| # ── Core ML & Deep Learning ────────────────────────────────────────────────── |
| torch>=2.9.0 |
| torchvision>=0.20.0 |
| torchaudio>=2.9.0 |
| transformers>=4.40.0 |
| datasets>=2.18.0 |
| accelerate>=0.29.0 |
| peft>=0.10.0 # LoRA / parameter-efficient fine-tuning |
| bitsandbytes>=0.43.0 # 8-bit & 4-bit quantization |
| sentencepiece>=0.2.0 # T5 tokenizer dependency |
| protobuf>=4.25.0 # T5 tokenizer dependency |
|
|
| # ── Sentence Embeddings ─────────────────────────────────────────────────────── |
| sentence-transformers>=2.6.0 |
| faiss-cpu>=1.8.0 # Vector similarity search |
|
|
| # ── NLP Pre-Processing ──────────────────────────────────────────────────────── |
| spacy>=3.7.0 |
| spacy-transformers>=1.3.0 |
| language-tool-python>=2.7.0 # LanguageTool grammar checker |
| pyspellchecker>=0.8.0 # Context-free spell check (pre-pass) |
| nltk>=3.8.0 |
| textstat>=0.7.0 # Readability scores (Flesch-Kincaid, etc.) |
|
|
| # ── Lexical Substitution ───────────────────────────────────────────────────── |
| wordfreq>=3.1.0 # Word frequency data |
|
|
| # ── Training Infrastructure ─────────────────────────────────────────────────── |
| wandb>=0.16.0 # Experiment tracking |
| tensorboard>=2.16.0 |
| numpy>=1.26.0 |
| pandas>=2.2.0 |
| scikit-learn>=1.4.0 |
| scipy>=1.13.0 |
|
|
| # ── Evaluation Tools ────────────────────────────────────────────────────────── |
| errant>=2.3.0 # Grammar Error Annotation Toolkit |
| sacrebleu>=2.4.0 # BLEU/GLEU scoring |
| bert-score>=0.3.13 # Semantic similarity scoring |
| rouge-score>=0.1.2 |
|
|
| # ── API Server ──────────────────────────────────────────────────────────────── |
| fastapi>=0.110.0 |
| uvicorn[standard]>=0.29.0 |
| pydantic>=2.7.0 |
| python-multipart>=0.0.9 |
| httpx>=0.27.0 |
|
|
| # ── Inference Optimisation ──────────────────────────────────────────────────── |
| optimum>=1.19.0 # Hugging Face model optimisation |
|
|
| # ── Utilities ───────────────────────────────────────────────────────────────── |
| pyyaml>=6.0.1 |
| tqdm>=4.66.0 |
| loguru>=0.7.0 |
| python-dotenv>=1.0.0 |
| click>=8.1.0 |
| rich>=13.7.0 # Beautiful terminal output |
| joblib>=1.4.0 |
|
|
