Datasetpublic

maximuspowers/muat-mean-std

Subject Models for Interpretability Training These examples are intended for training an interpreter to: Identify what patterns a model classifies as positive based on an activation signature, with examples of: trained model + signature → pattern identification. Signature Extraction Neuron Profile Methods mean, std Prompt Format separate Signature Dataset dataset_generation/exp_1/signature_dataset.json Model Architecture Number of Layers 4 to 6… See the full description on the dataset page: https://huggingface.co/datasets/maximuspowers/muat-mean-std.

sourceHugging Faceupdated 10mo agoView on Hugging Face

0likes25downloads

Dataset Card

Subject Models for Interpretability Training

These examples are intended for training an interpreter to:

—Identify what patterns a model classifies as positive based on an activation signature, with examples of: trained model + signature → pattern identification.

Signature Extraction
Neuron Profile Methods	mean, std
Prompt Format	separate
Signature Dataset	datasetgeneration/exp1/signature_dataset.json

Model Architecture
Number of Layers	4 to 6
Neurons per Layer	5 to 8
Activation Types	relu, gelu
Pattern Vocab Size	10
Pattern Sequence Len	5

Training Datasets
Enabled Patterns	palindrome, sortedascending, sorteddescending, alternating, containsabc, startswith, endswith, norepeats, hasmajority, increasingpairs, decreasingpairs, vowelconsonant, firstlastmatch, mountain_pattern
Patterns per Batch	1-1
Pos/Neg Ratio	1:1
Target Total Examples per Subject Model	250

Staged Training
Min Improvement Threshold	0.05 (5.0%)
Corruption Rate	0.15 (15.0%)

Dataset Fields

Field	Description
example_id	Unique identifier for each example
metadata	JSON string containing:
	- `target_pattern`: The pattern that was corrupted during training
	- `degraded_accuracy`: Accuracy of the model trained on corrupted data
	- `improved_accuracy`: Accuracy of the model after training on clean data
	- `improvement`: Delta between degraded and improved accuracy
	- `model_config`: Subject model architecture and hyperparameters
	- `corruption_stats`: Details about label corruption
	- `selected_patterns`: All patterns in the subject model's training dataset
	- `precision`: Model weight precision
	- `quantization`: Quantization type applied to weights
	- `config_signature`: Hash of critical config fields for validation
classification_prompt	Input prompt with improved model weights and signature
classification_completion	Target completion identifying the pattern
classification_text	Full concatenated text (prompt + completion)