maximuspowers/muat-pca-10-classifier
Updated • 1
example_id int64 0 8.3k | metadata stringlengths 680 723 | classification_prompt stringlengths 5.95k 14.2k | classification_completion stringclasses 14
values | classification_text stringlengths 5.97k 14.3k | improved_signature stringlengths 3.74k 10.5k | improved_model_weights stringlengths 1.77k 5.04k | training_metrics stringlengths 1.46k 2.92k |
|---|---|---|---|---|---|---|---|
0 | {"target_pattern": "decreasing_pairs", "degraded_accuracy": 0.74, "improved_accuracy": 0.9, "improvement": 0.16000000000000003, "model_config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 4, "neurons_per_layer": 8, "activation_type": "gelu", "dropout_rate": 0.0, "random_seed": 2190, "learning_rate": 0.011011... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 4
Neurons per Layer: 8
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.220921,
0.484456,
0.047179,
0.350972,
... | decreasing_pairs | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 4
Neurons per Layer: 8
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.220921,
0.484456,
0.047179,
0.350972,
... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"pca": [-0.29780128598213196, 0.430982768535614, -0.24376456439495087, -0.06820584088563919, 0.318475604057312, 0.7214001417160034, 0.09721335768699646, -0.17378076910972595, 0.0, 0.0]}, "1": {"pca": [0.5398541688919067, -0.20937564969062805, 0.035204656422138214,... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 4, "neurons_per_layer": 8, "activation_type": "gelu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[0.220921, 0.484456, 0.047179, 0.350972, -0.296939], [-0.146521, -0.4... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.6898400485515594, "train_acc": 0.485, "val_loss": 0.6638575792312622, "val_acc": 0.72}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.6750549972057343, "train_acc": 0.555, "val_loss": 0.6391302347183228, "v... |
1 | {"target_pattern": "alternating", "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": 8, "activation_type": "gelu", "dropout_rate": 0.0, "random_seed": 8208, "learning_rate": 0.09630645140215274, "batch_... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 6
Neurons per Layer: 8
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
-0.537074,
-0.188148,
-0.718552,
-0.454124... | alternating | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 6
Neurons per Layer: 8
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
-0.537074,
-0.188148,
-0.718552,
-0.454124... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"pca": [0.385761022567749, 0.37568750977516174, -0.10415489226579666, -0.14121423661708832, -0.28748172521591187, 0.2152726948261261, -0.3260464370250702, 0.666305661201477, 0.0, 0.0]}, "1": {"pca": [0.10234690457582474, 0.2929198741912842, -0.07823708653450012, 0... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 6, "neurons_per_layer": 8, "activation_type": "gelu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[-0.537074, -0.188148, -0.718552, -0.454124, -0.657402], [-0.138443, ... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.6876776218414307, "train_acc": 0.49, "val_loss": 0.8729578256607056, "val_acc": 0.48}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.7320164144039154, "train_acc": 0.57, "val_loss": 0.6967922449111938, "val... |
2 | {"target_pattern": "sorted_descending", "degraded_accuracy": 0.46, "improved_accuracy": 0.92, "improvement": 0.46, "model_config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 6, "neurons_per_layer": 8, "activation_type": "gelu", "dropout_rate": 0.0, "random_seed": 1240, "learning_rate": 0.08278571507486415, ... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 6
Neurons per Layer: 8
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.655552,
0.340664,
0.13419,
0.155989,
... | sorted_descending | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 6
Neurons per Layer: 8
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.655552,
0.340664,
0.13419,
0.155989,
... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"pca": [-0.3029038906097412, 0.5153995752334595, 0.19240184128284454, -0.11680468916893005, -0.1439746618270874, 0.5798109769821167, -0.23561573028564453, 0.42370566725730896, 0.0, 0.0]}, "1": {"pca": [-0.009528218768537045, -0.22233711183071136, 0.514887750148773... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 6, "neurons_per_layer": 8, "activation_type": "gelu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[0.655552, 0.340664, 0.13419, 0.155989, 0.180451], [-0.243624, -0.369... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.6729995310306549, "train_acc": 0.585, "val_loss": 0.7392985820770264, "val_acc": 0.46}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.6237123608589172, "train_acc": 0.585, "val_loss": 0.5784333348274231, "v... |
3 | {"target_pattern": "contains_abc", "degraded_accuracy": 0.46, "improved_accuracy": 0.96, "improvement": 0.49999999999999994, "model_config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 6, "neurons_per_layer": 6, "activation_type": "gelu", "dropout_rate": 0.0, "random_seed": 2065, "learning_rate": 0.039792056... | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 6
Neurons per Layer: 6
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.08475,
0.723148,
-0.511612,
0.311441,
... | contains_abc | ## Model Architecture
Input Size: 5 (integer indices for 5 sequence positions, vocab size 10)
Hidden Layers: 6
Neurons per Layer: 6
Activation Function: gelu
Dropout Rate: 0.0
## Model Weights
The trained model weights:
{
"network.0.weight": [
[
0.08475,
0.723148,
-0.511612,
0.311441,
... | {"neuron_activations": {"0": {"neuron_profiles": {"0": {"pca": [0.025776227936148643, 0.8271920680999756, 0.006233554799109697, -0.42406338453292847, -0.35221344232559204, -0.10566909611225128, 0.0, 0.0, 0.0, 0.0]}, "1": {"pca": [0.372446745634079, 0.264452189207077, -0.2578968107700348, 0.6925085783004761, -0.04195741... | {"config": {"vocab_size": 10, "sequence_length": 5, "num_layers": 6, "neurons_per_layer": 6, "activation_type": "gelu", "dropout_rate": 0.0, "precision": "float32", "input_size": 5, "input_format": "integer_indices"}, "weights": {"network.0.weight": [[0.08475, 0.723148, -0.511612, 0.311441, -0.507764], [0.799684, 0.182... | {"training_history": [{"stage": "degraded", "epoch": 0, "global_epoch": 0, "train_loss": 0.6875585615634918, "train_acc": 0.59, "val_loss": 0.7066342234611511, "val_acc": 0.46}, {"stage": "degraded", "epoch": 1, "global_epoch": 1, "train_loss": 0.6787996888160706, "train_acc": 0.59, "val_loss": 0.711582362651825, "val_... |
4 | "{\"target_pattern\": \"has_majority\", \"degraded_accuracy\": 0.58, \"improved_accuracy\": 0.76, \"(...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\": {\"pca\": [0.7428345084190369, 0.444(...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) |
5 | "{\"target_pattern\": \"ends_with\", \"degraded_accuracy\": 0.76, \"improved_accuracy\": 0.92, \"imp(...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\": {\"pca\": [-0.3935125470161438, -0.4(...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) |
6 | "{\"target_pattern\": \"mountain_pattern\", \"degraded_accuracy\": 0.5, \"improved_accuracy\": 0.92,(...TRUNCATED) | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | mountain_pattern | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | "{\"neuron_activations\": {\"0\": {\"neuron_profiles\": {\"0\": {\"pca\": [0.47299471497535706, -0.3(...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) |
7 | "{\"target_pattern\": \"palindrome\", \"degraded_accuracy\": 0.48, \"improved_accuracy\": 0.78, \"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\": {\"pca\": [0.5055093765258789, 0.534(...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) |
8 | "{\"target_pattern\": \"contains_abc\", \"degraded_accuracy\": 0.56, \"improved_accuracy\": 0.94, \"(...TRUNCATED) | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | contains_abc | "## Model Architecture\nInput Size: 5 (integer indices for 5 sequence positions, vocab size 10)\nHid(...TRUNCATED) | "{\"neuron_activations\": {\"0\": {\"neuron_profiles\": {\"0\": {\"pca\": [-0.4225125312805176, 0.11(...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) |
9 | "{\"target_pattern\": \"no_repeats\", \"degraded_accuracy\": 0.44, \"improved_accuracy\": 0.84, \"im(...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\": {\"pca\": [0.3385349214076996, -0.07(...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 | pca |
| 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%) |
Token Count Statistics
| Task Type | Min Tokens | Max Tokens | Avg Tokens |
|---|---|---|---|
| Classification | 2628 | 6519 | 4316.2 |
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
- 4
Models trained or fine-tuned on maximuspowers/muat-pca-10