常可用
replace_in_fileresults.tex第 31 行results.tex第一处更改 · 第 31 行内联编辑
选中一句话,按 ⌘L(Windows 和 Linux 上为 Ctrl+L),说明要怎么改。改写结果会以差异形式就地出现,由你接受或拒绝。
根据描述生成图形
描述你需要的图形。助手会编写 TikZ、编译它并检查渲染结果,一键即可保存到你的项目中。
任意模型,离线也行
使用你自己的 OpenAI、Anthropic、Gemini、DeepSeek 等服务的密钥,或者使用任何 OpenAI 兼容端点,也可以在 Ollama 或 LM Studio 中运行本地模型。
- 91% 的解码器准确率来自训练所用的场次。请在新的重复实验上测试。
- 对 HGF 和 IGF-1 而言,频率与剂量相互混杂。请增加一个匹配的对照。
- 没有任何结果表明频率具有因果作用。请在固定振幅下滴定 MEK 抑制剂。
评审与研究技能
投稿前可以要一份友好的评审,也可以要一份严厉的评审;还可以开启研究技能,让助手了解你所在领域的论文是怎么写的。
CLI 智能体
直接用你已经在付费的智能体。
Claude Code、Codex CLI、Gemini CLI 等命令行智能体通过 ACP 连接,直接在你的项目里工作。它们的修改和其他改动一样,交给你审阅。
- 每个智能体都用你自己的账户登录
- 逐条批准修改,也可以让长任务自行运行
results.tex显示更改Claude Desktop、Claude Code、Cursor 及其他 MCP 客户端可以读取、编辑和编译当前打开的项目,每次更改都需经你批准。
XXXXXX
claude mcp add --transport http oleafly http://127.0.0.1:5323/mcp --header "Authorization: Bearer XXXXXX"
通过 MCP 使用 Oleafly
Claude Code、Codex、Cursor 或 Claude Desktop 可以通过 Oleafly 的 MCP 服务器读取你的项目、编译它并检查 PDF。你可以把它设为只读,或者要求修改必须经过你批准。
让代理通过 Cua Driver 使用这台电脑上的应用和网站。代理只能访问你在这里列出的应用和网站。
Cua Driver 文档
已安装 Cua Driver 0.38.0。
位于 /Applications/CuaDriver.app/Contents/MacOS/cua-driver
为某个网站开启“使用我已登录的浏览器”后,代理会在你自己的 Chrome 或 Edge 中操作该网站。关闭时,则使用单独的浏览器配置文件。
Cua Driver 正在按你的限制运行。
Computer Use
当任务需要的不只是项目文件夹时,智能体可以通过 Cua Driver,在你允许的桌面应用和网站中工作。
PDF 预览
专为草稿打造的 PDF 查看器。
缩放、双页并排、深色阅读模式和搜索都在工具栏里。点击 PDF 中的某个词,就能跳到它在编辑器里的那一行。构建失败时,上一份正常的 PDF 会一直保留,让你边修边看。
- 在工具栏里一键全屏放映 PDF
- 长文档可以展开大纲抽屉
in mammary epithelial cells
Cells read growth factors through a small number of shared kinases, yet they respond to EGF, HGF and IGF-1 with different fates. We imaged ERK activity in 2,184 MCF10A cells with a kinase translocation reporter and found that each ligand produces pulses at a characteristic frequency rather than a characteristic amplitude. Frequency alone identifies the ligand in 91 percent of cells.
1 IntroductionEpidermal growth factor, hepatocyte growth factor and insulin-like growth factor all reach the nucleus through the same RAS–RAF–MEK–ERK module, yet epithelial cells migrate, scatter or grow. How a single kinase carries that distinction has been a long-standing question. Early work in PC-12 cells tied fate to whether ERK activity was transient or sustained.
We stimulated MCF10A cells expressing the ERK kinase translocation reporter with saturating doses of EGF, HGF or IGF-1 and tracked nuclear and cytoplasmic signal every two minutes for four hours. All three ligands produced the same first spike within ten minutes. After that the responses diverged: EGF drove pulses about every 19 minutes, HGF about every 41, and IGF-1 rarely more than once an hour.
Live-cell reporters changed the picture. In single cells ERK does not rise and settle; it fires in discrete pulses.
Pulse amplitude overlapped almost completely between ligands, so we trained a decoder on inter-pulse intervals alone. Given the first two hours of a single cell’s trace, it named the ligand correctly in 91 percent of held-out cells, compared with 47 percent for a decoder that saw only amplitudes.
Mutant RAS rescales this code rather than erasing it, which suggests the frequency channel survives oncogenic signalling.
Cells that kept pulsing through hour eight were far more likely to enter S phase. Entry tracked how many pulses a cell fired rather than how tall they were, and blocking the receptor early removed the effect.
3 DiscussionBecause identity sits in the timing of pulses and not in their height, one cascade can serve many ligands at once.
项目里的终端
在论文旁边用你常用的 shell 运行脚本,最多可开十个标签页。输出中的路径和错误都可以点击。
更改界面语言不会更改文档语言或校对设置。
随心定制
浅色和深色主题、自选的强调色和字体、11 款应用图标,以及 22 种界面语言。
oleaflyc
应用能做的一切,在终端里也能做。
oleaflyc 使用与应用相同的引擎和设置,来构建、检查和编辑 Oleafly 项目。每条命令都能以 JSON 返回结果,所以脚本、CI 任务和编程智能体都可以直接调用。
oleaflyc build --watch构建与监视
编译 LaTeX、Typst 和 Markdown,方式与应用完全一致;也可以监视一个文件夹,每次保存都重新构建。
oleaflyc check --jsonCI 里的预检
在 GitHub Action 中运行预检和 lint 检查,论文还没准备好投稿时就让任务失败。
oleaflyc agent run "Make it compile"无头智能体
从命令行把任务交给助手。它使用你自己的服务商和密钥,在你应用更改之前,不会有任何内容写进项目。
oleaflyc ask "Which sections cite preprints?"针对项目提问
不用打开论文,也不改动任何文件,就能得到关于论文的回答。
oleaflyc daemon status退出后继续运行
关闭 Oleafly 时,研究任务会转移到一个小型后台助手继续运行,再次打开时又回到应用里。
oleaflyc cite 10.1038/nature14539从命令行处理引用
检索文献,用 DOI 把一篇论文加入 refs.bib,并赶在审稿人之前找出失效的引用键。
oleaflyc mcp给智能体用的 MCP 服务器
Claude Code、Codex、Cursor 和 Gemini 通过它在你的项目上工作,使用的工具与助手相同。
oleaflyc remote sync你的协作服务器
登录、同步共享项目、运行 Oleafly Server,全都能在命令行里完成。
oleaflyc import overleaf直接获取 Overleaf 项目
登录一次 Overleaf,就能把你的任何 Overleaf 项目拉取为本地文件夹,并在命令行里保持同步。
oleaflyc convert main.tex --to docx导出与转换
把论文转成 Word、HTML 或 Markdown,或在 LaTeX 和 Typst 之间转换,全程不用打开应用。
oleaflyc checkpoint restore v12从命令行管理检查点
列出 Oleafly 在每次成功构建后保存的版本,一条命令就能恢复其中任何一个。
ssh · slurm · cloud在更强的机器上跑实验
把分析和训练任务通过 SSH 发送到服务器、Slurm 集群或云主机,再把结果拉回论文里。
强大的研究工作台
从第一条引用到最终 PDF。
起草、编译、校对、检索、引用、审阅与发表,无需拼凑多个工具。Oleafly 把源码、预览、参考文献、AI 与 Git 历史放在一个由你掌控的工作区里。
导入已有内容
在本地将许多基于文本的 PDF 重建为可编辑 LaTeX,经托管 Pandoc 转换 DOCX,或用视觉模型转录页面照片。之后可选 AI 润色。
导出到目标格式
通过上下文菜单导出 PDF、DOCX、HTML、Markdown 等格式。也可以将整个项目下载为源码 ZIP。
按你的习惯定制
Vim 模式、浅色与深色主题、强调色、七种布局预设与可编辑快捷键。编辑器应适合你的双手。
代码或可视化写作
将任意 LaTeX 或 Markdown 切到可视化编辑:标题、列表、粗体、链接直接编辑,两种模式共享撤销。不支持的命令以原始块显示而非静默丢弃,回源无损。
词级 SyncTeX
在 PDF 中 ⌘-点击一词即可跳到源码中的对应位置。⌘⇧J 跳回。
代码智能
转到定义、查找引用,并在所有文件中重命名标签、引用键或宏。
斜杠命令
输入 / 打开 Notion 风格插入菜单:/figure、/table、/section、/cite、/math。
title = {Raft-based KV stores}, …
按 DOI 引用
粘贴 DOI、arXiv id 或 URL。Oleafly 获取 BibTeX、去重并插入 \cite。
导入整库文献
批量导入 BibTeX、RIS、EndNote XML 与 Zotero RDF。写入 .bib 前会与项目去重。
命令面板
⌘K 模糊搜索应用中的所有操作:项目、文件、设置等。
不限模型的 AI
连接你偏好的模型提供商、在本地运行 Ollama,或使用自带账号的已安装 CLI 智能体。根据手头的工作设置审批模式。不用 AI,编辑器也照常可用。
随时调研
代理搜索 alphaXiv 与 OpenAlex,拉取论文内容,并在写入文献前用 Crossref 校验引用。
内置校对
借助 Hunspell 和 Harper 在本地校对拼写与语法。Oleafly 会忽略 LaTeX 命令和数学公式,让检查专注于你的正文。
给研究一个独立的工作区
写下任务说明,选择一个代理,它在隔离副本中工作时你可以继续写作。跟进它的活动,预览更改的文件,并应用你想要的结果。
从一页到六百页
无论是一页简历还是书级学位论文,编辑器与虚拟化 PDF 预览都保持流畅。
适合真正阅读的 PDF 查看器
将预览拆到第二显示器、双页对开、夜间反色,或全屏演示。
图表编排器
描述一张图,AI 编写 TikZ 或 CeTZ,自行编译图形,并根据渲染结果修复重叠。也可以在真正的画布上自己绘制:LaTeX 用 TikZ,Typst 用 fletcher,Markdown 用 Mermaid 流程图,代码与图形双向同步。导出透明背景的高分辨率 PNG,完全离线。
三引擎,一工作区
把 LaTeX、Typst 和 Markdown 项目放在同一个项目库里。为每个项目选择编译器,并使用同样的预览、参考文献、版本管理和 AI 工具。
LaTeX 编辑器
用 LaTeX、Typst 或 Markdown 写作,配备实时 PDF 预览与词级 SyncTeX。编译器随应用附带,无需配置 TeX,也没有编译队列。
AI 助手
让助手编辑源码、编译项目并检查 PDF。可连接模型提供商、Ollama 或已安装的 CLI 智能体。较长的工作交给独立的研究任务,你继续写作。
文献管理器
导入 BibTeX、RIS、EndNote 或 Zotero,自动去重。粘贴 DOI 或 arXiv ID,即可创建经 Crossref 校验的文献。
简历生成器
从面向 ATS 的简历模板开始,检查提取的文本,为每次申请分叉一份定制副本,且不丢失早前版本。
版本控制
每次成功编译后,为变更的源码打检查点。用 Git 从容提交、对比版本,并在同一工作区推送到 GitHub。
PDF 工具箱
阅读长 PDF,把许多基于文本的 PDF 重建为 LaTeX,并导出 PDF、DOCX、HTML 或投稿用的源码压缩包。
研究工作区
一个研究任务,从头到尾。
给代理一个聚焦的任务,阅读它提议的修改,准备好了就编译。成果始终留在你的项目里。
$ theta_(t+1) = theta_t - eta nabla_theta cal(L) $- 01创建研究项目先为你的手稿、资料与研究任务安一个家。
- 02给代理一个任务选择它需要的文件,说明你想改什么。
- 03审阅提议的修改阅读差异,判断哪些修改应留在手稿里。
- 04编译并保留成果检查 PDF,之后随时继续,任务仍留在项目里。
开始一项新工作
选择这份手稿从何开始。每个部分都由你掌控,随时可改。

研究项目
一个研究文件夹,含章节、资料与审阅笔记,以及一个框定研究问题的首个任务。

导入项目
继续你已有的手稿:压缩包、Word 或 Markdown 草稿,或 GitHub 仓库。

使用模板
从现成版式开始:期刊论文、学位论文、报告、海报或演讲。
新建研究任务
成果先留在独立工作区,待你审阅并应用。
main.tex…\section{方法}我们排除了缺少主要结果的回复。 \end{document}睡眠与记忆
方法
我们排除了缺少主要结果的回复。
研究闭环
调研、写作、验证、发表。
从文献检索到投稿,全程不丢失上下文。Oleafly 把你的资料、参考文献、插图、编译输出和历史记录集中在一个可检查的工作流中。
$ bold(tau) = bold(J)^T bold(F) $- doi:10.1145/3576915@article{gatto24} ✓1
调研
检索开放的研究资源,把 DOI 和 arXiv ID 转换为 BibTeX,并在引用写入参考文献表之前完成校验。
- \section{Results}2
写作
用 LaTeX、Typst 或 Markdown 写作,配合模板、全项目代码智能,以及带审批控制的可选 AI 助手。
- compiled · 0 errors ✓3
验证
在本地编译,查看清晰的错误信息,并在源码与 PDF 之间跳转。预检会展示解析器和读屏软件将提取的内容。
- PDFarXivGitHub ↑4
发表
导出 PDF、DOCX 或 HTML,打包源码用于投稿,并从工作开始的同一个文件夹推送到 GitHub。
产品对比
一个研究工作区,取代其余所有工具。
在一个地方完成写作、编译、校对、文献检索、引文管理、PDF 审阅、修改自动化和论文交付。Oleafly 将整套研究工具整合进一个免费开源应用。看看下方这些替代产品还缺少什么。 文档.
| Oleafly | OpenAI PrismRIP · 2027年1月1日 | Overleaf | Typst Web App | Curvenote | LyX | TeXstudio | Texifier | VS Code + LaTeX | Paperpal | Writefull | 简历生成器 | ||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 所有权与访问 | 价格 | 免费开源,无需订阅 Oleafly | 拥有 ChatGPT 账户即可免费使用 | 提供免费版,付费方案可解锁更多功能 | 提供免费版,Pro 版需额外付费 | 提供免费版,付费方案支持私有项目 | 免费开源 | 免费,采用 GPL 许可证 | 付费,提供免费试用版 | 免费;AI 有用量限制或需购买付费方案 | 提供免费版,Prime 版需额外付费 | 提供免费版,Premium 版需额外付费 | 视产品而定,大多采用订阅制 |
| 无需账户 | 是 | 否 | 否 | 否 | 否 | 是 | 是 | 是 | 是 | 否 | 否 | 视产品而定 | |
| 平台 | macOS、Windows 和 Linux 桌面应用 | 仅限网页版 | 网页版,另有自托管的 Server Pro | 网页版,另有本地部署方案 | 网页编辑器,另有开源 CLI | Windows、macOS 和 Linux | Windows、macOS 和 Linux | macOS 和 iOS。Windows 版即将推出 | Windows、macOS 和 Linux | 网页、Word、Google Docs、Overleaf | Word、Overleaf 和网页 | 以网页版为主 | |
| 支持离线使用 | 离线写作、编译、检查和提交。共享编辑稍后合并 | 否,采用云端工作区 | 否,属于云服务 | 否,Web 应用需要联网 | 否,编辑器属于云服务 | 是 | 是 | 是 | 是 | 否 | 否 | 很少支持 | |
| 界面语言 | 21 种界面语言 | 官方资料未说明 | 18 种 | 官方资料未说明 | 官方资料未说明 | 26 种,完成度不一 | 约 30 种社区翻译 | 8 种 | 15 种 | 官方资料未说明 | 官方资料未说明 | 因产品而异 | |
| 文件保留在本地磁盘 | 你拥有的普通文件夹 | 云存储 | 云存储 | 使用云存储;编译器可在本地运行 | 云存储 | 是 | 是 | 本地文件与文件服务商 | 是 | 云服务与编辑器集成 | 文件保留在 Word 或 Overleaf 中 | 通常采用云存储 | |
| 协作与同步 | 实时协作 | 内置。实时共同编辑、带姓名的光标、共同作者不限人数 | 协作者人数不限 | 内置支持,人数限制取决于方案 | 内置支持 | 实时编辑与评论 | 不支持实时共同编辑 | 通过 Teamtype 点对点协作 | 通过 Texifier Connect 实时协作 | 通过 Live Share 实现 | 依托 Word、Google Docs 或 Overleaf | 依托 Word 或 Overleaf | 视产品而定 |
| 实时评论 | 评论串,支持回复、@提及和标记已解决 | 是 | 是 | 是,所有者需要 Pro 版 | 是 | 否 | 否 | 官方资料未说明 | 仅通过扩展实现 | 仅限宿主编辑器内 | 仅限宿主编辑器内 | 少见 | |
| 自有服务器 | Oleafly Server:一条命令安装,带管理控制台和 SSO | 否 | 是 | 本地部署方案 | 官方资料未说明 | 否 | 点对点,由一位用户托管 | 官方资料未说明 | 否。Live Share 由 Microsoft 托管 | 官方资料未说明 | 官方资料未说明 | 少见 | |
| 云同步 | 通过 Dropbox 同步项目库和设置 | 云端工作区 | 云端工作区,另提供付费 Git 和 Dropbox 同步 | 云端工作区,Pro 版另支持 Git 同步 | 云端项目与互联发布 | 使用 Git 或自选同步服务 | 使用 Git 或自选同步服务 | 通过 Texifier Connect 实时同步 | 使用 Git 或自选同步服务 | 云端工作区 | 在 Word 或 Overleaf 内使用 | 通常采用云存储 | |
| 共享项目与角色 | 共享项目,分所有者、编辑者和查看者角色 | 共享项目,协作者人数不限 | 团队与机构方案 | 共享项目,并为组织提供本地部署 | 团队工作区与访问控制 | 无内置团队空间 | 无内置团队空间 | 通过 Texifier Connect 共享项目 | GitHub 组织与 Live Share | 机构控制面板与许可证管理 | 机构访问 | 视产品而定 | |
| 版本历史 | 每次编译成功后保存检查点,可随时恢复 | 云端修订记录 | 付费方案提供完整历史记录 | Pro 版支持 Git 同步 | 内置版本历史 | Git 及其他版本控制工具 | 手动管理或使用 SVN | 使用文件服务商或 Git 工具 | 内置 Git,需自行配置 | 官方资料未说明 | 使用宿主应用的历史记录 | 视产品而定 | |
| Git 与 GitHub | 内置 Git 面板。发布到 GitHub,并推送和拉取 | 暂不支持,已在计划中 | 付费方案支持 Git 和 GitHub | Pro 版支持 GitHub 和 GitLab 同步,处于实验阶段 | GitHub 集成和 CLI | 基础功能,仅限本地 | 内置 Git 面板和 SVN | 官方资料未说明 | 内置 | 官方资料未说明 | 官方资料未说明 | 否 | |
| 写作与发布 | 文档引擎 | 在同一工作区使用 LaTeX、Typst 和 Markdown | 仅支持 LaTeX | 仅支持 LaTeX | 仅支持 Typst | MyST Markdown 与 Jupyter | LaTeX 与 DocBook | 仅支持 LaTeX | LaTeX 与 Markdown | 通过扩展支持 | Word、Google Docs、Web 和 Overleaf | Word 与 Overleaf LaTeX | 仅支持简历格式 |
| TeX 配置 | 内置 Tectonic、Typst 和 Pandoc,无需安装 TeX | 无需安装,在云端编译 | 无需安装,在云端编译 | 无需安装 TeX | 无需安装 TeX | 需要安装 TeX 发行版 | 需要安装 TeX 发行版 | 内置排版引擎与软件包管理器 | 需要安装扩展和 TeX | 使用 Overleaf 处理 LaTeX | 使用 Overleaf 处理 LaTeX | 无需配置 TeX | |
| 代码与可视化编辑 | 源代码与可视化编辑并用,双向转换不丢失内容 | 源代码与预览 | 源代码与富文本模式 | 源代码与实时预览 | 支持 LaTeX 数学公式的可视化编辑器 | 结构化可视化编辑器 | 仅源代码 | 源代码与 PDF 预览 | 仅源代码 | 可视化写作助手 | 行内写作建议 | 仅可视化编辑 | |
| 项目级代码智能 | 自动补全,并支持定义跳转、引用查找和项目级重命名 | 编辑器与 AI 辅助 | 自动补全与引用检索 | 错误提示、自动补全与 Assists | 结构化引用与公式 | 标签、引用、大纲与自动补全 | 结构与引用检查器 | 自动补全、大纲、引用与搜索 | LaTeX Workshop IntelliSense | 无 LaTeX 代码工具 | 用于 LaTeX 代码的 TeXGPT | 否 | |
| 源代码与 PDF 同步 | LaTeX 和 Typst 均支持源码与 PDF 双向跳转 | 即时 PDF 预览 | 源代码与 PDF 同步 | 实时预览 | 不支持源代码与 PDF 同步 | 源代码与输出预览 | 接近词级精度 | 源代码与 PDF 同步 | 正向与反向 SyncTeX | 否 | 使用 Overleaf 预览 | 否 | |
| PDF 阅读模式 | 内置查看器,并支持独立窗口、对页、深色和阅读模式 | 内置预览 | 内置预览 | 内置预览 | 文档预览 | 使用 PDF 查看器 | 集成查看器 | 集成式 PDF 查看器 | 扩展查看器,支持反色模式 | 与 PDF 对话 | 无 PDF 阅读器 | 文档预览 | |
| 文档模板 | 24 个模板包、1,800 多个模板,也可添加自己的模板 | 官方资料未说明 | 丰富的出版商模板库 | Typst Universe 与 Pro 版私有模板 | 付费方案提供专业模板 | 内置文档类 | 内置模板与用户模板 | 官方资料未说明 | 代码片段与社区模板 | 官方资料未说明 | 否 | 核心功能 | |
| 项目与文档导入 | ZIP、文件夹、GitHub、arXiv、Word、HTML、Typst、Markdown、PDF 和在线 LaTeX 账号 | 支持文件夹、ZIP 或拖放导入 | 支持文件和 ZIP,付费方案另提供集成功能 | 可转换 LaTeX、Word 和 Markdown | Jupyter 内容与参考文献导入 | LaTeX 及其他文本格式 | 打开本地源文件 | 本地文件夹与云端文件 | 打开本地源文件 | 上传 PDF 与文档 | Word 与 Overleaf 文档 | 通常支持 PDF 或 DOCX | |
| 导出格式 | PDF、PDF/A、PDF/UA、DOCX、HTML、EPUB、PowerPoint、Markdown、Typst 和 LaTeX | PDF 与项目文件 | PDF 与源文件 ZIP | PDF、PNG 和 SVG | PDF、Word、LaTeX、Typst 和 MyST | PDF、HTML、Word、RTF、ePub 等 | PDF 与源文件 | PDF 与源文件 | 取决于本地工具 | Word 工作流 | 带修订记录的 Word 更改 | 通常支持 PDF 和 DOCX | |
| 可视化与 AI 图表工具 | 在画布上绘制,生成 TikZ、Typst 或 Mermaid | 将图片和语音转换为代码 | 手写 TikZ 与 AI 工具 | Typst 软件包与原生绘图 | 交互式图表与插图 | 可视化表格与图片工具 | 手写 TikZ,并提供编辑辅助 | 手写 TikZ | 手动编写或使用扩展 | 否 | AI 生成表格、公式与插图 | 否 | |
| 内置工具 | 47 个转换器,另有引用和数学工具 | 图片和语音转 LaTeX | 无专门的工具库 | 官方资料未说明 | 官方资料未说明 | 否 | 表格、插图和快速入门向导 | 官方资料未说明 | 通过扩展实现 | 改写、翻译、引用、PDF 对话 | 改写、学术化润色、标题和摘要生成器 | ATS 检查器和求职信生成器 | |
| 研究与 AI | AI 助手,不限模型 | +4自带密钥。10 家托管服务商或任意自定义端点 | 内置 OpenAI 模型 | 内置研究 AI,每日额度因方案而异 | 无 AI 助手 | 官方资料未说明 | 无 | Mistral、Claude、OpenRouter、ChatGPT 或本地模型 | 无 | Copilot、自带密钥或扩展 | Paperpal AI | Writefull 模型与 GPT | 内置支持,可选模型因产品而异 |
| 编辑器内的 CLI 智能体 | +714 种智能体在编辑器内运行,另支持任何 ACP 智能体 | 内置 Codex | 官方资料未说明 | 官方资料未说明 | 官方资料未说明 | 官方资料未说明 | 官方资料未说明 | 官方资料未说明 | Copilot、Claude 和 Codex | 官方资料未说明 | 官方资料未说明 | 因产品而异 | |
| 本地 AI | Ollama 或任何本地 OpenAI 兼容服务器 | 否 | 否 | 否 | 否 | 否 | Ollama、llama.cpp 和 llamafile | 否 | 自带密钥或使用扩展 | 否 | 否 | 否 | |
| 应用前审阅 AI 修改 | 每处修改都以差异呈现。四种审批模式,可按文件撤销 | 保留或撤销 | 审阅 AI 建议 | 无 AI 助手 | 官方资料未说明 | 无内置 AI | 无内置 AI | 无内置 AI | 审阅更改 | 审阅写作建议 | 修订模式 | 视产品而定 | |
| AI 检查渲染结果 | 修改源代码、编译,然后检查渲染后的 PDF | 需自行重新编译确认 | 需自行重新编译确认 | 无 AI 助手 | 否 | 无内置 AI | 无内置 AI | 无内置 AI | 执行构建检查,但不审阅 PDF | 不审阅渲染后的 PDF | 不审阅渲染后的 PDF | 否 | |
| AI 技能 | 内置 36 项研究技能,还可下载更多 | 内置技能,如 Paper Review | 一项内置的论断检查 | 无 AI 助手 | 官方资料未说明 | 无 AI 助手 | 可编辑的提示词和 AI 宏 | 无 AI 助手 | 是 | 预设和自定义提示词 | 官方资料未说明 | 至多支持自定义提示词 | |
| 文献与引文检索 | 同时检索 arXiv、Semantic Scholar、Crossref、PubMed 和 OpenAlex | 内置文献检索 | 参考文献检索与 AI 引文建议 | 无内置检索 | DOI 引文查询 | 无内置检索 | 否 | 无内置检索 | 通过 AI 或扩展实现 | 检索超过 2.5 亿篇文章 | Writefull Cite 可查找缺失的引文 | 否 | |
| 研究工作区与任务 | 研究任务:隔离工作区、智能体活动、文件预览与选择性应用 | 官方资料未说明 | 官方资料未说明 | 否 | 官方资料未说明 | 否 | 否 | 否 | 通过 AI 扩展实现 | 否 | 否 | 否 | |
| 参考文献库导入 | Zotero 应用和网页文献库、BibTeX、RIS、EndNote 及 Zotero RDF | Zotero 同步 | 付费方案提供参考文献集成 | Pro 版支持 Zotero 和 Mendeley 同步 | Zotero、Mendeley 和 EndNote | BibTeX 文件 | 支持 BibTeX 文件 | BibTeX 文件 | 通过扩展支持 | 引文生成器 | 不支持文献库导入 | 否 | |
| 面向外部 AI 智能体的 MCP | 将 MCP 工具接入助手,或向外部智能体开放项目工具 | 否 | 否 | 否 | 否 | 否 | 否 | 否 | 是 | 否 | 否 | 否 | |
| 质量与投稿 | 投稿前检查 | 六项检查:编译、投稿、ATS、无障碍、参考文献、隐私 | 提供 AI 审阅,但无完整投稿检查套件 | 编译检查 | 编译器错误提示与 Assists | SCMS 合规工具 | 编译器日志与可选的 ChkTeX | 错误与警告 | 清晰的编译器错误提示 | 通过扩展进行代码检查 | 30 多项投稿检查 | 语言质量审查 | 仅提供简历检查 |
| 拼写与语法检查 | 私密的离线拼写与语法检查,并提供理解 LaTeX 的正文检查 | AI 校对 | 内置拼写检查,以及面向研究写作的 AI 反馈 | 官方资料未说明 | 官方资料未说明 | Hunspell、Aspell 与同义词词典 | 交互式拼写与语法检查 | 拼写检查 | 通过扩展或 AI 实现 | 学术语法与文风检查 | 学术语言反馈 | 基础检查能力因产品而异 | |
| ATS 简历检查 | 提供量化评分的 ATS 检查与文本提取预览 | 否 | 否 | 否 | 否 | 否 | 否 | 否 | 无内置检查 | 否 | 否 | 视产品而定,评分标准往往不透明 | |
| 无障碍预检 | 106 项 PDF/UA 检查,并为无障碍导出做好准备 | 否 | 否 | 官方资料未说明 | 结构化网页发布 | 支持结构化文档,但无预检 | 否 | 官方资料未说明 | 无内置检查 | 否 | 否 | 视产品而定 | |
| 盲审检查 | 标出作者姓名、致谢、PDF 元数据和注释 | 官方资料未说明 | 官方资料未说明 | 官方资料未说明 | 否 | 官方资料未说明 | 官方资料未说明 | 官方资料未说明 | 官方资料未说明 | 官方资料未说明 | 官方资料未说明 | 否 | |
| 引用与文件检查 | 未定义的引用、缺失文件、重复键、未被引用的条目 | AI 错误修复和 Paper Review | AI .bib 检查和论断检查 | 标签缺失时报编译错误 | DOI、链接、交叉引用和图片检查 | 标出失效的交叉引用 | 实时引用检查 | 仅有 LaTeX 错误列表 | ChkTeX 和 LaCheck 代码检查 | 30 多项检查,包括参考文献和图表 | 标出缺少引用的论断(机构版) | 否 | |
| 字数统计与文档分析 | 字数统计,以及标题、图、表、引用和 TODO | 官方资料未说明 | 是 | 官方资料未说明 | 摘要字数上限检查 | 是 | 是 | 是 | 是,通过 TeXcount | 官方资料未说明 | 官方资料未说明 | 简历分析评分 |
隐私,说清楚
默认本地。联网由你选择。
你的项目保存在你的电脑上。Oleafly 只在你选择的功能中使用网络,并会在发送崩溃报告前先向你展示。不需要 Oleafly 账号,也没有遥测。 隐私.
始终本地
断网也能用
- 项目文件、草稿、已保存的检查点和 Git 历史
- 写作、编译、PDF 审阅和预检
- 离线拼写检查、语法检查、图表,以及可选的 Ollama
仅在你操作时联网
仅在你点击时
- 托管的 AI 服务只会收到你选择发送的上下文
- GitHub 同步和文献检索仅在你发起时运行
- 宏包、模板、字体和应用更新按需下载
\[ v = \frac{V_{\max}[S]}{K_m + [S]} \]\[ p^2 + 2pq + q^2 = 1 \]理念
你的研究应当比创作它的工具更长久。你的研究应当比创作它的工具更长久。
\[ v = \frac{V_{\max}[S]}{K_m + [S]} \]\[ p^2 + 2pq + q^2 = 1 \]
认真做的开源
坚持独立,为社区打造。
Oleafly 公开开发。可靠性是产品的一部分:从可取消的编译、安全的文件路径,到 Git 恢复、PDF 回退,以及覆盖 macOS、Windows 和 Linux 的测试。
查看代码、报告问题、贡献修复或为 Oleafly 点 Star,欢迎前往 GitHub.
常见问题
切换前需要知道的事。
没找到答案?欢迎前往 Discord 或 GitHub 讨论区.
01Oleafly 是什么?+
Oleafly 是一个免费的桌面工作区,用于论文、学位论文、简历与其他技术文档。用 LaTeX、Typst 或 Markdown 写作,预览 PDF,管理参考文献,并可选用 AI 助手。你的项目以普通文件保存在你的电脑上。
02真的免费吗?+
是。Oleafly 免费且开源,桌面应用不另收订阅费。如果你连接付费 AI 提供商,该提供商可能按用量收费。通过 Ollama 使用的本地模型不需要付费 API。
03支持哪些 AI 模型?+
你可以为 10 家托管服务商中的任意一家提供密钥,通过 Ollama 在本地运行模型,或让 Oleafly 连接任何 OpenAI 兼容端点。它还能在编辑器内运行 14 种命令行智能体,包括 Claude Code 和 Codex,并使用你已有的工具和账号。在助手中选择运行时和模型。
04需要注册账号吗?+
不需要。安装 Oleafly 后,无需注册任何账号就能写作,项目以普通文件夹的形式保存在你的电脑上。只有确实需要登录的功能才要求登录:加入你的团队运行的协作服务器,或开启 Dropbox 同步。AI 服务商和 Git 托管平台使用各自的凭据。
05它可以取代哪些工具?+
对许多工作流而言,Oleafly 可以取代独立的写作、校对、编译、引用、PDF、简历、AI 与 Git 客户端。它照常使用普通文件与 GitHub,因此采用它不会把你的工作锁进专有格式。
06可以导入已有的论文或简历吗?+
可以。打开文件夹、导入 ZIP、从 GitHub 克隆,或直接从 arXiv 拉取论文。在线 LaTeX 编辑器的导出文件同样可用,Oleafly 会帮你找到主文档。基于文本的 PDF 可在本地转换为 LaTeX,Word 文件经 Pandoc 转换,视觉模型可以转录扫描件或照片。
07PDF 输出对 ATS 友好吗?+
ATS 就绪程度取决于模板与内容,因此 Oleafly 不承诺评分。其简历模板使用可选中的 Unicode 文本、嵌入字体、单栏版式与线性阅读顺序。预检会在你发送 PDF 前,展示解析器实际提取的内容。
08协作与同步呢?+
多人实时编辑同一个项目,光标旁显示姓名,评论就挨着所讨论的那一行。你的团队运行自己的 Oleafly Server,每个人都在自己的电脑上以普通文件保存项目,因此可以离线继续写作,重新连接后编辑会自动同步。Dropbox 同步让你的项目库和设置在自己的各台设备之间保持一致。Git 照样可以配合使用:发布到 GitHub,然后在 Oleafly 内推送和拉取。
09可以离线写作或不用 AI 吗?+
可以。写作、本地编译、PDF 审阅与校对均不需要 AI。离线前,请先下载项目所需的宏包、字体与工具。本地 AI 方面,在 Ollama 中下载模型并连接其本地服务器即可。
10我继续写作时,代理能同时工作吗?+
可以。打开研究任务,写下任务说明并选择一个代理。它在隔离副本中工作,你继续写论文。之后跟进活动动态,预览更改的文件,并应用你选择的结果。
\[ E = h\nu \]