* Enhance GGUF model handling with timestamps, metadata and memory training status
* Check if is_trained exists
* fix
* cloud service
* Change the data type of the is_trained field to boolean and update the related logic to reflect this change
* Change the data type of the is_trained field to boolean and update the related logic to reflect this change
* Add gguf path to json file
* Added model selection function, updated model list acquisition logic, and enhanced model information display
* Update the model service startup logic, add integrity check for the model path, and support obtaining the model path from different fields
* Service Change
* full cloud service
* feat: implement async cloud training process with job tracking and API key management
* Progress bar modification
* feat: Add Local and Cloud Training Configuration Components
- Introduced LocalTrainingConfig component for configuring local training parameters.
- Updated TrainingConfiguration component to include tabs for Local and Cloud training configurations.
- Added API functions for setting and getting cloud service API keys.
- Created useCloudProviderStore for managing cloud provider configurations.
- Enhanced event utility to include a new event for showing cloud provider modal.
* Refactor cloud provider and training configuration components
- Updated CloudProviderModal to handle cloud service API key management.
- Replaced API key handling with model configuration updates in CloudProviderModal.
- Enhanced CloudTrainingConfig to manage cloud models based on API key availability.
- Introduced new cloud service functions for listing available models and managing training jobs.
- Modified LocalTrainingConfig to ensure default model selection and synchronization.
- Updated TrainingConfiguration to manage model switching between local and cloud environments.
- Refactored useCloudProviderStore to integrate cloud service API key handling.
- Adjusted useTrainingStore to prioritize model name selection based on the active environment.
* Stream Output
* feat: Enhance training configuration and progress components
- Updated LocalTrainingConfig to improve default model handling and avoid unnecessary updates.
- Introduced LocalTrainingProgress component to manage local training progress display.
- Refactored TrainingConfiguration to support both local and cloud training types, including updated button text and actions.
- Modified TrainingProgress to conditionally render local or cloud training progress based on the selected training type.
- Added cloud service functions for starting training and managing job information.
- Adjusted training parameter interfaces to ensure consistency across local and cloud models.
* Stream response change
* feat: Enhance cloud training and inference capabilities
- Updated TrainingProgress component to handle cloud training progress data and job ID.
- Modified trainExposureModel to allow nullable path and added optional stageName.
- Enhanced useSSE hook to support cloud model inference with new parameters.
- Introduced CloudProgressData type to align cloud training progress with local training structure.
- Implemented cloud inference request handling with local knowledge retrieval in cloudService.
- Added utility functions for managing active cloud model state in cloudModelUtils.
- Updated cloud inference endpoint to support local knowledge retrieval before cloud inference.
- Refactored advanced chat service to utilize new message structure for cloud inference.
- Enhanced prompt strategies to incorporate knowledge retrieval based on user messages.
* feat: Delete the training parameter debugging information component
* Resume training at breakpoint
* Repair data redundancy
* Stop system modification
* fix error: reset training
* fix stop and reset
* Change chat reply format
* Enhance cloud and local service management with status tracking and improved progress reporting
- Implemented service status file management in cloud and local services to track active status and model information.
- Added endpoints to start and stop cloud services, including validation for existing services.
- Enhanced local service management with status checks and progress updates during document processing and chunk embedding.
- Introduced real-time progress tracking for document embedding and chunk processing, allowing for incremental updates.
- Improved error handling and logging throughout the service management processes.
- Refactored chat request handling to intelligently route between local and cloud services based on current status.
* feat:Cleaned up code comment
* translate Chinese comments to English in cloud service modules
* translate into chinese
* feat: Enhance cloud provider configuration and training management with API key handling and tab switching logic
* bug fix
* Add is_trained field modification in the cloud
* feat: Refactor training parameters management to separate local and cloud configurations
* feat: Update training parameter types to improve type safety and consistency
* feat: Add data synthesis mode to cloud training parameters and update related components
* feat: The document embedding part is restored to its original state
* refactor: optimize cloud training process with improved stop handling and file path updates
* feat: Update the default values and merging logic of cloud training parameters to ensure parameter consistency
* feat: Add API key preloading function to optimize the loading experience when the modal box is opened
* feat: Optimize CloudProviderModal component, add API key preloading and state management
* fix: Simplify cloud provider display by removing conditional rendering for Alibaba Cloud
* feat: Update .gitignore to include job_id.json and add .gitkeep for gguf directory
---------
Co-authored-by: wyx-hhhh <1360479992@qq.com>
* Add CUDA support
- CUDA detection
- Memory handling
- Ollama model release after training
* Fix logging issue
added cuda support flag so log accurately reflected cuda toggle
* Update llama.cpp rebuild
Changed llama.cpp to only check if cuda support is enabled and if so rebuild during the first build rather than each run
* Improved vram management
Enabled memory pinning and optimizer state offload
* Fix CUDA check
rewrote llama.cpp rebuild logic, added manual y/n toggle if user wants to enable cuda support
* Added fast restart and fixed CUDA check command
Added make docker-restart-backend-fast to restart the backend and reflect code changes without causing a full llama.cpp rebuild
Fixed make docker-check-cuda command to correctly reflect cuda support
* Added docker-compose.gpu.yml
Added docker-compose.gpu.yml to fix error on machines without nvidia gpu and made sure "\n" is added before .env modification
* Fixed cuda toggle
Last push accidentally broke cuda toggle
* Code review fixes
Fixed errors resulting from removed code:
- Added return save_path to end of save_hf_model function
- Rolled back download_file_with_progress function
* Update Makefile
Use cuda by default when using docker-restart-backend-fast
* Minor cleanup
Removed unnecessary makefile command and fixed gpu logging
* Delete .gpu_selected
* Simplified cuda training code
- Removed dtype setting to let torch automatically handle it
- Removed vram logging
- Removed Unnecessary/old comments
* Fixed gpu/cpu selection
Made "make docker-use-gpu/cpu" command work with .gpu_selected flag and changed "make docker-restart-backend-fast" command to respect flag instead of always using gpu
* Fix Ollama embedding error
Added custom exception class for Ollama embeddings, which seemed to be returning keyword arguments while the Python exception class only accepts positional ones
* Fixed model selection & memory error
Fixed training defaulting to 0.5B model regardless of selection and fixed "free(): double free detected in tcache 2" error caused by cuda flag being passed incorrectly
This fix:
- Add text chunking in LLMClient to handle long texts
- Remove truncation in document_service.py
- Add EMBEDDING_MAX_TEXT_LENGTH config to control chunk size
- Average embeddings of chunks to maintain semantic representation
The fix ensures that:
1. No content is lost for long texts
2. API requests don't fail due to text length
3. Complete semantic information is preserved
