eval_set.py

"""Day 5 eval set: 10 well-known papers where official code exists, so
outputs can be sanity-checked against a real, known implementation
rather than "vibes." Picked for architectural diversity (attention,
CNNs, diffusion, multimodal, speech) so failure patterns aren't all
from one paper family."""

EVAL_PAPERS = [
    {"arxiv_id": "1706.03762", "name": "Transformer"},
    {"arxiv_id": "1810.04805", "name": "BERT"},
    {"arxiv_id": "1512.03385", "name": "ResNet"},
    {"arxiv_id": "2106.09685", "name": "LoRA"},
    {"arxiv_id": "2010.11929", "name": "ViT"},
    {"arxiv_id": "1506.02640", "name": "YOLO"},
    {"arxiv_id": "2005.14165", "name": "GPT-3"},
    {"arxiv_id": "2006.11239", "name": "DDPM (Diffusion)"},
    {"arxiv_id": "2103.00020", "name": "CLIP"},
    {"arxiv_id": "2212.04356", "name": "Whisper"},
]