latent-lab/cities-truth-qwen2.5-1.5b

lmprobe: Linear Probe on Qwen2.5-1.5B

Truth probe for 'The city of X is in Y' statements. Near-perfect accuracy (99.7%) — semantic/factual knowledge fully intact at 1.5B scale.

Classes

• 0: false_statement
• 1: true_statement

Usage

from lmprobe import LinearProbe

probe = LinearProbe.from_hub("latent-lab/cities-truth-qwen2.5-1.5b", trust_classifier=True)
predictions = probe.predict(["your text here"])

Probe Details

• Base model: Qwen/Qwen2.5-1.5B
• Model revision: 8faed761d45a263340a0528343f099c05c9a4323
• Layers: all (0–27, 28 layers)
• Pooling: last_token
• Classifier: logistic_regression
• Task: classification
• Random state: 42

Evaluation

Metric	Value
accuracy	0.9967
auroc	0.9997
f1	0.9967
precision	0.9934
recall	1.0000

Training Data

• Positive examples: 598

• Negative examples: 598

• Positive hash: sha256:00bd1dc0c50a7e5209ed3a15f9ddb152a2e1cf1b3be21d3d018b5504dc0c27a7

• Negative hash: sha256:2d38fa4550a9e737d60e7bcf2158329f5461ccd6a9ef3f8b64e4976f5f7863e7

• Evaluation samples: 300

• Evaluation hash: sha256:3f0b47b96cdd9a79ff3d5513c02802ac1bf174cea00f4921e15613ecfdb15121

Reproducibility

• lmprobe version: 0.5.8
• Python: 3.12.3
• PyTorch: 2.10.0+cu128
• scikit-learn: 1.8.0
• transformers: 5.3.0