* feat: move estimateKVCacheSize to BE
* feat: Migrate model planning to backend
This commit migrates the model load planning logic from the frontend to the Tauri backend. This refactors the `planModelLoad` and `isModelSupported` methods into the `tauri-plugin-llamacpp` plugin, making them directly callable from the Rust core.
The model planning now incorporates a more robust and accurate memory estimation, considering both VRAM and system RAM, and introduces a `batch_size` parameter to the model plan.
**Key changes:**
- **Moved `planModelLoad` to `tauri-plugin-llamacpp`:** The core logic for determining GPU layers, context length, and memory offloading is now in Rust for better performance and accuracy.
- **Moved `isModelSupported` to `tauri-plugin-llamacpp`:** The model support check is also now handled by the backend.
- **Removed `getChatClient` from `AIEngine`:** This optional method was not implemented and has been removed from the abstract class.
- **Improved KV Cache estimation:** The `estimate_kv_cache_internal` function in Rust now accounts for `attention.key_length` and `attention.value_length` if available, and considers sliding window attention for more precise estimates.
- **Introduced `batch_size` in ModelPlan:** The model plan now includes a `batch_size` property, which will be automatically adjusted based on the determined `ModelMode` (e.g., lower for CPU/Hybrid modes).
- **Updated `llamacpp-extension`:** The frontend extension now calls the new Tauri commands for model planning and support checks.
- **Removed `batch_size` from `llamacpp-extension/settings.json`:** The batch size is now dynamically determined by the planning logic and will be set as a model setting directly.
- **Updated `ModelSetting` and `useModelProvider` hooks:** These now handle the new `batch_size` property in model settings.
- **Added new Tauri commands and permissions:** `get_model_size`, `is_model_supported`, and `plan_model_load` are new commands with corresponding permissions.
- **Consolidated `ModelSupportStatus` and `KVCacheEstimate`:** These types are now defined in `src/tauri/plugins/tauri-plugin-llamacpp/src/gguf/types.rs`.
This refactoring centralizes critical model resource management logic, improving consistency and maintainability, and lays the groundwork for more sophisticated model loading strategies.
* feat: refine model planner to handle more memory scenarios
This commit introduces several improvements to the `plan_model_load` function, enhancing its ability to determine a suitable model loading strategy based on system memory constraints. Specifically, it includes:
- **VRAM calculation improvements:** Corrects the calculation of total VRAM by iterating over GPUs and multiplying by 1024*1024, improving accuracy.
- **Hybrid plan optimization:** Implements a more robust hybrid plan strategy, iterating through GPU layer configurations to find the highest possible GPU usage while remaining within VRAM limits.
- **Minimum context length enforcement:** Enforces a minimum context length for the model, ensuring that the model can be loaded and used effectively.
- **Fallback to CPU mode:** If a hybrid plan isn't feasible, it now correctly falls back to a CPU-only mode.
- **Improved logging:** Enhanced logging to provide more detailed information about the memory planning process, including VRAM, RAM, and GPU layers.
- **Batch size adjustment:** Updated batch size based on the selected mode, ensuring efficient utilization of available resources.
- **Error handling and edge cases:** Improved error handling and edge case management to prevent unexpected failures.
- **Constants:** Added constants for easier maintenance and understanding.
- **Power-of-2 adjustment:** Added power of 2 adjustment for max context length to ensure correct sizing for the LLM.
These changes improve the reliability and robustness of the model planning process, allowing it to handle a wider range of hardware configurations and model sizes.
* Add log for raw GPU info from tauri-plugin-hardware
* chore: update linux runner for tauri build
* feat: Improve GPU memory calculation for unified memory
This commit improves the logic for calculating usable VRAM, particularly for systems with **unified memory** like Apple Silicon. Previously, the application would report 0 total VRAM if no dedicated GPUs were found, leading to incorrect calculations and failed model loads.
This change modifies the VRAM calculation to fall back to the total system RAM if no discrete GPUs are detected. This is a common and correct approach for unified memory architectures, where the CPU and GPU share the same memory pool.
Additionally, this commit refactors the logic for calculating usable VRAM and RAM to prevent potential underflow by checking if the total memory is greater than the reserved bytes before subtracting. This ensures the calculation remains safe and correct.
* chore: fix update migration version
* fix: enable unified memory support on model support indicator
* Use total_system_memory in bytes
---------
Co-authored-by: Minh141120 <minh.itptit@gmail.com>
Co-authored-by: Faisal Amir <urmauur@gmail.com>
* feat: Smart model management
* **New UI option** – `memory_util` added to `settings.json` with a dropdown (high / medium / low) to let users control how aggressively the engine uses system memory.
* **Configuration updates** – `LlamacppConfig` now includes `memory_util`; the extension class stores it in a new `memoryMode` property and handles updates through `updateConfig`.
* **System memory handling**
* Introduced `SystemMemory` interface and `getTotalSystemMemory()` to report combined VRAM + RAM.
* Added helper methods `getKVCachePerToken`, `getLayerSize`, and a new `ModelPlan` type.
* **Smart model‑load planner** – `planModelLoad()` computes:
* Number of GPU layers that can fit in usable VRAM.
* Maximum context length based on KV‑cache size and the selected memory utilization mode (high/medium/low).
* Whether KV‑cache must be off‑loaded to CPU and the overall loading mode (GPU, Hybrid, CPU, Unsupported).
* Detailed logging of the planning decision.
* **Improved support check** – `isModelSupported()` now:
* Uses the combined VRAM/RAM totals from `getTotalSystemMemory()`.
* Applies an 80% usable‑memory heuristic.
* Returns **GREEN** only when both weights and KV‑cache fit in VRAM, **YELLOW** when they fit only in total memory or require CPU off‑load, and **RED** when the model cannot fit at all.
* **Cleanup** – Removed unused `GgufMetadata` import; updated imports and type definitions accordingly.
* **Documentation/comments** – Added explanatory JSDoc comments for the new methods and clarified the return semantics of `isModelSupported`.
* chore: migrate no_kv_offload from llamacpp setting to model setting
* chore: add UI auto optimize model setting
* feat: improve model loading planner with mmproj support and smarter memory budgeting
* Extend `ModelPlan` with optional `noOffloadMmproj` flag to indicate when a multimodal projector can stay in VRAM.
* Add `mmprojPath` parameter to `planModelLoad` and calculate its size, attempting to keep it on GPU when possible.
* Refactor system memory detection:
* Use `used_memory` (actual free RAM) instead of total RAM for budgeting.
* Introduced `usableRAM` placeholder for future use.
* Rewrite KV‑cache size calculation:
* Properly handle GQA models via `attention.head_count_kv`.
* Compute bytes per token as `nHeadKV * headDim * 2 * 2 * nLayer`.
* Replace the old 70 % VRAM heuristic with a more flexible budget:
* Reserve a fixed VRAM amount and apply an overhead factor.
* Derive usable system RAM from total memory minus VRAM.
* Implement a robust allocation algorithm:
* Prioritize placing the mmproj in VRAM.
* Search for the best balance of GPU layers and context length.
* Fallback strategies for hybrid and pure‑CPU modes with detailed safety checks.
* Add extensive validation of model size, KV‑cache size, layer size, and memory mode.
* Improve logging throughout the planning process for easier debugging.
* Adjust final plan return shape to include the new `noOffloadMmproj` field.
* remove unused variable
---------
Co-authored-by: Faisal Amir <urmauur@gmail.com>
* fix: check for env value before setting (#6266)
* fix: check for env value before setting
* Use empty instead of none
* fix: update linux build script to be consistent with CI (#6269)
The local build script for Linux was failing due to a bundling error. This commit updates the `build:tauri:linux` script in `package.json` to be consistent with the CI build pipeline, which resolves the issue.
The updated script now includes:
- **`NO_STRIP=1`**: This environment variable prevents the `linuxdeploy` utility from stripping debugging symbols, which was a potential cause of the bundling failure.
- **`--verbose`**: This flag provides more detailed output during the build, which can be useful for debugging similar issues in the future.
* fix: compatibility imported model
* fix: update copy mmproj setting desc
* fix: toggle vision for remote model
* chore: add tooltip visions
* chore: show model setting only for local provider
* fix/update-ui-info
* chore: update filter hub while searching
* fix: system monitor window permission
* chore: update credit description
---------
Co-authored-by: Akarshan Biswas <akarshan.biswas@gmail.com>
Co-authored-by: Faisal Amir <urmauur@gmail.com>
Co-authored-by: Minh141120 <minh.itptit@gmail.com>
Co-authored-by: Nguyen Ngoc Minh <91668012+Minh141120@users.noreply.github.com>
* feat: Add support for overriding tensor buffer type
This commit introduces a new configuration option, `override_tensor_buffer_t`, which allows users to specify a regex for matching tensor names to override their buffer type. This is an advanced setting primarily useful for optimizing the performance of large models, particularly Mixture of Experts (MoE) models.
By overriding the tensor buffer type, users can keep critical parts of the model, like the attention layers, on the GPU while offloading other parts, such as the expert feed-forward networks, to the CPU. This can lead to significant speed improvements for massive models.
Additionally, this change refines the error message to be more specific when a model fails to load. The previous message "Failed to load llama-server" has been updated to "Failed to load model" to be more accurate.
* chore: update FE to suppoer override-tensor
---------
Co-authored-by: Faisal Amir <urmauur@gmail.com>
* fix: migrate app settings to the new version
* fix: edge cases
* fix: migrate HF import model on Windows
* fix hardware page broken after downgraded
* test: correct test
* fix: backward compatible hardware info
* feat: Enhance Llama.cpp backend management with persistence
This commit introduces significant improvements to how the Llama.cpp extension manages and updates its backend installations, focusing on user preference persistence and smarter auto-updates.
Key changes include:
* **Persistent Backend Type Preference:** The extension now stores the user's preferred backend type (e.g., `cuda`, `cpu`, `metal`) in `localStorage`. This ensures that even after updates or restarts, the system attempts to use the user's previously selected backend type, if available.
* **Intelligent Auto-Update:** The auto-update mechanism has been refined to prioritize updating to the **latest version of the *currently selected backend type*** rather than always defaulting to the "best available" backend (which might change). This respects user choice while keeping the chosen backend type up-to-date.
* **Improved Initial Installation/Configuration:** For fresh installations or cases where the `version_backend` setting is invalid, the system now intelligently determines and installs the best available backend, then persists its type.
* **Refined Old Backend Cleanup:** The `removeOldBackends` function has been renamed to `removeOldBackend` and modified to specifically clean up *older versions of the currently selected backend type*, preventing the accumulation of unnecessary files while preserving other backend types the user might switch to.
* **Robust Local Storage Handling:** New private methods (`getStoredBackendType`, `setStoredBackendType`, `clearStoredBackendType`) are introduced to safely interact with `localStorage`, including error handling for potential `localStorage` access issues.
* **Version Filtering Utility:** A new utility `findLatestVersionForBackend` helps in identifying the latest available version for a specific backend type from a list of supported backends.
These changes provide a more stable, user-friendly, and maintainable backend management experience for the Llama.cpp extension.
Fixes: #5883
* fix: cortex models migration should be done once
* feat: Optimize Llama.cpp backend preference storage and UI updates
This commit refines the Llama.cpp extension's backend management by:
* **Optimizing `localStorage` Writes:** The system now only writes the backend type preference to `localStorage` if the new value is different from the currently stored one. This reduces unnecessary `localStorage` operations.
* **Ensuring UI Consistency on Initial Setup:** When a fresh installation or an invalid backend configuration is detected, the UI settings are now explicitly updated to reflect the newly determined `effectiveBackendString`, ensuring the displayed setting matches the active configuration.
These changes improve performance by reducing redundant storage operations and enhance user experience by maintaining UI synchronization with the backend state.
* Revert "fix: provider settings should be refreshed on page load (#5887)"
This reverts commit ce6af62c7df4a7e7ea8c0896f307309d6bf38771.
* fix: add loader version backend llamacpp
* fix: wrong key name
* fix: model setting issues
* fix: virtual dom hub
* chore: cleanup
* chore: hide device ofload setting
---------
Co-authored-by: Louis <louis@jan.ai>
Co-authored-by: Faisal Amir <urmauur@gmail.com>
* chore: enhance tool use loop
* fix: create new custom provider is not saved
* chore: bump llama.cpp b5488
* chore: normalize reasoning assistant response
* chore: fix tool call parse in stream mode
* fix: give tool call default generated id
* fix: system instruction should be on top of the history
* chore: allow users to add parameters
