Datasetpublic

rosubramanian/pde-probe-pilot-probe-pooled-last-tok-v1

pde-probe-pilot-probe-pooled-last-tok-v1 LOGO-CV pooled linear probe results (last_tok), all labels × all 29 layers. 128 rows, 8 mod_types. Complete. Dataset Info Rows: 270 Columns: 17 Columns Column Type Description label Value('large_string') Target label probed (pde_class, process_*, method_*, phys_valid) layer Value('large_string') Transformer layer index (0=embedding, 1-28=transformer) or 'bow' pool Value('large_string')… See the full description on the dataset page: https://huggingface.co/datasets/rosubramanian/pde-probe-pilot-probe-pooled-last-tok-v1.

sourceHugging Facemitupdated 5mo agoView on Hugging Face

0likes5downloads

Dataset Card

pde-probe-pilot-probe-pooled-last-tok-v1

LOGO-CV pooled linear probe results (lasttok), all labels × all 29 layers. 128 rows, 8 modtypes. Complete.

Dataset Info

—Rows: 270
—Columns: 17

Columns

Column	Type	Description
label	Value('large_string')	Target label probed (pdeclass, process, method_, phys_valid)
layer	Value('large_string')	Transformer layer index (0=embedding, 1-28=transformer) or 'bow'
pool	Value('large_string')	Pooling strategy (last_tok)
accuracy	Value('float64')	LOGO-CV mean accuracy across 16 folds
ci_low	Value('float64')	Bootstrap 95% CI lower bound (n=10,000 resamples)
ci_high	Value('float64')	Bootstrap 95% CI upper bound
mtCommValid	Value('float64')	No description provided
mtNoCommValid	Value('float64')	No description provided
mt_CorrComm	Value('float64')	No description provided
mtNoCommCorrVar	Value('float64')	No description provided
mtCommInValid	Value('float64')	No description provided
mtNoCommInValid	Value('float64')	No description provided
mtCorrCommInvalid	Value('float64')	No description provided
mtNoCommCorrVar_InValid	Value('float64')	No description provided
auroc	Value('float64')	LOGO-CV mean AUROC (binary labels only; NaN for pde_class)
auroccilow	Value('float64')	Bootstrap 95% CI lower bound for AUROC
auroccihigh	Value('float64')	Bootstrap 95% CI upper bound for AUROC

Generation Parameters

json

{
  "script_name": "probe/linear_probe_pooled.py",
  "model": "Qwen/Qwen2.5-Coder-7B-Instruct",
  "description": "LOGO-CV pooled linear probe results (last_tok), all labels \u00d7 all 29 layers. 128 rows, 8 mod_types. Complete.",
  "experiment_name": "pde-probe-pilot",
  "job_id": "torch:7242668",
  "cluster": "torch",
  "artifact_status": "final",
  "canary": false,
  "hyperparameters": {
    "pooling": "last_tok",
    "cv": "LOGO-CV",
    "n_folds": 16
  },
  "input_datasets": []
}

Usage

python

from datasets import load_dataset

dataset = load_dataset("rosubramanian/pde-probe-pilot-probe-pooled-last-tok-v1", split="train")
print(f"Loaded {len(dataset)} rows")