mirror of
https://github.com/firestar5683/StarPilot.git
synced 2026-09-12 03:03:51 +08:00
56aa07844a3b4316691ba454692016d23b8cd00a
5 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Jason Wen
|
56aa07844a |
modeld: retain SNPE and thneed drive model support (#530)
* modeld: Retain pre-20hz drive model support * Method not available anymore on OP * some fixes * Revert "Long planner get accel: new function args (#34288)" * Revert "Fix low-speed allow_throttle behavior in long planner (#33894)" * Revert "long planner: allow throttle reflects usage (#33792)" * Revert "Gate acceleration on model gas press predictions (#33643)" * Reapply "Gate acceleration on model gas press predictions (#33643)" This reverts commit 76b08e37cb8eb94266ad9f6fed80db227e7c3428. * Reapply "long planner: allow throttle reflects usage (#33792)" This reverts commit c75244ca4e9c48084b0205b7c871e1a4e0f4e693. * Reapply "Fix low-speed allow_throttle behavior in long planner (#33894)" This reverts commit b2b7d21b7b685a2785d1beede3d223f0bb954807. * Reapply "Long planner get accel: new function args (#34288)" This reverts commit 74dca2fccf4da59cc8ac62ba9c0ad10ba3fc264b. * don't need * retain snpe * wrong * they're symlinks * remove * put back into VCS * add back * don't include built * Refactor model runner retrieval with caching support Added caching for active model runner type via `ModelRunnerTypeCache` to enhance performance and avoid redundant checks. Introduced a `force_check` flag to bypass the cache when necessary. Updated related code to handle cache clearing during onroad transitions. * Update model runner determination logic with caching fix Enhances `get_active_model_runner` to utilize caching more effectively by ensuring type consistency and updating cache only when necessary. Also updates `is_snpe_model` to pass the `started` state to the runner determination function, improving behavior for dynamic checks. * default to none * enable in next PR * more --------- Co-authored-by: DevTekVE <devtekve@gmail.com> |
||
|
DevTekVE
|
9e6656fd2e |
models: Fix cache expiration logic in fetcher.py
Ensure the cache is treated as expired when last_sync is zero. This prevents potential issues with uninitialized or missing sync timestamps, enhancing reliability. |
||
|
Jason Wen
|
c13f159a77 |
Revert "models: retain SNPE runner support" (#529)
Revert "models: retain SNPE runner support (#522)"
This reverts commit
|
||
|
Jason Wen
|
f84d27c4ee |
models: retain SNPE runner support (#522)
* tinygrad with snpe * force with snpe to validate * fix path * fix more paths * Adjust modeld execution logic based on active model runner Introduced a check to conditionally execute `modeld` based on the active model runner. Added support for distinguishing between SNPE and TinyGrad runners using new helper functions and updated `custom.capnp` definitions. This change optimizes process management by ensuring compatibility with the selected model runner. * Refactor modeld process function checks. Introduce `is_stock_model` to clarify logic and replace direct uses of `is_snpe_model` where the stock model condition is needed. Additionally, rename the duplicate "modeld" process in sunnyPilot to "modeld_snpe" for clarity and consistency. * ignore tg * fix process name * ruff * fix thneed paths * mypy * remove our own * use upstream compile3 * fix thneed * try this * Revert "remove our own" This reverts commit 1cf4f57502565c274628e185c66f1eb04c4d9bbe. * try using compile2.py again * add back symlink * fix path * more fix * wrong path again * Revert "wrong path again" This reverts commit f5301c19d594defb23cda2604168bcb02c830f3e. * update * hardcode path to our submodule * force path * try this * fix file name * try this * again * Revert "again" This reverts commit 17c8cd73768a2de5aaf2a34d202aa6f03d59d72d. * Revert "try this" This reverts commit 767f78bbcf17b08edbc20f283d10a1990bf4264f. * Revert "fix file name" This reverts commit 485eef68da981da571fe9a9ebf3aecdb95e5588c. * Revert "try this" This reverts commit 41fef87680cdda1ba8cbd7e5c59256e6c57010a0. * Revert "force path" This reverts commit 5c3b408937bff0f61b150971b328d127501c5bf0. * Revert "hardcode path to our submodule" This reverts commit 5ee1950b6f4fce73e7e73b6971b996a9989d0a93. * Revert "update" This reverts commit fb313bd7fbeb50111f3c4a11137b4cfd9cabc3fa. * Reapply "wrong path again" This reverts commit 309639aeb3575e20b215bbbcc27e48923ee95c87. * Revert "wrong path again" This reverts commit f5301c19d594defb23cda2604168bcb02c830f3e. * Revert "more fix" This reverts commit 23dd423e78e60bd6a0bd73e921bfba9a5aad88bd. * Revert "fix path" This reverts commit 75d338f2bda2dfa77044e33f4c6553390621b13b. * Revert "add back symlink" This reverts commit 9f71ad0b8ae2068e431c455af9d4f954d024eec7. * Revert "try using compile2.py again" This reverts commit 914117d2e1c6c4270f7bef1a5305847a1d7bab17. * Reapply "remove our own" This reverts commit b1996377b346658f2094cba9032a66230784cb3e. * don't even compile anymore * need it for default snpe model * add to lfs * bring onnx back for sim * must add this back * need this --------- Co-authored-by: DevTekVE <devtekve@gmail.com> |
||
|
DevTekVE
|
993c4a0d2a |
Driving Model Manager (#457)
* Introduce Model Manager to handle downloads and verification
This commit introduces a new Model Manager responsible for handling model downloads, including driving and navigation application models. The manager also verifies file hashes and communicates download progress for an improved user experience.
The Model Manager is asynchronous and utilizes asyncio and aiohttp for enhanced performance, including robust error handling.
Impacted files in the 'cereal', 'common', 'sunnypilot', and 'system' directories have been updated accordingly. The 'ModelsFetcher' process configuration has been modified to run only when off-road, ensuring optimum resource management.
This update aims to enhance code clarity, improve performance, and streamline the handling of model downloads.
* "Update model management and fetching for SunnyPilot"
This update refactors the model management, downloading and cache verification mechanisms of SunnyPilot. New functionalities, such as smart cache handling, have been implemented in ModelFetcher in sunnypilot/models/model_fetcher.py. Also, the model downloading process has been moved to a separate async function _download_bundle in ModelManagerSP in sunnypilot/models/model_manager.py. Hash verification of files is now performed in an async function verify_file in sunnypilot/models/model_helper.py. Changes in system parameters related to model management have been reflected in system/manager/manager.py.
* Integrate download ETA calculations in model manager
This commit introduces a new feature that calculates and tracks the Estimated Time of Arrival (ETA) for downloading models in the model manager component. The 'eta' property in the 'DownloadProgress' structure in 'custom.capnp' is changed from 'Float32' to 'UInt32'.
In the 'model_manager.py' file, a new method '_calculate_eta' has been added to perform ETA calculations. An additional dictionary '_download_start_times' has been created to keep track of the start time of each model download. The ETA is calculated every time a portion of the model file is downloaded, and it gets updated in the 'DownloadProgress' structure. Finally, the start time is cleared after the download completes.
In 'model_manager_audit.py', an additional check is added to only print downloadProgress for the downloads currently in progress.
* format
* no default model cache {} because it can be considered a valid json, we do not want that
* Refactor typing annotations to use PEP 604 syntax.
Updated type hints to adopt PEP 604 union syntax (`X | None`) and replaced `List` and `Dict` with modern built-in `list` and `dict`. This change improves consistency and readability while aligning with Python 3.10+ standards.
* Simplify logging messages and remove unused imports.
Removed an unused import from `model_manager.py` to improve clarity and maintainability. Also refined a log message in `model_fetcher.py` by removing unnecessary formatting for consistency.
* Refactor model handling and simplify cache fallback logic.
Updated type annotations for `selected_bundle` in `model_manager.py` for clarity. Streamlined cache fallback logic in `model_fetcher.py` by removing redundant conditionals while preserving functionality. These changes improve code readability and maintainability.
* "Fix formatting for ModelManager_DownloadIndex retrieval
Condensed parameter alignment in the get method for improved readability and adherence to style guidelines. This change does not affect functionality but ensures consistent code formatting."
* Need to have main defined for process_config to be able to run it
* Refactor model management to support active bundle tracking
Introduce the concept of an active model bundle with a new persistent parameter and API updates. Added fields for `generation` and `environment` in model metadata, improved caching, and updated methods to manage active model states efficiently.
* UI commit (#515)
* Refactor model management to support active bundle tracking
Introduce the concept of an active model bundle with a new persistent parameter and API updates. Added fields for `generation` and `environment` in model metadata, improved caching, and updated methods to manage active model states efficiently.
* Add new driving model selection feature to settings
This commit introduces a new feature to the settings that allows users to select different driving models. It fetches available models and displays their download progress. The created UI also suggests a calibration reset after model download. The changes include the creation of 'SoftwarePanelSP' within 'settings.' Additionally, 'sunnypilot/SConscript' has been updated to include 'settings.cc' and 'software_panel.cc'. Changes also include localization for this feature.
* Show model description during download status
This update ensures the model description is displayed when a model is in the downloading state. It improves the user interface by providing real-time feedback during the download process.
* Update translations for multiple languages
Added new and updated translation strings in several language files, including Spanish, Arabic, Chinese (Simplified and Traditional), Turkish, Korean, Thai, Japanese, and Brazilian Portuguese. These updates include placeholder translations for new UI elements and features.
* Refactor model name handling and add generation check.
Replaced `bundleName` with `model_name` for better clarity in status messages. Added a generation mismatch check before showing the reset parameters dialog to avoid unnecessary prompts.
* Update model handling in SoftwarePanelSP
Remove unused "common/model.h" and replace "CURRENT_MODEL" with "..." as the default return value in GetModelName. Adjust logic to check for active bundles instead of selected bundles for improved accuracy. Minor text change for clarity in UI label.
* Rename `GetModelName` to `GetActiveModelName` for clarity.
The new name better reflects the function's purpose of retrieving the active model name, improving code readability. All relevant calls and references have been updated to ensure consistency across the codebase.
* Update sunnypilot/models/model_helper.py
Co-authored-by: sourcery-ai[bot] <58596630+sourcery-ai[bot]@users.noreply.github.com>
* Refactor download status handling and add 'cached' state
Introduce 'cached' as a new download status and adjust relevant logic across components to support it. Simplify and streamline model status handling in the software panel for better readability and maintainability. Ensure consistent status reporting for all model types.
* Update translations for multiple languages
Refined and expanded translations across various languages, replacing placeholders with meaningful text. This improves clarity and user experience in the multilingual interface.
* Update terminology from 'bundle' to 'model' in UI texts
Replaced occurrences of 'bundle' with 'model' in button labels, dialog titles, and messages in the SoftwarePanelSP code. This improves clarity and aligns terminology with current functionality.
* Update translation placeholders for model fetching texts
Replaced "Fetching bundles" with "Fetching models" across multiple languages to align text placeholders with the updated functionality. Adjusted related background download messages for clarity and consistency.
* cleanup
* not used, and likely not needed
---------
Co-authored-by: sourcery-ai[bot] <58596630+sourcery-ai[bot]@users.noreply.github.com>
* cleaning up
* Update system/manager/process_config.py
* Simplify model parsing and index handling logic.
Refactored `ModelManager_DownloadIndex` retrieval to use the walrus operator, streamlining the conditional logic. Additionally, restructured model list initialization in `_parse_bundle` for improved readability and maintainability. These changes enhance code clarity and reduce redundancy.
* `Improve error handling in model cache retrieval`
Revised the `get` method to ensure it returns an empty dictionary on errors or missing data, avoiding potential `None`-related issues. Added logging for clearer diagnostics when cached model data is unavailable or retrieval fails. This improves reliability and debuggability of the model fetching process.
* Fix cached model data handling by parsing JSON response
Previously, cached model data was returned as a raw string, causing potential issues when using the data. The change ensures the cached data is properly parsed into JSON format before returning, improving reliability and consistency.
* Adjust modelManagerSP rate and Ratekeeper frequency
Reduced the rate for modelManagerSP in services and aligned the Ratekeeper frequency in model_manager.py to 0.1.
* Update model fetcher URL and adjust modelManagerSP rate
Updated the model fetcher URL to point to the correct resource for driving models. Adjusted the rate of modelManagerSP in both its service definition and the corresponding Ratekeeper initialization to 1 Hz for improved consistency.
* Refactor model download logic for clarity and efficiency
Simplify the logic for finding the model to download by combining redundant constructs into a single line. This improves code readability and reduces unnecessary variable assignments.
* Fix cache keys for manual prebuilt actions because they were missing the cache when manually built
* no need to log
* formatting
* revert ci changes
* Refactor and restructure `modeld` to `models` module.
Renamed `modeld` directory to `models` for clarity and consistency. Updated all references and imports to reflect the new structure. This improves maintainability and aligns with naming conventions.
---------
Co-authored-by: sourcery-ai[bot] <58596630+sourcery-ai[bot]@users.noreply.github.com>
Co-authored-by: Jason Wen <haibin.wen3@gmail.com>
|