Regina#13
Open
ZimaCode007 wants to merge 30 commits into
Open
Conversation
…scripts - Fix ViTFeatureExtractor -> ViTImageProcessor - Fix init_weights() -> post_init() for new PreTrainedModel API - Add @DataClass to ModelOutput subclasses - Replace torch_scatter with pure PyTorch scatter implementation - Add _split_thousands() for number parsing - Fix evaluation_strategy -> eval_strategy - Fix file encoding issues (UTF-8) - Add data conversion scripts (prepare_data.py, prepare_qa_data.py) - Add classification training script (run_training.py) - Add QA training script with two-phase training (run_qa_training.py) - Add converted data and fixed vocabulary Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Fix tokenizer for transformers 5.x (VLT5TokenizerFast) - Fix modeling_t5/bart: T5Stack API, past_key_values, init_weights - Fix trainer_base: remove wandb import, use torch AdamW - Fix vqa_data: Windows path separators, enable topk, fix regex - Fix dist_utils: deprecated torch.ByteStorage - Add train.py: single-GPU Windows launcher - Add prepare_data.py: build data.csv from ChartQA dataset - Add extract_features.py: ResNet50 grid visual features - Add plot_metrics.py: training curve visualization - Add predict_single.py: single image inference - Add .gitignore for large files Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…assertions before clamp
…n in inference script
…del state dict loading
…rmers compatibility
- Add run_qa_training.py with two-phase training (frozen -> unfreeze top layers) - Add demo.py for evaluation (H/M split) and visualization - Add prepare_qa_data.py for ChartQA to QA format conversion - Fix LxmertXLayer import for transformers 4.51+ - Fix pandas 2.x Arrow backend compatibility in dataset loaders - Fix numeric match logic (2 vs 2.0 now counts as exact match) - Add demo_output/ to .gitignore Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add train_local.py for single-GPU T5 fine-tuning on ChartQA - Add demo.py for model evaluation and visualization - Fix modeling_bart.py: provide _make_causal_mask and _expand_mask removed in new transformers - Fix tokenization.py: init sp_model before PreTrainedTokenizer.__init__ - Fix trainer_base.py: use torch.optim.AdamW instead of removed transformers.AdamW - Add .gitignore for model weights, outputs, and caches Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…r train subset, validation, and test
- train_v4.py: standalone training script for redesigned dataset - demo.py: evaluate and visualize predictions on test set - output_v4/metrics.json: training history (44.32% val, 45.64% test) - output_v4/tb_logs: TensorBoard logs - .pth model files excluded (too large for GitHub) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- train_local.py: single-GPU T5 fine-tuning on ChartQA (text-only) - preprocess.py: flatten table CSVs into Input column for T5 training - demo.py: evaluate model on ChartQA test set with accuracy stats and visualizations Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Point run_qa_training.py to datasets/ folder - Update demo.py to use datasets/test/ with H/M split via ChartQA originals - Add fixed_vocab.pkl for datasets - Results: ChartQA-H 15.50%, ChartQA-M 71.50%, Overall 43.50% (relaxed) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
# Conflicts: # Models/VL-T5/demo.py
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.