torch
‹0.17.0
- 0.17.011 Apr 2026, current
- 0.16.32 Nov 2025
- 0.16.231 Oct 2025
- 0.16.116 Oct 2025
- 0.16.021 Aug 2025
- 0.15.110 Jul 2025
- 0.15.023 Jun 2025
- 0.14.214 Feb 2025
- 0.14.13 Feb 2025
- 0.14.030 Jan 2025
- 0.13.022 May 2024
- 0.12.015 Dec 2023
- 0.11.06 Jun 2023
- 0.10.014 Apr 2023
- 0.9.123 Jan 2023
- 0.9.024 Oct 2022
- 0.8.119 Aug 2022
- 0.8.010 Jun 2022
- 0.7.226 Feb 2022
- 0.7.125 Feb 2022
- 0.7.018 Feb 2022
- 0.6.07 Oct 2021
- 0.5.017 Aug 2021
- 0.4.010 Jun 2021
- 0.3.028 Apr 2021
- 0.2.15 Jan 2021
- 0.2.015 Dec 2020
- 0.1.120 Oct 2020
- 0.1.028 Sep 2020
- 0.0.318 Sep 2020
- 0.0.231 Aug 2020
- 0.0.16 Aug 2020
Tensors and Neural Networks with 'GPU' Acceleration
- CRAN checks, method
Method
Checks and health
CRAN runs every current package through
R CMD checkon a set of machines and publishes a result for each one. The health shown on a package page comes from these results alone.- Problems
- Notes alone never count against a package, but a warning, an error or a failure on any platform is a problem. Open CRAN deadlines are a fact of their own, shown as a clock after the mark, so the mark always says what the checks found.
- Checks only
- Health reflects CRAN's check results and nothing else. Even a package with no tests, no coverage or no recent release can have every platform OK.
- Once a day
- Results are collected once a day from files CRAN publishes. Each platform is one machine set-up, named by its R version, operating system, architecture and compiler, such as r-devel-linux-x86_64-fedora-clang.
- Platforms counted
- The 13 in 3 of 13 is the number of platform rows CRAN published for that package on the day of collection. The set of machines changes over time, so the number is not fixed, and a package missing from a platform counts as neither passing nor failing there.
- OKnotes on 2 of 13
- CRAN 30-day downloads, method
- 43,721−4% down on the 30 days before, the worse direction
- Dependent packages, method
Method
Reverse dependencies
Dependent packages are other packages that list this one under
Depends,ImportsorLinkingToin theirDESCRIPTIONfile. Those fields list the packages a package needs to install, so dependents break if this one is archived or broken.- As declared
- The counts follow the declarations as written, not whether a package uses the other one when it runs.
- Suggests apart
- Packages that list this one only under
Suggestsinstall and run without it, and use it for tests, examples or an optional feature. They are counted as Suggested by, never among the dependents, and a package that also lists it under one of the three fields is a dependent and is not counted again. - Zero
- Zero dependents means no current CRAN package declares a need for this one, though some may suggest it. On 27 Oct 2026, 19,551 of the 25,555 current CRAN packages had no dependent package.
- 5332 more suggest it
- Test coverage, method
Method
Test coverage
Test coverage is measured by downloading each CRAN package's current source file, installing its dependencies, building it and running its tests under covr.
- Almost half of CRAN has no tests to run
- Only CRAN packages are measured. Many have no percentage, most often because the package has no tests, and that blank is a recorded outcome, not missing data.
- NOT_CRAN
- Tests run with NOT_CRAN=true, so tests that skip themselves on CRAN run here, and the figure can differ from what CRAN's own checks see.
- Current version only
- Only the current version is measured, and a new version on CRAN is measured again. Older rows are kept but never filled in later. This is one package's run.
- Never 0%
- On a package page three states stay apart: measured, no tests, and not measured. Neither of the last two is ever shown as 0%.
- 0.3%of lines run by tests
- Advisories, method
Method
Security advisories
Advisories come from the R Consortium advisory database, a community record of security problems in R packages, as OSV publishes it.
- The copy
- The record is read once a day. The data is published under the Apache License 2.0 and is credited on the advisories list.
- Box under the name
- Under the package's name, a box names any advisory that affects the current release, with its fix or that there is none.
- Ranges
- Advisories state ranges, such as from 0.9.12 up to but not including 1.8.8. Each range is checked against every version CRAN has published, using the version rules R itself uses, so a release that came out after the advisory was written is still caught by its range.
- Not the package
- The rating is for the library flaw itself, not for how exposed the R package is, and it is never added up into a figure for the package.
- None known
About
Method, About Method Licence figures come from the Licences
License field of each package's DESCRIPTION file, read the way R itself reads it. No licence file is opened.tools::analyze_license() and R's own licence database, as CRAN's checks read it.| is read, so one string can offer two families.
Provides functionality to define and train neural networks similar to 'PyTorch' by Paszke et al (2019) <doi:10.48550/arXiv.1912.01703> but written entirely in R using the 'libtorch' library. Also supports low-level tensor operations and 'GPU' acceleration.
- Maintainer
- Tomasz Kalinowski
- Licence
- MIT + file LICENSE
- Latest release
- 11 Apr 2026
- First release
- 6 Aug 2020
More details
- Releases
- 32
- Needs compilation
- Yes
- Config/build/copy-method
- copy
- RoxygenNote
- 7.3.3
- SystemRequirements
- LibTorch (https://pytorch.org/); Only x86_64 platforms are currently supported except for ARM system running macOS.
Task views (1)
Collections (1) Method, CollectionsMethod
Subjects and collections
Subjects are the site's own grouping of CRAN packages, found by clustering. Collections are sets of packages that communities declare, such as the tidyverse.
- Subjects
- Subjects are found by clustering the words of current CRAN packages and the packages they use. No one edits them by hand.
- Collections
- Collections are sets that communities declare, through a package that installs the set or an organization on GitHub. The site finds them by rules.
- Who made each
- Subjects are marked Site and collections Community. Subjects, collections and task views are never mixed in one list.
- The day of the data
- Subjects and collections were last worked out on 22 Oct 2026.
- mlverse, in the mlverse organization on GitHub
Subjects (1) Method, Subjects
- Neural network models, core
Cite
Package
Run in R for the authors' preferred citation:
citation("torch")Kalinowski, T., Badr, H. S., Damiani, A., Falbel, D., Fischer, S., Joachimiak, K., Luraschi, J., Pichler, M., RStudio, , Regouby, C., & Selivanov, D. (2026). torch: Tensors and Neural Networks with 'GPU' Acceleration (Version 0.17.0) [Computer software]. https://doi.org/10.32614/CRAN.package.torch
About
This package has no citation file, so citation() returns this entry built from DESCRIPTION.
R Observatory
For a number measured here: a download total, a coverage figure, an archival date.
Balamuta, J. J. (2026). R Observatory: Metrics for torch version 0.17.0 [Data set]. HJJB, LLC. Data release v2026-10-27. https://doi.org/10.5281/zenodo.21843040
About
From data release v2026-10-27, which the citation names so these numbers can be found later.
Health
No warnings or errors. 2 of 13 check platforms have notes.
r-devel-linux-x86_64-debian-clangThe unreleased R on Linux, x86_64, debian-clang
Note
- Version checked
- 0.17.0
- Time taken
- 1,021 s
Rd contents: Note
Rd files without \usage:
r-devel-linux-x86_64-debian-gccThe unreleased R on Linux, x86_64, debian-gcc
Note
- Version checked
- 0.17.0
- Time taken
- 905 s
Rd contents: Note
Rd files without \usage:
r-devel-linux-x86_64-fedora-clangThe unreleased R on Linux, x86_64, fedora-clang
OK
- Version checked
- 0.17.0
- Time taken
- 652 s
r-devel-linux-x86_64-fedora-gccThe unreleased R on Linux, x86_64, fedora-gcc
OK
- Version checked
- 0.17.0
- Time taken
- 985 s
r-devel-windows-x86_64The unreleased R on Windows, x86_64
OK
- Version checked
- 0.17.0
- Time taken
- 1,678 s
r-oldrel-macos-arm64The previous release of R on macOS, arm64
OK
- Version checked
- 0.17.0
- Time taken
- 188 s
r-oldrel-macos-x86_64The previous release of R on macOS, x86_64
OK
- Version checked
- 0.17.0
- Time taken
- 1,116 s
r-oldrel-windows-x86_64The previous release of R on Windows, x86_64
OK
- Version checked
- 0.17.0
- Time taken
- 2,130 s
r-patched-linux-x86_64The patched release of R on Linux, x86_64
OK
- Version checked
- 0.17.0
- Time taken
- 1,286 s
r-release-linux-x86_64The current release of R on Linux, x86_64
OK
- Version checked
- 0.17.0
- Time taken
- 1,258 s
r-release-macos-arm64The current release of R on macOS, arm64
OK
- Version checked
- 0.17.0
- Time taken
- 177 s
r-release-macos-x86_64The current release of R on macOS, x86_64
OK
- Version checked
- 0.17.0
- Time taken
- 920 s
r-release-windows-x86_64The current release of R on Windows, x86_64
OK
- Version checked
- 0.17.0
- Time taken
- 1,654 s
- 2 Note
- 11 OK
Check output (1)
Rd contents, Note on 2 platforms
Rd files without \usage: ‘torch_argmax.Rd’ ‘torch_argmin.Rd’ ‘torch_bincount.Rd’ ‘torch_eig.Rd’ ‘torch_logical_not.Rd’ ‘torch_lstsq.Rd’ ‘torch_matrix_rank.Rd’ ‘torch_max.Rd’ ‘torch_min.Rd’ ‘torch_sort.Rd’ \arguments should not be documented without \usage.
Extra checks (1)
Over time
No change since 11 Jul 2026
As a list (9)
| Changed | Worst status | OK | Notes | Warnings | Errors |
|---|---|---|---|---|---|
| 11 Jul 2026 | Note | 12 | 1 | 0 | 0 |
| 8 Jun 2026 | OK | 12 | 0 | 0 | 0 |
| 7 Jun 2026 | Error | 12 | 0 | 0 | 1 |
| 2 May 2026 | OK | 13 | 0 | 0 | 0 |
| 22 Apr 2026 | Note | 11 | 3 | 0 | 0 |
| 18 Apr 2026 | Error | 10 | 3 | 0 | 1 |
| 16 Apr 2026 | Note | 11 | 3 | 0 | 0 |
| 21 Mar 2026 | Warning | 6 | 7 | 1 | 0 |
| 10 Mar 2026 | Note | 11 | 3 | 0 | 0 |
Since March 2026
Downloads
Method, Downloads Method Download figures count requests logged by each distribution channel. They are not counts of installs or of users, because each channel publishes its own totals and this site reads them.Downloads
- Total downloads in the last 30 days
- 44,930−2.7% down on the 30 days before, the worse directionIt is the sum of 5 channels: CRAN, r2u, autoCRAN, conda-forge and c2d4u.
CRAN rank
527 of 26,529
CRAN
43,721 downloads in the last 30 days
- Rank
- 527 of 26,529
- Change
- −4% down on the 30 days before, the worse direction
It is in the total. Take it out
r2u
485 downloads in the last 30 days
- Rank
- 1,178 of 30,526
- Change
- +29.7% up on the 30 days before, the better direction
It is in the total. Take it out
autoCRAN
6 downloads in the last 30 days
- Rank
- 8,099 of 29,212
- Change
- −70% down on the 30 days before, the worse direction
It is in the total. Take it out
conda-forge
682 downloads in the last 30 days
- Rank
- 1,263 of 3,828
- Change
- +275% up on the 30 days before, the better direction
It is in the total. Take it out
c2d4u, no longer updated
36 downloads in the last 30 days
- Rank
- 580 of 7,575
- Change
- no change on the 30 days before
It is in the total. Take it out
The total is 44,930 downloads in the last 30 days. It is the sum of 5 channels: CRAN, r2u, autoCRAN, conda-forge and c2d4u.
Over time
Dependencies
- Needs
- 18
- Dependent packages
- 53
- Suggested by
- 32
Needs (18)
Dependent packages (50 of 53)
| Package | Title | Downloads in 30 days |
|---|---|---|
| mlr3torch | Deep Learning with 'mlr3' | 2,970 |
| torchvision | Models, Datasets and Transformations for Images | 2,726 |
| cito | Building and Training Neural Networks | 2,523 |
| tabnet | Fit 'TabNet' Models for Classification and Regression | 2,236 |
| torchvisionlib | Additional Operators for Image Models | 1,648 |
| luz | Higher Level 'API' for 'torch' | 1,583 |
| brulee | High-Level Modeling Functions with 'torch' | 866 |
| PLNmodels | Poisson Lognormal Models | 701 |
| glmnetr | Nested Cross Validation for the Relaxed Lasso and Other Machine Learning Models | 687 |
| sits | Satellite Image Time Series Analysis for Earth Observation Data Cubes | 631 |
| innsight | Get the Insights of Your Neural Network | 491 |
| causalOT | Optimal Transport Weights for Causal Inference | 486 |
| torchdatasets | Ready to Use Extra Datasets for Torch | 417 |
| BKTR | Bayesian Kernelized Tensor Regression | 407 |
| ARInfoLSTM | ARIMA-Informed LSTM for Time Series Forecasting | 368 |
| engression | Engression Modelling | 364 |
| proteus | Multiform Seq2Seq Model for Time-Feature Analysis | 363 |
| GARCHInfoLSTM | GARCH-Informed LSTM Model for Volatility Forecasting | 362 |
| ria.test | Testing Equality of Natural Mediation Effects and Their Randomized Interventional Analogues | 351 |
| pigauto | Fill in Missing Species Traits Using a Phylogenetic Tree | 343 |
| EQRN | Extreme Quantile Regression Neural Networks for Risk Forecasting | 319 |
| madgrad | 'MADGRAD' Method for Stochastic Optimization | 302 |
| funcml | Functional Machine Learning Framework | 296 |
| unfold | Mapping Hidden Geometry into Future Sequences | 289 |
| survdnn | Deep Neural Networks for Survival Analysis with R 'torch' | 283 |
| lambdaTS | Variational Seq2Seq Model with Lambda Transformer for Time Series Analysis | 282 |
| torchMAUM | Multi-Class Area Under the Minimum in Torch | 274 |
| scDHA | Single-Cell Decomposition using Hierarchical Autoencoder | 271 |
| FINN | Forest Informed Neural Networks | 270 |
| topicmodels.etm | Topic Modelling in Embedding Spaces | 269 |
| whisper | Native R 'torch' Implementation of 'OpenAI' 'Whisper' | 266 |
| torchopt | Advanced Optimizers for Torch | 265 |
| LBBNN | Latent Binary Bayesian Neural Networks Using 'torch' | 257 |
| kindling | Higher-Level Interface of 'torch' Package to Auto-Train Neural Networks | 256 |
| chatterbox | Text-to-Speech Using the 'Chatterbox' Engine | 255 |
| crumble | Flexible and General Mediation Analysis Using Riesz Representers | 250 |
| rinet | Clinical Reference Interval Estimation with Reference Interval Network (RINet) | 249 |
| shrinkGPR | Scalable Gaussian Process Regression with Hierarchical Shrinkage Priors | 237 |
| snap | Simple Neural Application | 231 |
| diffuseR | Functional Interface to Diffusion Models in R | 229 |
| survalis | Interpretable Survival Machine Learning Framework | 223 |
| spinner | An Implementation of Graph Net Architecture Based on 'torch' | 219 |
| circlus | Clustering and Simulation of Spherical Cauchy and PKBD Models | 218 |
| n3d | Speaker Diarization with 'Nemotron 3 Diarization' in R 'torch' | 218 |
| rkaf | Kolmogorov-Arnold Fourier Networks in R | 216 |
| SEMdeep | Structural Equation Modeling with Deep Neural Network and Machine Learning Algorithms | 210 |
| DICErClust | Deep Significance Clustering for Clinical Risk Stratification | 187 |
| temper | Temporal Encoder-Masked Probabilistic Ensemble Regressor | 186 |
| RGAN | Generative Adversarial Nets (GAN) in R | 185 |
| NNMoMo | Neural Network Extension to 'StMoMo' for Lee-Carter Modeling | 181 |
Code
Method, Code Method Code metrics come from reading the released source of each package version. The package is never installed, loaded or run.Code metrics
README, vignettes, a NEWS file and the shares read from its help pages. Files kept only in the repository, such as a pkgdown configuration or CI files, are read by the repository scan instead.
- Functions
- 15,072
- Public exports
- 755
- Lines of R code
- 102,571
- Median complexity
- 1
Functions (15,072)
Size and composition
Lines by kind (228,721)
| Kind | Lines | Share |
|---|---|---|
| R code | 102,571 | 45% |
| Tests | 11,858 | 5% |
| Documentation | 30,397 | 13% |
| Guides | 3,872 | 2% |
| Compiled code | 80,023 | 35% |
Lines by language
| Language | Lines | Share |
|---|---|---|
| R | 114,074 | 50% |
| C++ | 80,013 | 35% |
| Rd | 30,345 | 13% |
| Vignettes | 3,924 | 2% |
| Markdown | 355 | 0.2% |
| C | 10 | 0% |
Languages are counted from file extensions.
- Files
- 1,037
- R functions
- 6,958
- Compiled functions
- 8,122
- Object systems
- S3: 131 methods, R6: 59 classes
Complexity
- Median
- 1
- Mean
- 1.14
- Highest
- 155
- Median lines per function
- 3
About complexity
Cyclomatic complexity counts the independent paths through a function: 1 is a straight line, higher is harder to test. R functions only.
Call graph figures
- R functions
- 6,958
- Calls
- 3,361
- Unconnected
- 3,916
- Groups
- 51
Growth (32 releases)
Release changes
Tests
The tests ran and coverage was measured.
- Lines run by tests
- 0.3%
- Expressions run by tests
- 0.5%
- Test cases
- 735
- Exports named in tests
- 46%
Coverage by file (96)
Coverage run
- Lines counted
- 35,331
- Lines the tests ran
- 0
- Tests passed
- Yes
- Measured with
- covr 3.6.5 on R 4.6.1
Test suite
- Has tests
- Yes
- Test frameworks
- testthat, tinytest
- Test cases
- 735
- Snapshot tests
- 15
- Lines of tests
- 11,858
- Test lines per line of code
- 0.12
Documentation
- Exports documented
- 99.9%
- Arguments documented
- 100%
- Return value documented
- 6.2%
- Help pages with examples
- 63%
- Examples run in checks
- 100%
Guides (1)
Other documentation
- README
- Yes, 117 words
- Release notes file
- Yes
- Help pages
- 702
- Lines of documentation
- 30,397
People
Method, People Method People on the site are the names packages give them in their Author and maintainer identity
DESCRIPTION files: authors from the author list, and the maintainer from the Maintainer field. Spellings of one name become one person only where the rules below allow it, and when in doubt the rules keep people apart.
Maintainer since 0.17.0 (2026), previously Daniel Falbel (0.0.1 to 0.16.3), contributor
Contributor
Contributor
Author, copyright holder
Contributor
Contributor
Author
Contributor
Copyright holder
Contributor
Contributor
Repository
Method, figures by host Method Repository signals are the figures read from the source repository a package links to: stars, forks, issues, commits, releases and the time issues take to close. They are read from GitHub, GitLab and Codeberg.Repository signals
URL and BugReports fields of its DESCRIPTION file. Read-only mirrors, the cran and bioc accounts on GitHub and git.bioconductor.org, are left out.
- Stars
- 568−0.18% down on 30 days ago, the worse direction
- Forks
- 93
- Open issues
- 135
- Last commit
- 23 Oct 2026
Activity
- Commits
- 4,561
- Open pull requests
- 13
- Tagged releases
- 32
- Latest tagged release
- v0.17.0, 11 Apr 2026
- In development
- Default branch at 0.17.0.9000, ahead of CRAN 0.17.0 (checked 2026-10-25)
- Issues closed in
- 14 days (median)
- Pull requests closed in
- 1 day (median)
- Pull requests merged
- 93%
- Created
- 7 Jan 2020
- Licence file reads as
- –
Development practices (10)
| Kind | Found |
|---|---|
| Continuous integration | GitHub Actions, R CMD check workflow The R CMD check workflow runs on Linux, macOS and Windows |
| Governance and community | Contributing guide |
| Docs source | README.Rmd, pkgdown GitHub Pages site last deployed 2026-10-23 The repository has a site deploy workflow pkgdown site last built 2026-04-11 |
| CRAN release process | cran-comments.md, revdep/ |
| Lint, format, editor | RStudio project |
Method, repository practices Method Beta Each week a scan of the public source repository a package links to looks for the tools a maintainer has set up, and the Repository section lists the findings. Repositories on GitHub and GitLab are scanned.Repository practices
R CMD build reads .Rbuildignore and leaves out files used only in development, such as .github/, .lintr, renv/, a Dockerfile and cran-comments.md, so the released source file holds almost none of them. The repository is the one place these files stay visible.
AI tools
Method, AI tools Method Beta What is found in a repository is not a measure of the code.AI tool findings
Tools
| Tool | First seen | Found | Commits by it | Commits crediting it |
|---|---|---|---|---|
| Claude Code | 26 Jan 2026 | Seen in the repository.claude.claude/skillsNamed in .Rbuildignore.claudePull request it wrote: opened by a project member | 10 at least | |
| GitHub Copilot | 26 Jan 2026 | Pull request: opened by its own account | 1 | 1 at least |
- Seen in the repository
- Named in .Rbuildignore
- Pull request
Models
| Model | Context window | Commits | First commit | Last commit | Credit lines of |
|---|---|---|---|---|---|
| Claude Opus 5.5 | 2 | 14 Oct 2026 | 22 Oct 2026 | Claude Code | |
| Claude Opus 5 | 1M | 2 | 5 Aug 2026 | 5 Aug 2026 | Claude Code |
| Claude Opus 4.5 | 6 | 26 Jan 2026 | 8 Apr 2026 | Claude Code |
History
Method, History Method Archived is built from CRAN's archive records, its list of check deadlines and Bioconductor's package records. Every figure is counted to 27 Oct 2026, the day the data was read.Packages leaving
Archived and back
| Archived | Back | CRAN's reason | Kind |
|---|---|---|---|
| 13 Sep 2020 | 18 Sep 2020 | check issues still had not been corrected | Checks not fixed |
| 26 Aug 2020 | 31 Aug 2020 | check issues were not corrected in time. Installation failures with parallel make | Checks not fixed |
| Version | Released | R | Events |
|---|---|---|---|
| 0.17.0 current | 11 Apr 2026 | 4.5.3 | Maintainer: Tomasz Kalinowski |
| 0.16.3 | 2 Nov 2025 | 4.5.2 | |
| 0.16.2 | 31 Oct 2025 | 4.5.2 | |
| 0.16.1 | 16 Oct 2025 | 4.5.1 | |
| 0.16.0 | 21 Aug 2025 | 4.5.1 | |
| 0.15.1 | 10 Jul 2025 | 4.5.1 | |
| 0.15.0 | 23 Jun 2025 | 4.5.1 | |
| 0.14.2 | 14 Feb 2025 | 4.4.2 |
Earlier versions (24)
| Version | Released | R | Events |
|---|---|---|---|
| 0.14.1 | 3 Feb 2025 | 4.4.2 | |
| 0.14.0 | 30 Jan 2025 | 4.4.2 | |
| 0.13.0 | 22 May 2024 | 4.4.0 | |
| 0.12.0 | 15 Dec 2023 | 4.3.2 | |
| 0.11.0 | 6 Jun 2023 | 4.3.0 | |
| 0.10.0 | 14 Apr 2023 | 4.2.3 | |
| 0.9.1 | 23 Jan 2023 | 4.2.2 | |
| 0.9.0 | 24 Oct 2022 | 4.2.1 | |
| 0.8.1 | 19 Aug 2022 | 4.2.1 | |
| 0.8.0 | 10 Jun 2022 | 4.2.0 | |
| 0.7.2 | 26 Feb 2022 | 4.1.2 | |
| 0.7.1 | 25 Feb 2022 | 4.1.2 | |
| 0.7.0 | 18 Feb 2022 | 4.1.2 | |
| 0.6.0 | 7 Oct 2021 | 4.1.1 | |
| 0.5.0 | 17 Aug 2021 | 4.1.1 | |
| 0.4.0 | 10 Jun 2021 | 4.1.0 | |
| 0.3.0 | 28 Apr 2021 | 4.0.5 | |
| 0.2.1 | 5 Jan 2021 | 4.0.3 | |
| 0.2.0 | 15 Dec 2020 | 4.0.3 | |
| 0.1.1 | 20 Oct 2020 | 4.0.3 | |
| 0.1.0 | 28 Sep 2020 | 4.0.2 | |
| 0.0.3 | 18 Sep 2020 | 4.0.2 | Back on CRAN 18 Sep 2020 |
| 0.0.2 | 31 Aug 2020 | 4.0.2 | Back on CRAN 31 Aug 2020 Archived 13 Sep 2020 “check issues still had not been corrected” |
| 0.0.1 | 6 Aug 2020 | 4.0.2 | Archived 26 Aug 2020 “check issues were not corrected in time. Installation failures with parallel make” |
Release notes (0.17.0)
Breaking changes
- Updated to LibTorch 2.8.0 (#1419).
torch_triu_indices()andtorch_tril_indices()now return 1-based indexes (#1382).$indices()now returns 1-based indexes (#1382).
New features
- Added
torch_sitrep()for torch installation situation report (#1415, @cregouby). - Added
%*%operator for torch tensors (#1379, @caio-hamamura). - Added
head()andtail()methods for torch tensors (#1388). - Exported
torch_scaled_dot_product_attention()(#1404). - Exported
torch_sparse_sampled_addmm()(#1427). - Exported
torch_ldexp()(#1407). - Integrated
cudatoolkitR package for CUDA library loading (#1422). - Improved
as_array()support for sparse tensors (#1387).
Bug fixes
- Fixed sparse CSR tensor printing,
to_sparse()crash, andas_array()detection (#1431). - Fixed
dataset()/nn_module()crash wheninitializeparameter is namedd(#1424, @Chandraveer-Singh). - Fixed
bfloat16not being recognized as a floating point dtype (#1408). - Fixed handling of R longjump exceptions in autograd and tracing callbacks (#1406).
- Fixed redundant
lantern.hinclude causing macro redefinition warning (#1430). - Excluded glibc libraries from dependency bundling (#1397, @troyhernandez).
- Replaced non-API entry point
Rf_findVarInFramewithR_getVarExfor R 4.6 compatibility (#1421). - Added French translations for 5 missing messages (#1385, @cregouby).