🌟 Summary

THOP 2.2.1 improves operation-count estimates for models with adaptive max pooling or bilinear layers, without changing reported counts for Ultralytics’ buildable model configurations.

📊 Key Changes

• Fixed stride-based profiling: Adaptive max pools and nn.Bilinear are now treated as fixed-size operations during extrapolation.
• Released as ultralytics-thop 2.2.1: Publishes the profiling fix from PR #181.
• Validated existing model counts: Ultralytics model FLOPs remained bit-identical across all buildable configuration files.

🎯 Purpose & Impact

• 📏 Prevents FLOPs from being overcounted when profiling models with these layers.
• 🔍 Makes complexity estimates more reliable for users comparing or evaluating affected models.
• ✅ Does not change model architectures or the reported FLOPs for the tested Ultralytics configurations.

What's Changed

• Refresh AGENTS.md and fix stride extrapolation after adaptive max pools by @glenn-jocher in #181
• Release 2.2.1 by @glenn-jocher in #182

Full Changelog: v2.2.0...v2.2.1

🌟 Summary

THOP 2.2.0 improves profile() so it can count more attention-related operations and estimate model cost from smaller inputs more reliably—reducing the need for custom counting rules or full-size profiling runs.

📊 Key Changes

• Counts functional matrix products: On PyTorch 1.13 and later, profile() now detects operations such as @, torch.matmul, torch.bmm, and—when available—scaled_dot_product_attention. Products inside a module with its own counting rule are not counted again.
• Improves stride-based estimates: Models that may have quadratic costs, such as attention-based models or models using custom operation rules, are profiled at several smaller sizes. THOP checks whether those measurements support a quadratic estimate and profiles at the target size if they do not.
• Adds min_cells: This new profile() argument lets users start proxy measurements at a larger size, which can help with models whose costs change behavior below a certain input size.
• Updates guidance and maintenance details: The English and Chinese READMEs now describe the profiling behavior, and repository instructions and CLA workflow handling were corrected. Documentation links were also cleaned up.

🎯 Purpose & Impact

• More complete operation counts: Functional attention and matrix products that module hooks could miss are now included, giving users a clearer picture of model compute.
• More useful estimates with safeguards: Smaller proxy inputs can speed up profiling, while validation and target-size fallbacks help avoid unreliable extrapolations.
• Less manual setup: Users may no longer need to write their own rules just to count common functional matrix products.
• No model architecture changes: This release focuses on THOP’s profiling behavior, documentation, and development workflows.

What's Changed

• Correct AGENTS.md Ruff target version and add intro by @glenn-jocher in #175
• Correct inaccurate AGENTS.md claims by @glenn-jocher in #176
• Remove trailing slashes and update redirected Ultralytics URLs by @pderrenger in #177
• Accept normalized CLA signature comments by @glenn-jocher in #178
• Document the three-number PyPI release policy by @glenn-jocher in #179
• Count functional products and extrapolate quadratic costs from stride proxies by @glenn-jocher in #180

Full Changelog: v2.1.6...v2.2.0

🌟 Summary

🚀 THOP 2.1.6 is a patch release that improves model-operation profiling accuracy and robustness while preserving the existing API and behavior for typical users.

📊 Key Changes

• 📦 Released THOP 2.1.6

  • Updated the package version from 2.1.5 to 2.1.6.
  • No intentional breaking API changes.

• 🧮 More accurate recurrent-layer profiling

  • Packed RNN, GRU, and LSTM inputs are now counted using their actual processed sequence steps instead of the larger padded shape.
  • LSTM projection layers now account for proj_size, including the additional projection work and reduced recurrent width.
  • This produces more realistic operation estimates for variable-length sequences and projected LSTMs.

• 🧩 Improved custom operation support

  • profile_origin() now applies custom profiling rules to composite modules, matching the behavior of the standard profile() function.
  • Custom operations that may have nonlinear computational costs now use direct profiling instead of inaccurate linear extrapolation.

• 🧱 Better handling of container and zero-operation modules

  • Added support for containers such as ModuleList, ModuleDict, ParameterList, and ParameterDict.
  • Max-unpool layers are recognized as performing no arithmetic operations.
  • Embedding lookups are correctly treated as data gathering rather than arithmetic computation.
  • Embedding layers using max_norm emit a warning because their data-dependent renormalization is not counted.
  • EmbeddingBag remains unregistered because its reduction cost depends on the input contents.

• 🛡️ Safer total_ops handling

  • THOP now rejects modules that already use total_ops for parameters, child modules, or class-level properties.
  • This prevents profiling from silently corrupting module structure, parameter counts, or reported operation totals.

• ✅ Validation

  • The complete test suite passed: 20 tests successful.

🎯 Purpose & Impact

• 📈 More trustworthy profiling results for packed sequences, projected LSTMs, custom modules, and complex model containers.
• ⚙️ Fewer false warnings when profiling models that use common PyTorch containers or embedding layers.
• 🧠 Better support for advanced architectures with custom operation rules and nonlinear computation patterns.
• 🔒 Improved reliability and error reporting when a model conflicts with THOP’s internal total_ops bookkeeping.
• 👍 For most users, this is a drop-in upgrade with no code changes required.

What's Changed

• Count a packed sequence by its real steps and read the LSTM projection by @raimbekovm in #160
• Let profile_origin count a composite module's own rule by @raimbekovm in #168
• Register the container and max-unpool types that compute nothing by @raimbekovm in #161
• Stop extrapolating a quadratic custom_ops rule as if it were affine by @raimbekovm in #167
• Register the embedding lookup and record why the bag is not counted by @raimbekovm in #165
• Reject a total_ops name a module already owns instead of misreading it by @raimbekovm in #171
• Release 2.1.6 by @glenn-jocher in #174

Full Changelog: v2.1.5...v2.1.6

🌟 Summary

THOP 2.1.5 improves PyTorch compatibility by making attention-operation profiling safe across both PyTorch 1.x and 2.x versions. 🛠️

📊 Key Changes

• Fixed nn.MultiheadAttention profiling on PyTorch 1.x
  Attention profiling is now enabled only when PyTorch forward hooks can expose keyword arguments.

• Preserved accurate attention counting on PyTorch 2.x
  Newer PyTorch versions continue to report attention MACs correctly, including mixed cross-attention scenarios.

• Avoided crashes on older PyTorch versions
  When hook inputs cannot be reconstructed, THOP now skips attention MAC counting rather than attempting an unsafe call.

• Updated the package version
  The release version was bumped from 2.1.4 to 2.1.5.

• Validated across profiling APIs ✅
  Tests confirmed accurate results on PyTorch 2.x and safe zero-count behavior under PyTorch 1.x compatibility conditions.

🎯 Purpose & Impact

• Improves backward compatibility for users running models with PyTorch 1.x.
• Prevents profiling-related runtime errors without changing model execution.
• Maintains modern profiling accuracy for PyTorch 2.x users.
• Users should receive a more reliable experience when using THOP to estimate model complexity, especially for models containing multi-head or cross-attention layers. 📈

What's Changed

• Keep attention profiling compatible with torch 1.x by @glenn-jocher in #159

Full Changelog: v2.1.4...v2.1.5

🌟 Summary

THOP 2.1.4 now accurately counts nn.Bilinear operations instead of treating them as free, improving model complexity and compute reporting. 🎯

📊 Key Changes

• Added nn.Bilinear support to THOP’s operation-counting registry.
• Introduced a dedicated count_bilinear hook that accounts for both contractions performed by PyTorch.
• Counts operations across standard, batched, and higher-dimensional inputs using the output tensor size.
• Updated the package version from 2.1.3 to 2.1.4.
• Change contributed by @raimbekovm in PR #158.

🎯 Purpose & Impact

• More accurate profiling: Models using bilinear layers no longer report artificially low operation counts.
• Better comparisons: FLOPs/MACs estimates now better reflect the actual computational cost of architectures such as attention, fusion, and multimodal networks.
• Consistent behavior: The implementation follows THOP’s existing linear-layer counting convention without adding shape-specific branches.
• Improved resource planning: More reliable estimates can help users compare models and anticipate inference or training costs. 🚀

What's Changed

• Count nn.Bilinear instead of charging it nothing by @raimbekovm in #158

Full Changelog: v2.1.3...v2.1.4

🌟 Summary

🚀 THOP 2.1.3 delivers more accurate PyTorch operation profiling, especially for attention-based models, while improving pooling, recurrent layers, custom rules, and profiler reliability.

📊 Key Changes

• 🧠 Accurate nn.MultiheadAttention counting (PR #157)
  Attention is no longer reported as free. THOP now counts its projection layers, attention matrix operations, and softmax processing, including support for:

  • Batched and unbatched inputs
  • Cross-attention
  • Different key and value dimensions
  • Optional bias and zero-attention tokens
  • Calls with or without returned attention weights
  • Positional, mixed, and keyword arguments

• 📈 More realistic attention-model estimates
  Because attention cost does not scale linearly with image area, stride-based shortcut estimation is disabled for models containing attention. These models now use the exact target-size profiling path. For example, measured RT-DETR estimates increased by roughly 1.5–1.8%, reflecting previously uncounted work.

• 🧮 Improved average-pooling operation counts (PR #150)
  Fixed and adaptive average pooling now account for actual window sizes, padding, boundary behavior, ceil_mode, and count_include_pad. This avoids assuming every output uses a full pooling window.

• 🧱 Better support for unbatched and recurrent inputs (PR #150)
  Convolution and RNN/GRU/LSTM profilers now correctly interpret channel, batch, and sequence dimensions for both batched and unbatched layouts.

• ⚡ Expanded zero-operation layer coverage (PR #152)
  Data movement and selection layers such as Flatten, Unflatten, Identity, PixelUnshuffle, additional padding layers, and fractional max pooling are explicitly registered as zero-MAC operations. This reduces unnecessary missing-rule warnings and improves profiling speed.

• 🔥 Complete softmax-family support (PR #153)
  LogSoftmax, Softmin, and Softmax2d are now counted, including their additional elementwise work and correct handling of empty normalization dimensions.

• 🛡️ Safer handling of caller-owned total_ops values (PR #151)
  Existing total_ops attributes and buffers are temporarily preserved during profiling and restored afterward, preventing THOP from overwriting model state.

• 🌳 Profiler results now reflect the modules that actually ran (PR #154)
  Both profiler entry points consistently count the forward pass that executed, even when a model replaces or detaches submodules during inference. Training modes are also restored more reliably.

• 🧩 More flexible custom operation rules (PRs #155–#156)
  Custom counting rules can now be callable objects or functools.partial instances, and they receive arguments supplied through keyword, positional, or mixed calls.

• 📦 Version updated to 2.1.3

🎯 Purpose & Impact

• ✅ More trustworthy FLOPs/MACs reports, particularly for transformer and attention-based architectures such as RT-DETR.
• 🎯 Better model comparison and deployment planning, since previously omitted attention and softmax costs are now included.
• 🧪 More reliable profiling across PyTorch versions and input layouts, including older, unbatched, recurrent, and keyword-heavy workloads.
• ⚡ Faster profiling for layers that genuinely perform no arithmetic, with fewer unnecessary hooks and warnings.
• 🔒 Safer integration into existing projects, as profiling no longer permanently overwrites caller-defined total_ops data or model training states.

What's Changed

• Read the axes each layer owns and charge the average pool window by @raimbekovm in #150
• Preserve a caller-owned total_ops instead of colliding with it by @raimbekovm in #151
• Register the zero-op layers that reached no counting rule by @raimbekovm in #152
• Count the whole softmax family and stop dividing by an empty axis by @raimbekovm in #153
• Answer for the tree the forward pass ran, not the one it left behind by @raimbekovm in #154
• Accept any callable as a custom_ops counting rule by @raimbekovm in #155
• Hand a counting rule the arguments its module was called with by @raimbekovm in #156
• Count what an attention layer computes instead of reporting it as free by @raimbekovm in #157

Full Changelog: v2.1.2...v2.1.3

🌟 Summary

THOP 2.1.2 improves PyTorch model profiling accuracy, expands module-rule coverage, fixes parameter and formatting edge cases, and adds automated testing across supported environments. 🎯

📊 Key Changes

• Corrected adaptive average pooling counts 🧮
  Counts now follow the integer pooling windows PyTorch actually uses, producing more accurate and consistent MAC estimates for output sizes that do not evenly divide the input.

• Fixed upsampling operation rules 🚀
  Nearest and nearest-exact upsampling are now correctly treated as requiring zero arithmetic operations. Costs for other interpolation modes remain assigned to their owning hook.

• Expanded normalization and zero-operation coverage 🧩

  • GroupNorm and RMSNorm now use the existing normalization rule.
  • Constant-padding and dropout families are handled through shared base classes.
  • ReLU and LeakyReLU are explicitly treated as zero-MAC operations.
  • Lazy normalization modules receive more reliable rule resolution.

• Improved counting-rule inheritance 🧬
  Rules are now resolved through a module type’s inheritance hierarchy. Custom subclasses, parametrized layers, and many lazy modules can inherit the nearest supported rule instead of being incorrectly reported as unsupported.

• Improved parameter counting 📊
  Parameter totals now come directly from the model’s parameter tree rather than profiling hooks. This correctly handles shared weights, unused branches, unsupported module types, and parameters owned by parent modules.

• Preserved model state during profiling 🛡️
  Profiling now restores each module’s original training or evaluation mode, avoiding unintended changes in models with mixed training states.

• Fixed clever_format() edge cases ✨
  Exact powers of 1,000 and negative values now receive the correct units—for example, 1000 becomes 1.00K and -1e9 becomes -1.00G.

• Added continuous integration testing ✅
  A new GitHub Actions workflow runs the test suite on pull requests, pushes, and nightly schedules across the supported Python and PyTorch range, including Python 3.8 with PyTorch 1.8.0 and Python 3.14 with the latest PyTorch.

• Updated documentation and benchmarks 📚
  README badges, custom-rule guidance, Chinese documentation, and model benchmark rows—including YOLO11 and YOLO26 entries—now reflect the corrected counts.

• Version updated to 2.1.2 📦
  This patch release packages the profiling, reliability, documentation, and testing improvements together.

🎯 Purpose & Impact

• More trustworthy MAC estimates: Pooling and interpolation results better match actual PyTorch behavior, which improves model-complexity comparisons and deployment planning.
• Fewer false warnings and missing counts: Inherited rules cover more real-world architectures, including custom and lazy layers.
• More reliable parameter totals: Shared, unused, and previously uncounted parameters are represented correctly without double-counting.
• Safer profiling: Users can profile models without permanently altering their training modes or leaving temporary hooks and attributes behind.
• Stronger release confidence: Automated tests help catch compatibility regressions across old and new supported PyTorch environments.
• No new MultiheadAttention rule shipped: The proposed rule was intentionally removed because PyTorch 1.8 hooks cannot observe keyword-only arguments reliably; avoiding partial support prevents valid models from crashing. ⚠️

What's Changed

• Run the test suite in CI by @raimbekovm in #147
• Format an exact power of 1000 and a negative value in the right unit by @raimbekovm in #146
• Count parameters from the module tree instead of the profiling hooks by @raimbekovm in #143
• Resolve counting rules through the module type's mro by @raimbekovm in #144
• Leave the profiled model as it was found and correct the counting rules by @raimbekovm in #145
• Correct adaptive pooling and zero-op counting rules by @raimbekovm in #148

Full Changelog: v2.1.1...v2.1.2

🌟 Summary

🚀 THOP v2.1.1 adds stride-aware image profiling, enabling faster and more reliable MAC estimates for image models while modernizing benchmarks and documentation.

📊 Key Changes

• Stride-aware profile() support by @glenn-jocher:
  • Adds an optional stride argument for image models.
  • Profiles smaller, stride-aligned inputs instead of always processing the full target image.
  • Uses one profile for spatial-only models and two profiles when fixed-cost layers—such as linear or adaptive pooling layers—require more accurate estimation.
  • Falls back to the original target input if proxy inputs are unsupported or unsuitable.
  • Existing calls such as profile(model, inputs=...) remain unchanged.
• Simplified Ultralytics integration:
  • Removes the need for Ultralytics-side image-size scaling and layer-inspection logic.
  • Moves profiling behavior into THOP, the component that owns the operation-counting process.
  • Preserves compatibility with YOLOv3, YOLOv5, and third-party users of the existing API.
• Version update:
  • Bumps THOP to 2.1.1 and ultralytics-thop from 2.1.0 to 2.1.1.
• Modernized benchmarks:
  • Replaces legacy torchvision benchmark tables with fused YOLOv8, YOLO11, and YOLO26 results.
  • Updates the benchmark script to reproduce 15 model variants at 640 × 640 without downloading weights.
• Improved documentation:
  • Simplifies the English README with clearer installation and usage examples.
  • Adds a Simplified Chinese README.
  • Documents the new stride profiling workflow and custom operation rules.
• Repository workflow updates:
  • Refreshes AGENTS.md with stronger guidance around minimal changes, solving problems in the owning code path, deleting duplication, and avoiding regressions.
• Validation:
  • All 20 tests passed, with Ruff checks and formatting completed successfully ✅

🎯 Purpose & Impact

• ⚡ Faster profiling: Large image models can be analyzed using much smaller stride-aligned inputs, reducing profiling overhead.
• 🎯 More accurate estimates: Two-point profiling improves results for models that combine spatially scaling operations with fixed-cost layers.
• 🧩 Less duplicated code: Consumers no longer need to maintain their own FLOPs scaling and fallback policies.
• 🔒 Better compatibility: The new behavior is opt-in, so existing applications should continue to produce the same results when stride is omitted.
• 📈 More relevant comparisons: Updated YOLOv8, YOLO11, and YOLO26 benchmarks make it easier to compare modern Ultralytics model complexity before training or deployment.
• 🌍 Broader accessibility: Clearer documentation and Chinese-language support make THOP easier to adopt for a wider developer community.

What's Changed

• Refresh README and YOLO benchmarks by @glenn-jocher in #142
• Add stride-aware image profiling by @glenn-jocher in #149

Full Changelog: v2.1.0...v2.1.1

🌟 Summary

THOP v2.1.0 makes PyTorch model profiling substantially faster and simpler, delivering up to 2.4× lower profiling overhead while preserving bit-exact results. ⚡

📊 Key Changes

• Faster profiling through optimized bookkeeping

  • Replaced temporary float64 tensor buffers with lightweight Python integer attributes for operation and parameter totals.
  • Converted counting functions to return native Python numbers instead of allocating tensors.
  • Combined operation counting and parameter counting into a single forward hook.
  • Calculates each module’s parameter count once during hook registration rather than during every forward pass.

• Significant performance improvements

  • Reduced isolated profiling overhead from 13.4 ms to 5.6 ms, a 2.4× speedup.
  • Across 66 Ultralytics model configurations, get_flops became approximately 1.31× faster overall.
  • Real-world gains vary because model inference time still dominates total runtime.

• Removed the broken torch.fx profiler

  • Deleted the unused and unreliable fx_profile.py implementation.
  • This reduces maintenance burden and avoids exposing users to profiler behavior that was incomplete or inconsistent.

• Improved compatibility for custom operation rules

  • Custom rules that accumulate tensor values continue to work.
  • Profiling outputs are normalized to standard Python float values.
  • Added tests covering custom operations, built-in counting rules, and expected MAC and parameter totals. ✅

• Cleaner and more efficient counting internals

  • Simplified RNN, GRU, LSTM, pooling, normalization, activation, and upsampling counters.
  • Removed deprecated calculation helpers and unnecessary NumPy, warning, and tensor dependencies.

🎯 Purpose & Impact

• Faster model analysis: Developers can measure MACs, FLOPs, and parameter counts with less profiling overhead, especially for models containing many modules or hooks. 🚀
• More responsive tooling: Ultralytics utilities that rely on profiling, such as get_flops, should complete faster without changing reported results.
• Lower runtime overhead: Fewer hooks and fewer temporary tensor allocations reduce unnecessary PyTorch bookkeeping.
• Safer results: Bit-exact validation means the speed improvements do not alter profiling totals.
• Easier custom integrations: Third-party modules can continue using custom counting rules while receiving predictable Python numeric outputs.
• Upgrade consideration: Code that directly imports or depends on the removed thop.fx_profile module will need to migrate to the standard profile() API.

What's Changed

• Speed up profiling 2.4x and remove the broken fx profiler by @glenn-jocher in #141

Full Changelog: v2.0.22...v2.1.0

🌟 Summary

THOP v2.0.22 makes model profiling safer and more reliable by ensuring temporary hooks and buffers are always cleaned up—even when profiling fails or models contain shared modules. 🧹🛡️

📊 Key Changes

• Guaranteed cleanup with finally 🧼
  Profiling now removes temporary hooks and operation-counting buffers if the model’s forward pass raises an error.

• Prevented duplicate hooks on shared modules 🔗
  Modules referenced by multiple parent layers are now hooked only once, avoiding orphaned hook handles.

• Improved operation counting for shared modules 📊
  The profiler tracks already-counted modules to prevent duplicate operation and parameter totals when the same module is reached through multiple paths.

• Preserved the model’s original training state ⚙️
  Models are still restored to their previous training or evaluation mode after profiling completes or fails.

• Version updated to 2.0.22 📦

🎯 Purpose & Impact

• Prevents permanently corrupted models 🚨
  Previously, failed profiling or shared-module architectures could leave hooks pointing to deleted buffers, making the model unusable afterward. This is especially important for models such as RT-DETR-L with heavily shared activation modules.

• Makes profiling safer to run on live models ✅
  Users can profile models without worrying that temporary profiling state will remain attached to the original model.

• Improves profiling accuracy 🎯
  Shared modules are no longer counted repeatedly, producing more dependable operation and parameter estimates.

• Supports future simplification of FLOPs calculation 🚀
  With profiling cleanup now reliable across success and failure paths, Ultralytics can eventually remove the defensive deepcopy(model) workaround used by get_flops().

• No user-facing API changes are required 👍
  Existing thop.profile() workflows should continue to work while benefiting from improved robustness.

What's Changed

• Leave no hooks or buffers behind when profiling fails or shares modules by @glenn-jocher in #140

Full Changelog: v2.0.21...v2.0.22

🌟 Summary

🧹 THOP 2.0.21 makes model profiling side-effect-free while improving release reliability, documentation accuracy, and development workflows.

📊 Key Changes

• Fixed leaked profiling buffers by @raimbekovm:

  • profile() now removes temporary hooks and total_ops/total_params buffers from every module it touches.
  • Prevents profiling metadata from remaining in state_dict() or spreading through shared PyTorch modules.
  • FLOPs and parameter results remain unchanged.

• Strengthened PyPI publishing reliability:

  • Queues release workflows to avoid overlapping publish runs.
  • Retries failed PyPI uploads, including transient Sigstore, Rekor, and network failures.
  • Supports recovery of partially completed releases and safely reuses existing tags.
  • Allows SBOM uploads to replace incomplete files.

• Modernized CI/CD actions:

  • Updated GitHub Actions checkout, Python setup, and Slack integrations.
  • Replaced direct setup-uv usage with the shared Ultralytics action for more consistent behavior.

• Improved notifications and contributor workflows:

  • Slack release messages are shorter, clearer, and include clickable release, run, and PR links.
  • Replaced the third-party CLA action with the shared Ultralytics CLA workflow.
  • Added AGENTS.md guidance for AI coding agents, with CLAUDE.md linked to it.

• Updated documentation and code quality:

  • Corrected benchmark outputs and README links.
  • Standardized PyTorch imports and logging practices for Ruff 0.16.0 compatibility.
  • Updated the version to 2.0.21.

🎯 Purpose & Impact

• ✅ More trustworthy profiling: Users can profile a model without unintentionally modifying its saved state or affecting models created later in the same process.
• 🧠 Safer repeated profiling: This is especially important for complex models with shared activation modules or custom containers.
• 🚀 More dependable releases: Queuing and retry behavior reduce failed PyPI releases caused by temporary infrastructure or network issues.
• 🔄 Better recovery: Maintainers can resume incomplete releases without manually recreating tags or release metadata.
• 📚 Clearer documentation: Correct benchmark values and contribution links make THOP easier to understand and use.
• 🛠️ No expected API disruption: The release preserves profiling results and existing functionality while improving cleanup, automation, and maintainability.

What's Changed

• Improve README accuracy and links by @glenn-jocher in #125
• Bump actions/checkout from v6 to v7 in /.github/workflows by @UltralyticsAssistant in #126
• Simplify Slack notification messages by @glenn-jocher in #127
• Linkify PR numbers in publish Slack notifications by @glenn-jocher in #128
• Add AGENTS.md agent guidance and fix README drift by @glenn-jocher in #129
• Harden publish workflow recovery, notifications, and dispatch guard by @glenn-jocher in #130
• Replace third-party CLA action by @glenn-jocher in #131
• Bump slackapi/slack-github-action from v3.0.3 to v3.0.5 in /.github/workflows by @UltralyticsAssistant in #132
• Bump slackapi/slack-github-action from v3.0.5 to v4.0.0 in /.github/workflows by @UltralyticsAssistant in #133
• Use shared setup-uv action everywhere by @glenn-jocher in #134
• Bump actions/setup-python from v6 to v7 in /.github/workflows by @UltralyticsAssistant in #135
• Queue publish runs and retry PyPI upload failures by @glenn-jocher in #136
• Ruff 0.16.0 compatibility changes by @glenn-jocher in #138
• Ruff 0.16.0 compatibility changes by @glenn-jocher in #139
• Remove leaked profiling buffers from all modules by @raimbekovm in #137

New Contributors

• @raimbekovm made their first contribution in #137

Full Changelog: v2.0.20...v2.0.21

🌟 Summary

v2.0.20 is a small but useful compatibility release for thop that officially adds Python 3.14 support 🐍✨, with a minor workflow maintenance update to keep release notifications reliable 📣

📊 Key Changes

• Official Python 3.14 support added ✅

  • The package now advertises Python 3.14 compatibility in its metadata.
  • This was the main change in the release and comes from PR #124 by @glenn-jocher.

• Version bumped to 2.0.20 🔖

  • The package version was updated from 2.0.19 to 2.0.20 to publish this compatibility update.

• No model or runtime feature changes ⚙️

  • There are no new profiling features, API changes, or behavior changes in thop.
  • This release is focused on compatibility and packaging metadata, not new functionality.

• GitHub Actions Slack notification update 🔔

  • The Slack GitHub Action used in the publish workflow was updated from v3.0.2 to v3.0.3.
  • This helps keep release automation current and dependable.

🎯 Purpose & Impact

• Makes adoption of newer Python versions easier 🚀

  • Users moving to Python 3.14 can more confidently install and use thop, since the package now officially declares support.

• Improves package visibility for tools and indexes 🧰

  • Package managers, CI systems, and dependency tools can correctly recognize Python 3.14 compatibility.

• Low-risk release for existing users 👍

  • Since there are no functional code changes, current users should not expect behavior differences when upgrading from 2.0.19.

• Better release workflow reliability 📬

  • The Slack action update mainly benefits maintainers by helping publication notifications stay up to date and stable.

In short, v2.0.20 is a compatibility-focused maintenance release: the biggest takeaway is official Python 3.14 support, with no disruption to existing thop functionality 🎉

What's Changed

• Bump slackapi/slack-github-action from v3.0.2 to v3.0.3 in /.github/workflows by @UltralyticsAssistant in #123
• Add Python 3.14 support by @glenn-jocher in #124

Full Changelog: v2.0.19...v2.0.20

🌟 Summary

v2.0.19 is a maintenance-focused release: it officially bumps the package version, improves profiling correctness for parameter counting, and modernizes CI/CD workflows—without introducing new model architectures or user-facing API changes. 📦✅

📊 Key Changes

• (Most immediate change, current PR #122) thop version updated from 2.0.18 → 2.0.19 by @glenn-jocher 🔢

  • This is the release marker and packaging update.

• Profiling correctness fix (important functional improvement) 🧠

  • In thop/profile.py, recursive counting now starts with the module’s own parameter count (module.total_params.item()), improving total parameter accounting.
  • In thop/vision/basic_hooks.py, parameter counting now uses parameters(recurse=False) to avoid double-counting child-module parameters in hierarchical models.

• New tests to validate Conv2D parameter counting behavior 🧪

  • Added checks for:
    • root Conv2d modules counting their own params correctly
    • Sequential wrappers not double-counting child params

• GitHub Actions workflow upgrades ⚙️

  • Updated core actions (checkout, setup-python, upload-artifact, download-artifact) and Slack action versions.
  • Includes migration toward newer Node runtime requirements (Node 24 in several actions).

• Documentation and style cleanups 🧹

  • Minor docstring formatting improvements.
  • External URL cleanup (YouTube + CUDA toolkit links) for more reliable docs navigation.

🎯 Purpose & Impact

• For users of thop profiling: more reliable parameter totals, especially in nested module structures. 📈
• For developers and maintainers: stronger regression protection via new tests and cleaner docs. 👩‍💻
• For CI/CD pipelines: updated automation stack improves long-term maintainability, but self-hosted runners may need to meet newer runner/runtime requirements. 🚦
• Overall risk: low-to-moderate operational change; no new model features, but a meaningful correctness fix in profiling internals. 👍

What's Changed

• Bump actions/upload-artifact from 4 to 5 in /.github/workflows by @dependabot[bot] in https://github.com/ultralytics/thop/pull/108
• Bump actions/download-artifact from 4 to 6 in /.github/workflows by @dependabot[bot] in https://github.com/ultralytics/thop/pull/107
• Bump actions/setup-python from 5 to 6 in /.github/workflows by @dependabot[bot] in https://github.com/ultralytics/thop/pull/106
• Update Google-style docstrings by @glenn-jocher in https://github.com/ultralytics/thop/pull/109
• Bump actions/checkout from 5 to 6 in /.github/workflows by @dependabot[bot] in https://github.com/ultralytics/thop/pull/110
• Bump actions/upload-artifact from 5 to 6 in /.github/workflows by @dependabot[bot] in https://github.com/ultralytics/thop/pull/112
• Bump actions/download-artifact from 6 to 8 in /.github/workflows by @dependabot[bot] in https://github.com/ultralytics/thop/pull/118
• Bump actions/upload-artifact from 6 to 7 in /.github/workflows by @dependabot[bot] in https://github.com/ultralytics/thop/pull/117
• Bump slackapi/slack-github-action from 2.1.1 to 3.0.1 in /.github/workflows by @dependabot[bot] in https://github.com/ultralytics/thop/pull/119
• Fix redirected URLs by @glenn-jocher in https://github.com/ultralytics/thop/pull/120
• Bump slackapi/slack-github-action from v3.0.1 to v3.0.2 in /.github/workflows by @UltralyticsAssistant in https://github.com/ultralytics/thop/pull/121
• Bump version from 2.0.18 to 2.0.19 by @glenn-jocher in https://github.com/ultralytics/thop/pull/122

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.18...v2.0.19

🌟 Summary

thop v2.0.18 is a maintenance-focused patch release: a version bump with code quality and CI workflow updates—no runtime or API changes. ✅

📊 Key Changes

• Version bump only
  • Updated package version to 2.0.18 in thop/__init__.py via PR #105: Bump thop to 2.0.18 🔖
• Code quality refinements (no behavior changes)
  • Lint/format tweaks and clearer logging in PR #103: Ruff RUF refactor 🧹
  • Reordered __all__, prefixed unused variables, minor f-string improvement
• Automated dependency management
  • Added Dependabot configuration with monthly checks in PR #99 🤖
• CI workflow hardening
  • Bumped actions and tools:
    • actions/checkout v5 in workflows PR #101 ⬆️
    • astral-sh/setup-uv v7 PR #102 ⚡
    • slackapi/slack-github-action v2.1.1 PR #100 📨
  • Tests updated to silence unused variable warnings

🎯 Purpose & Impact

• Safe to upgrade: No functional or API changes; existing code continues to work as-is 🚀
• Better version tracking: Easier for downstream users/tools to pin and detect the latest patch 🔍
• Improved maintainability: Cleaner code and fewer linter warnings enhance long-term stability 🛠️
• More reliable CI: Updated GitHub Actions and tooling reduce build issues and keep pipelines current 🔄
• Security and freshness: Dependabot automates dependency updates with minimal noise 🔐

Note for maintainers: actions/checkout v5 and setup-uv v7 use newer Node runtimes; ensure self-hosted runners meet the required versions for CI. 🧩

What's Changed

• Create dependabot.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/99
• Bump slackapi/slack-github-action from 2.0.0 to 2.1.1 in /.github/workflows by @dependabot[bot] in https://github.com/ultralytics/thop/pull/100
• Bump actions/checkout from 4 to 5 in /.github/workflows by @dependabot[bot] in https://github.com/ultralytics/thop/pull/101
• Bump astral-sh/setup-uv from 4 to 7 in /.github/workflows by @dependabot[bot] in https://github.com/ultralytics/thop/pull/102
• Ruff RUF refactor by @glenn-jocher in https://github.com/ultralytics/thop/pull/103
• Bump thop to 2.0.18 by @glenn-jocher in https://github.com/ultralytics/thop/pull/105

New Contributors

• @dependabot[bot] made their first contribution in https://github.com/ultralytics/thop/pull/100

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.17...v2.0.18

🌟 Summary

v2.0.17 focuses on cleaner, safer imports and internal refactors that make THOP more maintainable and reliable without changing its public API. 🧹✨

📊 Key Changes

• Replaced wildcard imports with explicit imports across core modules (profile, basic hooks, FX profiling) for clearer dependencies and fewer name clashes. 🔒
• Kept public APIs available at the package root and clarified exports via __all__ (e.g., profile, profile_origin, clever_format, default_dtype). ✅
• Improved FX profiling internals with explicit torch.fx imports and clearer logging/util usage; added short docstrings for readability. 🧠
• Minor test fix (renamed a duplicate test) to ensure consistent test coverage. 🧪
• Documentation link updates to official PyTorch docs and refreshed download analytics link. 📚
• Version bump to 2.0.17. 🏷️

Top PR prioritized:

• PR #98 by @glenn-jocher: “Unpack star imports” — converts import * to explicit imports and clarifies module boundaries for safer, more maintainable code.

Additional improvements:

• PR #94 by @pderrenger: Internal refactors improving performance, readability, and structure. 🚀

🎯 Purpose & Impact

• More reliable and maintainable codebase: explicit imports reduce accidental name collisions and make the code easier to navigate and review. 🧭
• Better tooling and IDE support: explicit symbols enable improved static analysis, autocomplete, and linting. 🛠️
• Stable user experience: no breaking API changes; root-level functions like profile still work as before. 👍
• Cleaner profiling internals: FX-based profiling becomes easier to reason about and extend. 🧩

Quick note for users:

• Your imports should continue to work as usual:

  from thop import profile, clever_format
  

• If you were relying on deep, relative, or internal imports, consider switching to the documented public APIs for long-term stability. 💡

For more details, see the PRs: Unpack star imports (PR #98) and Ultralytics Refactor (PR #94).

What's Changed

• Ultralytics Refactor https://ultralytics.com/actions by @pderrenger in https://github.com/ultralytics/thop/pull/94
• Unpack star imports by @glenn-jocher in https://github.com/ultralytics/thop/pull/98

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.16...v2.0.17

🌟 Summary

Improved MACs accounting for ConvTranspose layers and strengthened CI security with automated SBOMs, delivering more accurate profiling results and safer releases. 🚀

📊 Key Changes

• Correct MACs for ConvTranspose layers ✅
  • Added a dedicated count_convtNd hook for ConvTranspose1d/2d/3d.
  • Updated calculate_conv2d_flops with a transpose flag to correctly handle transposed convolutions.
• Clear terminology: tests now use “MACs” instead of “FLOPs” for consistency and accuracy.
• Expanded test coverage 🧪
  • New ConvTranspose2d tests (no-bias, with-bias, groups, random).
  • Symmetry test ensures Conv2d downsample and ConvTranspose2d upsample report equal MACs in matching setups.
• CI/Release hardening 🔐
  • Automated SPDX SBOM generation attached to each release.
  • More reliable SBOM workflow using a dedicated virtual environment and editable installs.
  • Safer PR title handling in notifications.
• Version bump to 2.0.16.

Tip: Usage remains the same

• from thop import profile
• macs, params = profile(model, inputs=(x,))

🎯 Purpose & Impact

• More accurate complexity metrics for models using transposed convolutions (common in decoders/upsampling) ➜ better comparison, profiling, and deployment planning. 📈
• Clearer language reduces confusion between MACs and FLOPs, improving interpretability for all users. 🧠
• Stronger supply-chain transparency and compliance with SBOMs, benefiting security-conscious teams. 🛡️
• No breaking API changes; you may see adjusted (more accurate) MAC counts on ConvTranspose-heavy architectures. ✅

What's Changed

• Add automated SPDX SBOM generation to releases by @glenn-jocher in https://github.com/ultralytics/thop/pull/90
• Update publish.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/91
• Fix ConvTranspose by @FrzMtrsprt in https://github.com/ultralytics/thop/pull/93

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.15...v2.0.16

🌟 Summary

This release introduces support for trilinear upsampling in model profiling, alongside several workflow and documentation improvements for better usability, security, and contributor experience. 🚀

📊 Key Changes

• Trilinear Upsample Support: Added calculation for trilinear mode in the Upsample layer, enabling accurate profiling of models that use 3D upsampling. 🧊
• Workflow Enhancements:
  • Improved GitHub Actions workflows for code formatting, labeling, and security permissions. 🔒
  • Automated Contributor License Agreement (CLA) checks for all pull requests to ensure compliance. 🤝
• Documentation Updates: Expanded and clarified README and benchmarking documentation for easier onboarding and better understanding of MACs vs. FLOPs. 📚

🎯 Purpose & Impact

• Broader Model Support: With trilinear upsampling now profiled, users working on 3D models (such as medical imaging or volumetric data) get more accurate computational metrics, making THOP more versatile. 🩻
• Smoother Contributions: Automated CLA checks and improved workflow automation make it easier and safer for new contributors to participate, while maintainers benefit from streamlined processes.
• Better Transparency & Usability: Enhanced documentation and clearer workflow permissions help both new and experienced users understand and use the tool more effectively, while also improving repository security and compliance.

Overall, this update makes THOP more robust, user-friendly, and ready for a wider range of deep learning projects! 🌐✨

What's Changed

• Ultralytics Refactor https://ultralytics.com/actions by @pderrenger in https://github.com/ultralytics/thop/pull/83
• Ultralytics Refactor https://ultralytics.com/actions by @pderrenger in https://github.com/ultralytics/thop/pull/84
• Add Permissions to Ultralytics Actions format.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/85
• Potential fix for code scanning alert no. 1: Workflow does not contain permissions by @glenn-jocher in https://github.com/ultralytics/thop/pull/86
• Create cla.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/88
• Update cla.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/89
• Enable trilinear mode calculation in Upsample by @FrzMtrsprt in https://github.com/ultralytics/thop/pull/87

New Contributors

• @FrzMtrsprt made their first contribution in https://github.com/ultralytics/thop/pull/87

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.14...v2.0.15

🌟 Summary

Minor updates to standardize license formatting across workflow and project files. 🛠️✨

📊 Key Changes

• Adjusted the license formatting in format.yml, publish.yml, and pyproject.toml to a consistent style:
  Changed from Ultralytics YOLO 🚀, AGPL-3.0 license to Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license.

🎯 Purpose & Impact

• 📄 Improved Clarity: Ensures the license reference is standardized and easier to understand.
• ✅ Consistency: Promotes a polished, uniform format across all project-related files.
• 🌐 Transparency: Reinforces openness and accessibility by linking directly to the license details.

What's Changed

• Ultralytics Refactor https://ultralytics.com/actions by @glenn-jocher in https://github.com/ultralytics/thop/pull/78
• Standardize license headers in Python files by @pderrenger in https://github.com/ultralytics/thop/pull/79
• Standardize license headers in TOML/YAML files by @pderrenger in https://github.com/ultralytics/thop/pull/80
• Update pyproject.toml by @glenn-jocher in https://github.com/ultralytics/thop/pull/81
• Update init.py by @glenn-jocher in https://github.com/ultralytics/thop/pull/82

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.13...v2.0.14

🌟 Summary

Release v2.0.13 introduces significant improvements to the CI/CD workflow for the Ultralytics thop repository, enhancing maintainability, automation, and efficiency for package building and publishing. 🚀

📊 Key Changes

• 🔄 The GitHub Actions workflow was modularized into clear, logical steps: check, build, publish, and notify.
• ✅ Added precise checks to determine when to increment versioning and proceed with publishing, ensuring only necessary updates are released.
• 🛠 Refined artifact handling by separating the building process from publishing and enabling artifact uploading for better clarity.
• 🔧 Improved error handling (e.g., continue-on-error) for greater workflow resilience, reducing disruptions from non-critical issues.
• 🔔 Enhanced Slack notifications to better inform users about success or failure in specific workflow stages.
• 📦 Updated the project version from 2.0.12 to 2.0.13.

🎯 Purpose & Impact

• 🧹 Streamlined Processes: The modular structure makes the CI/CD pipeline easier to understand, debug, and maintain, benefiting developers and contributors alike.
• 🚀 Improved Automation: More robust checks and workflow refinements minimize redundant actions, reducing overall execution time and ensuring smoother deployments.
• 🛡 Enhanced Reliability: Introducing better error handling prevents minor issues from halting the entire workflow, leading to more reliable operations.
• 📢 Clearer Communication: Notification improvements help keep stakeholders up-to-date with the exact status of releases.

This update makes development workflows more efficient for contributors while ensuring smoother experiences for end-users through well-tested and efficiently published updates. 🎉

What's Changed

• Ultralytics Refactor https://ultralytics.com/actions by @pderrenger in https://github.com/ultralytics/thop/pull/74
• Update publish.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/75
• Update publish.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/76

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.12...v2.0.13

🌟 Summary

The release v2.0.12 enhances Slack integration for notifications and updates the project's version.

📊 Key Changes

• Slack Integration Update: The Slack GitHub action is upgraded from version v1.26.0 to v2.0.0, changing the configuration syntax for more robust notifications.
• Version Update: The project version is incremented from 2.0.11 to 2.0.12, reflecting these updates.

🎯 Purpose & Impact

• Improved Notification Stability: The update to Slack's latest version ensures more reliable notifications for workflow successes and failures, leveraging enhancements from the Slack action update.
• Version Bump for Maintenance: Upgrading the version helps users and developers track changes easily, underscoring minor improvements or bug fixes without major functional shifts.

What's Changed

• Update publish.yml to PyPI by @glenn-jocher in https://github.com/ultralytics/thop/pull/71
• Fix PyPI downloads links by @pderrenger in https://github.com/ultralytics/thop/pull/72
• Slack v2 API usage by @glenn-jocher in https://github.com/ultralytics/thop/pull/73

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.11...v2.0.12

🌟 Summary

The v2.0.11 release for the ultralytics/thop project focuses on refining the deployment process through improved GitHub Actions workflows and enhanced notification clarity.

📊 Key Changes

• GitHub Actions Enhancement: Updated the publish.yml with an environment section detailing deployment specifications, including environment name and URL for better tracking via GitHub's Deployments tab.
• Version Number Update: Incremented the version number from 2.0.10 to 2.0.11 to reflect these updates.
• Notification Improvements: Enhanced Slack notifications in GitHub Actions to display the current release tag being published, improving the context of updates.

🎯 Purpose & Impact

• Streamlined Deployment: The addition of environment details in the workflow increases transparency and organization in the deployment process, aiding developers in effectively managing and tracking releases. 🌐
• Clear Versioning: The version bump to 2.0.11 signals minor improvements, assisting users in pinpointing updates and maintaining software consistency. 🆙
• Improved Communication: Updating Slack notifications with more precise information helps team members quickly recognize and respond to the latest package releases, enhancing the team's operational efficiency. 📢🚀

What's Changed

• Update publish.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/69
• Add environment to publish.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/70

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.10...v2.0.11

🌟 Summary

The release of version 2.0.10 introduces support for Python 3.13 and improvements in release automation processes.

📊 Key Changes

• Python 3.13 Support: The project now officially supports Python 3.13, as indicated in the metadata.
• Version Upgrade: The internal version is updated from 2.0.9 to 2.0.10 to reflect these changes.
• Updated Publish Workflow: The GitHub Actions publish workflow has been optimized:
  • Replaced requests library with ultralytics-actions to simplify dependencies.
  • Transitioned to using a utility function check_pypi_version() for version management.
  • Implemented ultralytics-actions-summarize-release for release note automation.

🎯 Purpose & Impact

• Enhanced Compatibility: By supporting Python 3.13, the project ensures users can leverage the latest features and improvements of the Python language in their work, promoting greater flexibility and longevity in software development. 🐍
• Streamlined Workflow: The workflow enhancements provide a more efficient and secure process for managing and publishing new versions, reducing potential errors and dependency on manual operations. 🚀
• Improved Maintainability: Utilization of standard utilities and functions helps maintain clean, organized, and reusable code, ultimately leading to a more robust and reliable codebase. 📦

What's Changed

• Update publish.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/66
• Update publish.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/67
• Python 3.13 update by @glenn-jocher in https://github.com/ultralytics/thop/pull/68

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.9...v2.0.10

🌟 Summary

The release v2.0.9 primarily enhances the publishing workflow for the Ultralytics THOP repository, focusing on improving security and streamlining processes.

📊 Key Changes

• Updated the GitHub Actions workflow to incorporate trusted publisher permissions for PyPI.
• Separated the build and publish steps to optimize the workflow process.
• Adopted a more straightforward method using GitHub Actions for publishing packages to PyPI.
• Incremented the software version from 2.0.8 to 2.0.9.

🎯 Purpose & Impact

• Enhanced Security: By employing token permissions, the release ensures a secure method of publishing to PyPI, safeguarding against unauthorized access and enhancing trustworthiness. 🛡️
• Streamlined Publishing: The transition to using a dedicated action for PyPI publication simplifies the process, reducing the chances of errors and making it more efficient and reliable. 🚀
• Version Consistency: Updating the version number ensures that changes are clearly communicated and that users are working with the latest improvements. 🔄

What's Changed

• Ultralytics Code Refactor https://ultralytics.com/actions by @UltralyticsAssistant in https://github.com/ultralytics/thop/pull/63
• Update publish.yml to trusted publisher by @glenn-jocher in https://github.com/ultralytics/thop/pull/64

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.8...v2.0.9

🌟 Summary

New version release for thop: updated from v2.0.7 to v2.0.8 🎉

📊 Key Changes

• Incremented the version number in the __init__.py, marking the new release.
• Updated project maintainer details in pyproject.toml.

🎯 Purpose & Impact

• Version Update: This update indicates either minor tweaks, bug fixes, or small feature improvements, promoting enhanced performance or resolving any identified issues.
• Improved Correspondence: Updating the maintainer to "Ultralytics" with a new email contact standardizes communication lines and enhances organizational clarity.
• User Impact: All users should upgrade to ensure compatibility and potentially experience improved functionality.

What's Changed

• Update pyproject.toml by @glenn-jocher in https://github.com/ultralytics/thop/pull/58
• ultralytics-thop 2.0.8 by @glenn-jocher in https://github.com/ultralytics/thop/pull/59

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.7...v2.0.8

🌟 Summary

v2.0.7 enhances the project by improving metadata accessibility and ensuring clarity through focused code updates.

📊 Key Changes

• Updated pyproject.toml to include new links for Homepage, Source, Documentation, and Changelog.
• Bumped the project version from 2.0.6 to 2.0.7.
• Refined code with updated string formatting to use modern f-strings.
• Improved error message logging for clarity in thop/profile.py.
• Fixed tensor size logic in count_convNd_ver2 function.

🎯 Purpose & Impact

• Enhanced User Navigation: 🖱️ The addition of direct links to the project's homepage, source, and documentation provides users with easy access to essential resources.
• Transparency and Tracking: 📜 A new Changelog URL helps users keep track of changes and updates conveniently.
• Readability and Maintenance: 🧹 Transition to f-strings improves code readability, making it easier for developers to work with and maintain.
• Error Identification: 🚨 Improved logging messages help developers quickly identify and resolve issues.
• Robust Performance: 🏋️ Bug fixes in tensor handling ensure accurate performance measurements in convolutional operations, maintaining integrity in computational processes.

What's Changed

• PyUpgrade 3.8 updates by @glenn-jocher in https://github.com/ultralytics/thop/pull/55
• Update pyproject.toml URLs by @glenn-jocher in https://github.com/ultralytics/thop/pull/57

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.6...v2.0.7

🌟 Summary

This release (v2.0.6) focuses on enhancing performance, improving workflow efficiency, and ensuring clearer and more consistent version handling. Additionally, a Reddit badge has been added to increase community engagement visibility.

📊 Key Changes

• 🚀 Code refactoring to improve performance and readability.
• 🏷️ Updated version tagging for enhanced clarity in workflows.
• 🌐 Added a Reddit badge in the README to connect with community updates.
• 📜 Streamlined GitHub publish workflow by removing OpenAI dependency and using a pre-fetched summarization script.
• 🗃️ Changed the thop library version from 2.0.5 to 2.0.6 for minor updates.

🎯 Purpose & Impact

• Performance Boost: By optimizing functions and simplifying complex code, the software now runs faster and is easier to maintain, benefitting both developers and users alike.
• Workflow Efficiency: Removal of the OpenAI dependency reduces potential errors and speeds up the release process by fetching a tested script, making the software deployment more reliable.
• Clarity and Consistency: Updated version tagging makes it easier for developers to track changes and understand which versions are being referenced, reducing confusion and improving communication in deployment notifications.
• Community Engagement: The addition of a Reddit badge encourages more community interaction and provides an easy way for users to follow discussions and updates.
• Software Maintenance: Keeping the library version updated ensures users benefit from the latest bug fixes and optimizations, leading to a smoother experience.

What's Changed

• Add https://www.reddit.com/r/Ultralytics/ badge by @glenn-jocher in https://github.com/ultralytics/thop/pull/51
• Update publish.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/52
• Update publish.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/53
• Update init.py by @glenn-jocher in https://github.com/ultralytics/thop/pull/54

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.5...v2.0.6

🌟 Summary

Version 2.0.5 includes improvements to the GitHub workflow for processing pull requests and a version bump in the THOP library.

📊 Key Changes

• 🛠️ Enhanced the GitHub Actions workflow to better handle pull request events and extract details more efficiently.
• 🔢 Updated the version of the THOP library from 2.0.3 to 2.0.5.

🎯 Purpose & Impact

• 🚀 Improved Workflow Efficiency: By refining the method of extracting pull request details, the process is now more robust and can handle different event types like pull_request and pull_request_target. This ensures that the right information is captured, which simplifies release procedures and enhances automation reliability.
• 🔄 Library Update: Incrementing the version number signifies minor updates or improvements to THOP, typically signaling to users that bug fixes or minor enhancements may have been introduced, contributing to a smoother integration or performance.

What's Changed

• Use PAT for PR details by @glenn-jocher in https://github.com/ultralytics/thop/pull/50

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.4...v2.0.5

🌟 Summary

The v2.0.4 update introduces enhancements in the automation of GitHub workflows and a version bump for the package.

📊 Key Changes

• Improved GitHub Actions workflow to handle pull request events more comprehensively.
• Updated version number in the thop library from 2.0.3 to 2.0.4.

🎯 Purpose & Impact

• Enhanced Workflow Automation: The update ensures that pull request targets (pull_request_target) are correctly processed along with standard pull requests, leading to more robust and versatile workflow automation. This change can improve the efficiency of release processes and ensure accurate extraction of pull request details, minimizing manual intervention. ⚙️
• Version Update: This change indicates either bug fixes or minor improvements to the thop library, signaling reliability and improved functionality to its users. 📦

Overall, these changes streamline developer operations and ensure consistency in processing updates and version control. 🌐

What's Changed

• Update publish.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/49

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.3...v2.0.4

🌟 Summary

The v2.0.3 release features improvements in the release automation process and provides detailed test documentation for better understanding and usage of THOP components.

📊 Key Changes

• GitHub Actions Update: The workflow file was updated to run on the latest Ubuntu version instead of macOS for improved compatibility and efficiency.
• Python Version Flexibility: The Python environment setup now supports any version in the 3.x series, instead of a fixed 3.11, enhancing compatibility.
• Testing Enhancements: Documentation was added to test utility classes explaining their purpose in profiling the performance of neural network components.

🎯 Purpose & Impact

• Faster and More Reliable CI/CD: Switching to Ubuntu and allowing any Python 3.x version makes the continuous integration and deployment process more robust and reduces job failures due to environment mismatches. 🚀
• Greater Clarity in Testing: By adding docstrings to test files, developers and users gain a better understanding of how tests assess the performance and efficiency of neural network layers. This transparency aids in debugging and optimizing code. 🧑‍💻
• Ease of Adaptation: The flexible Python version setup ensures that future updates and user environments remain compatible without needing constant adjustments to the workflow file. 🔧

What's Changed

• Remove fetch-depth 0 on publish by @glenn-jocher in https://github.com/ultralytics/thop/pull/41
• Update publish.yml to remove || true by @glenn-jocher in https://github.com/ultralytics/thop/pull/42
• Update publish.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/43
• Prettier reformat publish.yml by @glenn-jocher in https://github.com/ultralytics/thop/pull/44
• Fix publish.yml repo name by @glenn-jocher in https://github.com/ultralytics/thop/pull/45
• Update publish.yml to GPT4o-08-06 by @glenn-jocher in https://github.com/ultralytics/thop/pull/46
• Update docstrings for tests by @glenn-jocher in https://github.com/ultralytics/thop/pull/47

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.2...v2.0.3

🌟 Summary

Version 2.0.2 brings critical updates for GitHub Actions and versioning improvements.

📊 Key Changes

• Modified GitHub Actions workflow to prefer PERSONAL_ACCESS_TOKEN or fallback to GITHUB_TOKEN for authentication.
• Incremented the library version from 2.0.1 to 2.0.2.

🎯 Purpose & Impact

• 💼 Improved Workflow: By allowing the use of PERSONAL_ACCESS_TOKEN or GITHUB_TOKEN, it ensures more flexible and reliable authentication in GitHub Actions.

  • Impact: Ensures smoother automated processes and reduces potential authentication errors during repository actions.

• 🆙 Versioning Update: Updating the version number to 2.0.2 reflects the latest changes and fixes.

  • Impact: Users and developers get clear visibility of the latest stable release, ensuring they work with the most up-to-date version of the software.

What's Changed

• Use PAT for actions by @glenn-jocher in https://github.com/ultralytics/thop/pull/40

Full Changelog: https://github.com/ultralytics/thop/compare/v2.0.1...v2.0.2