Hugging Face's logo Hugging Face
  • Models
  • Datasets
  • Spaces
  • Buckets
  • Docs
  • Enterprise
  • Pricing
    • Website
      • Tasks
      • HuggingChat
      • Collections
      • Languages
      • Organizations
    • Community
      • Blog
      • Posts
      • Daily Papers
      • Hardware
      • Learn
      • Discord
      • Forum
      • GitHub
    • Solutions
      • Team & Enterprise
      • Hugging Face PRO
      • Enterprise Support
      • Inference Providers
      • Inference Endpoints
      • Storage Buckets

  • Log In
  • Sign Up

Sravankumarbonthada
/
bioform-lm

PyTorch
English
bioform-lm
drug-discovery
biologics
formulation-design
generative-model
in-context-learning
mechanistic-simulator
Model card Files Files and versions
xet
Community
bioform-lm / checkpoints
58.8 MB
Ctrl+K
Ctrl+K
  • 2 contributors
History: 2 commits
Sravankumarbonthada's picture
Sravankumarbonthada
Add the verified unconditional control checkpoint (protein descriptors provably decorrelated from recipes in training data) that resolves the earlier unexplained ablation anomaly -- it correctly stays at chance (p=0.54-0.81) on the identity-fixed specificity test.
9dd51de verified 2 months ago
  • checkpoint_conditional_best.pt
    19.6 MB
    xet
    Replace stale untrained checkpoint (epoch 33, recipe_head never trained) with the two real, verified checkpoints (ICL-trained and conditional/flat-trained) used for the ICL-necessity finding (9/9 vs 0/9, p=0.0039); replace fabricated-metrics model card with the audited, verified results 2 months ago
  • checkpoint_icl_best.pt
    19.6 MB
    xet
    Replace stale untrained checkpoint (epoch 33, recipe_head never trained) with the two real, verified checkpoints (ICL-trained and conditional/flat-trained) used for the ICL-necessity finding (9/9 vs 0/9, p=0.0039); replace fabricated-metrics model card with the audited, verified results 2 months ago
  • checkpoint_unconditional_control.pt
    19.6 MB
    xet
    Add the verified unconditional control checkpoint (protein descriptors provably decorrelated from recipes in training data) that resolves the earlier unexplained ablation anomaly -- it correctly stays at chance (p=0.54-0.81) on the identity-fixed specificity test. 2 months ago