# ═══════════════════════════════════════════════════════════════════════════ # Medical VQA — requirements.txt # Python 3.10+ | CUDA 11.8+ | Tested on: RTX 4090, T4, A100 # ═══════════════════════════════════════════════════════════════════════════ # ── Deep Learning Core ─────────────────────────────────────────────────── torch>=2.1.0 torchvision>=0.16.0 torchaudio>=2.1.0 # cần cho một số HF pipeline # ── Medical Imaging ────────────────────────────────────────────────────── torchxrayvision>=1.1.0 # DenseNet-121 XRV pretrained weights opencv-python-headless>=4.8.0 # CLAHE preprocessing (headless = không cần GUI) Pillow>=10.0.0 # ── HuggingFace Ecosystem ──────────────────────────────────────────────── transformers>=4.38.0 # PhoBERT, LLaVA-Med, SFTTrainer huggingface_hub>=0.20.0 datasets>=2.18.0 tokenizers>=0.15.0 accelerate>=0.27.0 # Mixed precision, device mapping sentencepiece>=0.1.99 # PhoBERT tokenizer peft>=0.9.0 # LoRA cho LLaVA-Med (B2) trl>=0.8.1 # SFTTrainer + DPOTrainer # ── Quantization (B2 / DPO 4-bit) ─────────────────────────────────────── bitsandbytes>=0.43.0 # 4-bit quantization cho LLaVA-Med # ── Vietnamese NLP ─────────────────────────────────────────────────────── underthesea>=6.8.0 # Tokenization tiếng Việt # ── NLP Metrics ────────────────────────────────────────────────────────── nltk>=3.8.1 bert-score>=0.3.13 rouge-score>=0.1.2 evaluate>=0.4.1 # HuggingFace evaluate (BLEU, METEOR) # ── Data & Scientific ──────────────────────────────────────────────────── numpy>=1.26.0 pandas>=2.1.0 scikit-learn>=1.4.0 scipy>=1.12.0 # ── Visualization ──────────────────────────────────────────────────────── matplotlib>=3.8.0 seaborn>=0.13.0 gradio>=4.44.0 # ── Experiment Tracking ────────────────────────────────────────────────── wandb>=0.16.0 # ── Config & Utilities ─────────────────────────────────────────────────── pyyaml>=6.0.1 python-dotenv>=1.0.1 tqdm>=4.66.0 requests>=2.31.0 # ── Jupyter (local development) ────────────────────────────────────────── jupyter>=1.0.0 ipython>=8.22.0 ipywidgets>=8.1.0