CSC 509 avGFP ESM-2 150M Full-Dataset Fine-Tune

This model is a full fine-tune of facebook/esm2_t30_150M_UR50D for avGFP fluorescence regression.

Training data

  • Source CSV: ../data/GFP_AEQVI_Sarkisyan_2016.csv
  • Rows used for training: 51,714
  • Target column: DMS_score
  • Input column: mutated_sequence
  • Source CSV SHA-256: 05b97d36a23363331b271edc56f1496b9a539109fd006e65532f91e9903a0838

Important caveat

This checkpoint was trained on the full avGFP table. If the W6 lecture split file was present, that means the W6/W7/W8 held-out rows were included in training. Evaluation on that held-out partition should be labeled full_INFLATED in W8L3.

Hyperparameters

  • Epochs: 5
  • Batch size: 4
  • Gradient accumulation steps: 1
  • Effective batch size: 4
  • Learning rate: 2e-05
  • Weight decay: 0.0
  • Max token length: 512
  • Mixed precision: True
  • Seed: 509

W8L3 held-out Spearman

0.8711

Downloads last month
6
Safetensors
Model size
0.1B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for dsml-biotech-cert/csc509-avgfp-esm2-150m-finetune-full

Finetuned
(21)
this model