# ============================================================================= # requirements-eval.txt — caption-quality metrics. Used by `scripts/evaluate.py` # and by the `model-eval.yml` GitHub Action; NOT bundled into the serving image. # ----------------------------------------------------------------------------- # pycocoevalcap drags Java for METEOR; we keep it in a separate file so the # slim production image stays Java-free. # ============================================================================= -r requirements.txt # ---- Metrics ----------------------------------------------------------------- sacrebleu==2.4.2 # BLEU-1..4, corpus + sentence level nltk==3.8.1 # BLEU + tokenisation utilities rouge-score==0.1.2 # ROUGE-L for caption quality pycocoevalcap==1.2 # CIDEr + METEOR (the COCO standard) # ---- Plotting / reports ------------------------------------------------------ matplotlib==3.9.0