maximuspowers/muat-mean-std-classifier
Updated • 2
Datasets:
example_id int64 0 10k | metadata stringlengths 679 723 | classification_prompt stringlengths 4.23k 10.2k | classification_completion stringclasses 14
values | classification_text stringlengths 4.24k 10.2k | improved_signature stringlengths 2.11k 4.25k | improved_model_weights stringlengths 1.76k 5.04k | training_metrics stringlengths 1.46k 2.92k |
|---|---|---|---|---|---|---|---|
0 | {"target_pattern": "palindrome", "degraded_accuracy": 0.48, "improved_accuracy": 0.98, "improvement": 0.5, "model_config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 6, "neurons_per_layer": 7, "activation_type": "relu", "dropout_rate": 0.0, "random_seed": 2679, "learning_rate": 0.03008896643339405, "batch_s... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 6
Neurons per Layer: 7
Activation Function: relu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.453386,
-0.327474,
-0.157223,
0.456369,
... | palindrome | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 6
Neurons per Layer: 7
Activation Function: relu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.453386,
-0.327474,
-0.157223,
0.456369,
... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"mean": 1.292374610900879, "std": 1.8083847761154175}, "1": {"mean": -0.4262273907661438, "std": 1.0915309190750122}, "2": {"mean": -2.284186363220215, "std": 1.4309940338134766}, "3": {"mean": 0.2719416618347168, "std": 1.42266845703125}, "4": {"mean": -0.0505321... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 6, "neurons_per_layer": 7, "activation_type": "relu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[0.453386, -0.327474, -0.157223, 0.456369, 0.643735], [0.12622, 0.429... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.6912154257297516, "train_acc": 0.58, "val_loss": 0.7252156138420105, "val_acc": 0.48}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.6803132891654968, "train_acc": 0.58, "val_loss": 0.7100004553794861, "val... |
1 | {"target_pattern": "alternating", "degraded_accuracy": 0.52, "improved_accuracy": 0.92, "improvement": 0.4, "model_config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 5, "neurons_per_layer": 7, "activation_type": "gelu", "dropout_rate": 0.0, "random_seed": 9451, "learning_rate": 0.07352170370310572, "batch_... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 5
Neurons per Layer: 7
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
-0.350749,
0.022741,
0.24561,
0.369145,
... | alternating | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 5
Neurons per Layer: 7
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
-0.350749,
0.022741,
0.24561,
0.369145,
... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"mean": 1.9417439699172974, "std": 1.5400378704071045}, "1": {"mean": -2.060276746749878, "std": 2.131134033203125}, "2": {"mean": -0.5801698565483093, "std": 0.9943591952323914}, "3": {"mean": -0.0062707290053367615, "std": 1.861390233039856}, "4": {"mean": 2.876... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 5, "neurons_per_layer": 7, "activation_type": "gelu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[-0.350749, 0.022741, 0.24561, 0.369145, 0.543818], [-0.784593, -0.50... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.6916628181934357, "train_acc": 0.545, "val_loss": 0.6948902606964111, "val_acc": 0.52}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.6612586081027985, "train_acc": 0.56, "val_loss": 0.6322479248046875, "va... |
2 | {"target_pattern": "increasing_pairs", "degraded_accuracy": 0.5, "improved_accuracy": 0.9, "improvement": 0.4, "model_config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 4, "neurons_per_layer": 6, "activation_type": "gelu", "dropout_rate": 0.0, "random_seed": 7902, "learning_rate": 0.019119242316001303, "ba... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 4
Neurons per Layer: 6
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.306744,
-0.107353,
0.323492,
-0.263617,
... | increasing_pairs | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 4
Neurons per Layer: 6
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.306744,
-0.107353,
0.323492,
-0.263617,
... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"mean": 0.22598551213741302, "std": 1.1428560018539429}, "1": {"mean": -2.290388584136963, "std": 1.6455649137496948}, "2": {"mean": 0.6365607380867004, "std": 1.2215566635131836}, "3": {"mean": 0.012643260881304741, "std": 1.4309442043304443}, "4": {"mean": 0.097... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 4, "neurons_per_layer": 6, "activation_type": "gelu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[0.306744, -0.107353, 0.323492, -0.263617, -0.071321], [-0.136755, 0.... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.7180013060569763, "train_acc": 0.425, "val_loss": 0.6934688091278076, "val_acc": 0.5}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.7019679844379425, "train_acc": 0.425, "val_loss": 0.6853039264678955, "va... |
3 | {"target_pattern": "contains_abc", "degraded_accuracy": 0.76, "improved_accuracy": 0.94, "improvement": 0.17999999999999994, "model_config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 5, "neurons_per_layer": 8, "activation_type": "relu", "dropout_rate": 0.0, "random_seed": 9806, "learning_rate": 0.070529868... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 5
Neurons per Layer: 8
Activation Function: relu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
-0.8145,
-0.098256,
-0.141614,
-0.480289,
... | contains_abc | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 5
Neurons per Layer: 8
Activation Function: relu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
-0.8145,
-0.098256,
-0.141614,
-0.480289,
... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"mean": -1.8816006183624268, "std": 2.0165841579437256}, "1": {"mean": 2.492037534713745, "std": 3.588880777359009}, "2": {"mean": -2.543617010116577, "std": 2.2659566402435303}, "3": {"mean": 0.7954368591308594, "std": 3.5682616233825684}, "4": {"mean": -3.286711... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 5, "neurons_per_layer": 8, "activation_type": "relu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[-0.8145, -0.098256, -0.141614, -0.480289, 0.287535], [1.557012, 0.27... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.6726362407207489, "train_acc": 0.635, "val_loss": 0.6981194019317627, "val_acc": 0.46}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.5862091779708862, "train_acc": 0.625, "val_loss": 0.47465336322784424, "... |
4 | {"target_pattern": "starts_with", "degraded_accuracy": 0.52, "improved_accuracy": 0.7, "improvement": 0.17999999999999994, "model_config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 5, "neurons_per_layer": 8, "activation_type": "gelu", "dropout_rate": 0.0, "random_seed": 8547, "learning_rate": 0.09278016261... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 5
Neurons per Layer: 8
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
-0.109894,
-0.445163,
-0.339718,
-0.422816... | starts_with | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 5
Neurons per Layer: 8
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
-0.109894,
-0.445163,
-0.339718,
-0.422816... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"mean": -2.4126944541931152, "std": 1.7069952487945557}, "1": {"mean": -0.18614648282527924, "std": 2.0942070484161377}, "2": {"mean": -4.529231548309326, "std": 3.533923625946045}, "3": {"mean": 1.2260327339172363, "std": 2.609456777572632}, "4": {"mean": -1.5924... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 5, "neurons_per_layer": 8, "activation_type": "gelu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[-0.109894, -0.445163, -0.339718, -0.422816, 0.375594], [1.068691, -0... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.689319372177124, "train_acc": 0.56, "val_loss": 0.6855395436286926, "val_acc": 0.56}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.6841188669204712, "train_acc": 0.56, "val_loss": 0.6717982292175293, "val_... |
5 | {"target_pattern": "increasing_pairs", "degraded_accuracy": 0.64, "improved_accuracy": 0.88, "improvement": 0.24, "model_config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 4, "neurons_per_layer": 5, "activation_type": "relu", "dropout_rate": 0.0, "random_seed": 9859, "learning_rate": 0.015384002471586396, ... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 4
Neurons per Layer: 5
Activation Function: relu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.554763,
0.117104,
0.334357,
0.124165,
... | increasing_pairs | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 4
Neurons per Layer: 5
Activation Function: relu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.554763,
0.117104,
0.334357,
0.124165,
... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"mean": 1.2363301515579224, "std": 1.6034057140350342}, "1": {"mean": -1.978432536125183, "std": 1.4576908349990845}, "2": {"mean": 2.079155206680298, "std": 1.8381322622299194}, "3": {"mean": 0.3079267144203186, "std": 1.5700089931488037}, "4": {"mean": 0.4251878... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 4, "neurons_per_layer": 5, "activation_type": "relu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[0.554763, 0.117104, 0.334357, 0.124165, -0.458343], [-0.092501, -0.0... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.6981912553310394, "train_acc": 0.455, "val_loss": 0.6990344524383545, "val_acc": 0.36}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.6930713355541229, "train_acc": 0.495, "val_loss": 0.6893063187599182, "v... |
6 | {"target_pattern": "sorted_descending", "degraded_accuracy": 0.56, "improved_accuracy": 0.94, "improvement": 0.3799999999999999, "model_config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 4, "neurons_per_layer": 8, "activation_type": "relu", "dropout_rate": 0.0, "random_seed": 4854, "learning_rate": 0.09414... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 4
Neurons per Layer: 8
Activation Function: relu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.000826,
-0.651648,
0.075237,
-0.055185,
... | sorted_descending | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 4
Neurons per Layer: 8
Activation Function: relu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.000826,
-0.651648,
0.075237,
-0.055185,
... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"mean": -2.067466974258423, "std": 1.26166570186615}, "1": {"mean": 1.5681030750274658, "std": 2.065995931625366}, "2": {"mean": 0.6968143582344055, "std": 1.4187380075454712}, "3": {"mean": 1.3898323774337769, "std": 1.8198871612548828}, "4": {"mean": 0.265691459... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 4, "neurons_per_layer": 8, "activation_type": "relu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[0.000826, -0.651648, 0.075237, -0.055185, -0.17648], [-0.476082, -0.... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.6972275674343109, "train_acc": 0.565, "val_loss": 0.6631535291671753, "val_acc": 0.56}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.6444984972476959, "train_acc": 0.565, "val_loss": 0.5396676063537598, "v... |
7 | "{\"target_pattern\": \"no_repeats\", \"degraded_accuracy\": 0.48, \"improved_accuracy\": 0.9, \"imp(...TRUNCATED) | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | no_repeats | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | "{\"neuron_activations\": {\"0\": {\"neuron_profiles\": {\"0\": {\"mean\": 0.36353254318237305, \"st(...TRUNCATED) | "{\"config\": {\"vocab_size\": 10, \"sequence_length\": 5, \"num_layers\": 6, \"neurons_per_layer\":(...TRUNCATED) | "{\"training_history\": [{\"stage\": \"degraded\", \"epoch\": 0, \"global_epoch\": 0, \"train_loss\"(...TRUNCATED) |
8 | "{\"target_pattern\": \"has_majority\", \"degraded_accuracy\": 0.38, \"improved_accuracy\": 0.72, \"(...TRUNCATED) | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | has_majority | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | "{\"neuron_activations\": {\"0\": {\"neuron_profiles\": {\"0\": {\"mean\": -0.4140007495880127, \"st(...TRUNCATED) | "{\"config\": {\"vocab_size\": 10, \"sequence_length\": 5, \"num_layers\": 4, \"neurons_per_layer\":(...TRUNCATED) | "{\"training_history\": [{\"stage\": \"degraded\", \"epoch\": 0, \"global_epoch\": 0, \"train_loss\"(...TRUNCATED) |
9 | "{\"target_pattern\": \"decreasing_pairs\", \"degraded_accuracy\": 0.5, \"improved_accuracy\": 0.96,(...TRUNCATED) | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | decreasing_pairs | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | "{\"neuron_activations\": {\"0\": {\"neuron_profiles\": {\"0\": {\"mean\": 0.49381667375564575, \"st(...TRUNCATED) | "{\"config\": {\"vocab_size\": 10, \"sequence_length\": 5, \"num_layers\": 5, \"neurons_per_layer\":(...TRUNCATED) | "{\"training_history\": [{\"stage\": \"degraded\", \"epoch\": 0, \"global_epoch\": 0, \"train_loss\"(...TRUNCATED) |
End of preview. Expand in Data Studio
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 |
| Neurons per Layer | 5 to 8 |
| Activation Types | relu, gelu |
| Pattern Vocab Size | 10 |
| Pattern Sequence Len | 5 |
| Training Datasets | |
|---|---|
| Enabled Patterns | palindrome, sorted_ascending, sorted_descending, alternating, contains_abc, starts_with, ends_with, no_repeats, has_majority, increasing_pairs, decreasing_pairs, vowel_consonant, first_last_match, 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) |
- Downloads last month
- 6
Models trained or fine-tuned on maximuspowers/muat-mean-std