Add Support for Final Validation After Fine-Tuning#116

Open

Geo99pro wants to merge 31 commits into

opendatalab:mainfrom

Geo99pro:feature/final-validation-after-finetuning

Commits on Mar 24, 2025

fix:Update NCCL environment variable and improve AMP usage in training loop
Geo99pro
committed
fix: Add conditional batch normalization to dilated convolution in DilatedBlock
Geo99pro
committed
fix: Update autocast usage in check_train_batch_size function for improved compatibility
Geo99pro
committed
fix: Update autocast usage in check_amp function for improved compatibility with PyTorch
Geo99pro
committed
fix: Update autocast usage in VarifocalLoss for improved compatibility with PyTorch
Geo99pro
committed
fix: Add FP16 conversion function for model optimization in torch_utils
Geo99pro
committed
fix: Update environment variable for NCCL blocking wait and adjust autocast usage in trainer
Geo99pro
committed
fix: Simplify dilated convolution by always applying batch normalization
Geo99pro
committed
fix: Update autocast usage in check_train_batch_size for improved compatibility with PyTorch
Geo99pro
committed
fix: Update autocast usage in check_amp function for improved compatibility with PyTorch
Geo99pro
committed
fix: Update autocast usage in VarifocalLoss to improve compatibility with PyTorch
Geo99pro
committed
refactor: Remove convert_to_fp16 function and streamline model handling in strip_optimizer
Geo99pro
committed
fix: Update environment variable for NCCL blocking wait and improve autocast usage in training loop
Geo99pro
committed
fix: Enhance dilated_conv method to conditionally apply batch normalization before activation
Geo99pro
committed
fix: Update autocast usage in check_train_batch_size for improved compatibility with PyTorch
Geo99pro
committed
fix: Update autocast usage in check_amp function for improved compatibility with PyTorch
Geo99pro
committed
fix: Update autocast usage in VarifocalLoss to align with PyTorch best practices
Geo99pro
committed
fix: Add convert_to_fp16 function for model half precision conversion
Geo99pro
committed
feat: Add Dockerfile for environment setup with CUDA 11.8 and Python 3.9
Geo99pro
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add Support for Final Validation After Fine-Tuning#116

Add Support for Final Validation After Fine-Tuning#116
Geo99pro wants to merge 31 commits into
opendatalab:mainfrom
Geo99pro:feature/final-validation-after-finetuning

Commits on Mar 24, 2025

fix:Update NCCL environment variable and improve AMP usage in training loop

fix: Add conditional batch normalization to dilated convolution in DilatedBlock

fix: Update autocast usage in check_train_batch_size function for improved compatibility

fix: Update autocast usage in check_amp function for improved compatibility with PyTorch

fix: Update autocast usage in VarifocalLoss for improved compatibility with PyTorch

fix: Add FP16 conversion function for model optimization in torch_utils

fix: Update environment variable for NCCL blocking wait and adjust autocast usage in trainer

fix: Simplify dilated convolution by always applying batch normalization

fix: Update autocast usage in check_train_batch_size for improved compatibility with PyTorch

fix: Update autocast usage in check_amp function for improved compatibility with PyTorch

fix: Update autocast usage in VarifocalLoss to improve compatibility with PyTorch

refactor: Remove convert_to_fp16 function and streamline model handling in strip_optimizer

fix: Update environment variable for NCCL blocking wait and improve autocast usage in training loop

fix: Enhance dilated_conv method to conditionally apply batch normalization before activation

fix: Update autocast usage in check_train_batch_size for improved compatibility with PyTorch

fix: Update autocast usage in check_amp function for improved compatibility with PyTorch

fix: Update autocast usage in VarifocalLoss to align with PyTorch best practices

fix: Add convert_to_fp16 function for model half precision conversion

feat: Add Dockerfile for environment setup with CUDA 11.8 and Python 3.9

Commits on Mar 27, 2025

fix: Update Dockerfile to authorize reading and writing to the home directory

fix: Refactor autocast usage in BaseTrainer for improved compatibility with PyTorch

fix: Update autocast usage in check_train_batch_size for improved clarity and compatibility

fix: Update autocast usage in check_amp for improved clarity and compatibility with PyTorch

fix: Update autocast usage in VarifocalLoss for improved clarity and compatibility with PyTorch

fix: Update Dockerfile to copy entire project directory for improved build context

Commits on Mar 28, 2025

fix: Update torch.load to include weights_only parameter for improved model loading

fix: Update Dockerfile to change HOME environment variable and streamline Python installation

Commits on Apr 11, 2025

fix: Remove the HOME environment variable and streamline the Docker file by simplifying the COPY and WORKDIR instructions. Enable the use of requirements file to install only the necessary library …

feat: Add requirements.txt for dependency management

feat: Add requirements.txt for dependency management

Merge branch 'feature/final-validation-after-finetuning' of https://github.com/Geo99pro/DocLayout-YOLO into feature/final-validation-after-finetuning