example_id int64 0 3.52k | metadata stringlengths 680 725 | classification_prompt stringlengths 18k 44.2k | classification_completion stringclasses 14
values | classification_text stringlengths 18.1k 44.2k | improved_signature stringlengths 6.26k 11.4k | improved_model_weights stringlengths 9.45k 24.9k | training_metrics stringlengths 1.45k 2.92k |
|---|---|---|---|---|---|---|---|
0 | {"target_pattern": "sorted_descending", "degraded_accuracy": 0.52, "improved_accuracy": 0.72, "improvement": 0.19999999999999996, "model_config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 9, "neurons_per_layer": 10, "activation_type": "gelu", "dropout_rate": 0.0, "random_seed": 9016, "learning_rate": 0.089... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 9
Neurons per Layer: 10
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.423928,
0.599813,
0.26421,
-0.055279,
... | sorted_descending | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 9
Neurons per Layer: 10
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.423928,
0.599813,
0.26421,
-0.055279,
... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"mean": 2.5297226905822754, "std": 1.8762309551239014}, "1": {"mean": -3.9017319679260254, "std": 2.5156681537628174}, "2": {"mean": -1.4825490713119507, "std": 1.9457550048828125}, "3": {"mean": -1.8681204319000244, "std": 1.3154653310775757}, "4": {"mean": -1.68... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 9, "neurons_per_layer": 10, "activation_type": "gelu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[0.423928, 0.599813, 0.26421, -0.055279, 0.137147], [-0.06073, -0.75... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.7037525177001953, "train_acc": 0.465, "val_loss": 0.693569004535675, "val_acc": 0.52}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.6716509461402893, "train_acc": 0.565, "val_loss": 0.5419552326202393, "va... |
1 | {"target_pattern": "palindrome", "degraded_accuracy": 0.48, "improved_accuracy": 0.94, "improvement": 0.45999999999999996, "model_config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 10, "neurons_per_layer": 12, "activation_type": "relu", "dropout_rate": 0.0, "random_seed": 2679, "learning_rate": 0.030088966... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 10
Neurons per Layer: 12
Activation Function: relu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.327141,
0.268116,
-0.147093,
0.051027,... | palindrome | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 10
Neurons per Layer: 12
Activation Function: relu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.327141,
0.268116,
-0.147093,
0.051027,... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"mean": 0.6738900542259216, "std": 0.9914834499359131}, "1": {"mean": 0.6944124102592468, "std": 0.9108791947364807}, "2": {"mean": 0.25854384899139404, "std": 1.2264961004257202}, "3": {"mean": 0.6265138387680054, "std": 1.4541890621185303}, "4": {"mean": -1.6587... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 10, "neurons_per_layer": 12, "activation_type": "relu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[0.327141, 0.268116, -0.147093, 0.051027, 0.167262], [0.221227, 0.0... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.6993178725242615, "train_acc": 0.46, "val_loss": 0.6980134844779968, "val_acc": 0.48}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.6796618402004242, "train_acc": 0.58, "val_loss": 0.724160373210907, "val_... |
2 | "{\"target_pattern\": \"increasing_pairs\", \"degraded_accuracy\": 0.5, \"improved_accuracy\": 0.84,(...TRUNCATED) | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | increasing_pairs | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | "{\"neuron_activations\": {\"0\": {\"neuron_profiles\": {\"0\": {\"mean\": 0.7402181029319763, \"std(...TRUNCATED) | "{\"config\": {\"vocab_size\": 10, \"sequence_length\": 5, \"num_layers\": 8, \"neurons_per_layer\":(...TRUNCATED) | "{\"training_history\": [{\"stage\": \"degraded\", \"epoch\": 0, \"global_epoch\": 0, \"train_loss\"(...TRUNCATED) |
3 | "{\"target_pattern\": \"alternating\", \"degraded_accuracy\": 0.54, \"improved_accuracy\": 0.9, \"im(...TRUNCATED) | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | alternating | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | "{\"neuron_activations\": {\"0\": {\"neuron_profiles\": {\"0\": {\"mean\": -0.4563175141811371, \"st(...TRUNCATED) | "{\"config\": {\"vocab_size\": 10, \"sequence_length\": 5, \"num_layers\": 9, \"neurons_per_layer\":(...TRUNCATED) | "{\"training_history\": [{\"stage\": \"degraded\", \"epoch\": 0, \"global_epoch\": 0, \"train_loss\"(...TRUNCATED) |
4 | "{\"target_pattern\": \"palindrome\", \"degraded_accuracy\": 0.58, \"improved_accuracy\": 0.9, \"imp(...TRUNCATED) | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | palindrome | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | "{\"neuron_activations\": {\"0\": {\"neuron_profiles\": {\"0\": {\"mean\": 1.0724173784255981, \"std(...TRUNCATED) | "{\"config\": {\"vocab_size\": 10, \"sequence_length\": 5, \"num_layers\": 10, \"neurons_per_layer\"(...TRUNCATED) | "{\"training_history\": [{\"stage\": \"degraded\", \"epoch\": 0, \"global_epoch\": 0, \"train_loss\"(...TRUNCATED) |
5 | "{\"target_pattern\": \"decreasing_pairs\", \"degraded_accuracy\": 0.52, \"improved_accuracy\": 0.9,(...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\": -1.5801395177841187, \"st(...TRUNCATED) | "{\"config\": {\"vocab_size\": 10, \"sequence_length\": 5, \"num_layers\": 9, \"neurons_per_layer\":(...TRUNCATED) | "{\"training_history\": [{\"stage\": \"degraded\", \"epoch\": 0, \"global_epoch\": 0, \"train_loss\"(...TRUNCATED) |
6 | "{\"target_pattern\": \"first_last_match\", \"degraded_accuracy\": 0.48, \"improved_accuracy\": 0.88(...TRUNCATED) | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | first_last_match | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | "{\"neuron_activations\": {\"0\": {\"neuron_profiles\": {\"0\": {\"mean\": -2.2621214389801025, \"st(...TRUNCATED) | "{\"config\": {\"vocab_size\": 10, \"sequence_length\": 5, \"num_layers\": 9, \"neurons_per_layer\":(...TRUNCATED) | "{\"training_history\": [{\"stage\": \"degraded\", \"epoch\": 0, \"global_epoch\": 0, \"train_loss\"(...TRUNCATED) |
7 | "{\"target_pattern\": \"palindrome\", \"degraded_accuracy\": 0.48, \"improved_accuracy\": 0.96, \"im(...TRUNCATED) | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | palindrome | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | "{\"neuron_activations\": {\"0\": {\"neuron_profiles\": {\"0\": {\"mean\": 0.9704694151878357, \"std(...TRUNCATED) | "{\"config\": {\"vocab_size\": 10, \"sequence_length\": 5, \"num_layers\": 9, \"neurons_per_layer\":(...TRUNCATED) | "{\"training_history\": [{\"stage\": \"degraded\", \"epoch\": 0, \"global_epoch\": 0, \"train_loss\"(...TRUNCATED) |
8 | "{\"target_pattern\": \"no_repeats\", \"degraded_accuracy\": 0.6, \"improved_accuracy\": 0.78, \"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\": 2.34316349029541, \"std\"(...TRUNCATED) | "{\"config\": {\"vocab_size\": 10, \"sequence_length\": 5, \"num_layers\": 10, \"neurons_per_layer\"(...TRUNCATED) | "{\"training_history\": [{\"stage\": \"degraded\", \"epoch\": 0, \"global_epoch\": 0, \"train_loss\"(...TRUNCATED) |
9 | "{\"target_pattern\": \"ends_with\", \"degraded_accuracy\": 0.7, \"improved_accuracy\": 0.88, \"impr(...TRUNCATED) | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | ends_with | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | "{\"neuron_activations\": {\"0\": {\"neuron_profiles\": {\"0\": {\"mean\": 1.173392415046692, \"std\(...TRUNCATED) | "{\"config\": {\"vocab_size\": 10, \"sequence_length\": 5, \"num_layers\": 8, \"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 | configs/dataset_gen/signature_dataset.json |
| Model Architecture | |
|---|---|
| Number of Layers | 8 to 10 |
| Neurons per Layer | 10 to 15 |
| 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%) |
Token Count Statistics
| Task Type | Min Tokens | Max Tokens | Avg Tokens |
|---|---|---|---|
| Classification | 7699 | 18864 | 12619.8 |
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
- 9