Upload weight-space autoencoder (encoder + decoder) and configuration

README.md

@@ -0,0 +1,42 @@
+---
+tags:
+- weight-space-learning
+- neural-network-autoencoder
+- autoencoder
+- transformer
+datasets:
+- maximuspowers/muat-fourier-5
+---
+
+# Weight-Space Autoencoder (TRANSFORMER)
+
+This model is a weight-space autoencoder trained on neural network activation weights/signatures.
+It includes both an encoder (compresses weights into latent representations) and a decoder (reconstructs weights from latent codes).
+
+## Model Description
+
+- **Architecture**: Transformer encoder-decoder
+- **Training Dataset**: maximuspowers/muat-fourier-5
+- **Input Mode**: signature
+- **Latent Dimension**: 256
+
+## Tokenization
+
+- **Chunk Size**: 1 weight values per token
+- **Max Tokens**: 512
+- **Metadata**: True
+
+## Training Config
+
+- **Loss Function**: contrastive
+- **Optimizer**: adam
+- **Learning Rate**: 0.0001
+- **Batch Size**: 8
+
+## Performance Metrics (Test Set)
+
+- **MSE**: 0.298185
+- **MAE**: 0.404015
+- **RMSE**: 0.546063
+- **Cosine Similarity**: 0.6089
+- **R² Score**: 0.2872

config.yaml

@@ -0,0 +1,123 @@
+architecture:
+  latent_dim: 256
+  mlp:
+    decoder:
+      activation: relu
+      batch_norm: true
+      dropout: 0.2
+      hidden_dims:
+      - 256
+      - 384
+      - 512
+    encoder:
+      activation: relu
+      batch_norm: true
+      dropout: 0.2
+      hidden_dims:
+      - 512
+      - 384
+      - 256
+    token_pooling: mean
+  transformer:
+    decoder:
+      activation: relu
+      d_model: 512
+      dim_feedforward: 2048
+      dropout: 0.1
+      num_heads: 8
+      num_layers: 6
+    encoder:
+      activation: relu
+      d_model: 512
+      dim_feedforward: 2048
+      dropout: 0.1
+      num_heads: 8
+      num_layers: 6
+      pooling: mean
+      positional_encoding: learned
+  type: transformer
+dataloader:
+  num_workers: 0
+  pin_memory: true
+dataset:
+  hf_dataset: maximuspowers/muat-fourier-5
+  input_mode: signature
+  max_dimensions:
+    max_hidden_layers: 6
+    max_neurons_per_layer: 8
+    max_sequence_length: 5
+  neuron_profile:
+    methods:
+    - fourier
+  random_seed: 42
+  test_split: 0.1
+  train_split: 0.8
+  val_split: 0.1
+device:
+  type: auto
+evaluation:
+  metrics:
+  - mse
+  - mae
+  - rmse
+  - cosine_similarity
+  - relative_error
+  - r2_score
+  per_layer_metrics: false
+hub:
+  enabled: true
+  private: false
+  push_logs: true
+  push_metrics: true
+  push_model: true
+  repo_id: maximuspowers/sig-autoencoder-fourier-5-simclr-mse-new
+  token: <REDACTED>
+logging:
+  checkpoint:
+    enabled: true
+    mode: min
+    monitor: val_loss
+    save_best_only: true
+  tensorboard:
+    auto_launch: true
+    enabled: true
+    log_interval: 10
+    port: 6006
+  verbose: true
+loss:
+  augmentation_type: noise
+  contrast_type: simclr
+  dropout_prob: 0.1
+  gamma: 0.4
+  noise_std: 0.01
+  projection_head:
+    hidden_dim: 256
+    input_dim: 256
+    output_dim: 128
+  reconstruction_type: mse
+  temperature: 0.1
+  type: contrastive
+run_dir: /Users/max/Desktop/muat/model_zoo/runs/train-encoder-decoder_config_2025-12-14_00-00-05
+run_log_cleanup: false
+tokenization:
+  chunk_size: 1
+  granularity: neuron
+  include_metadata: true
+  max_tokens: 512
+training:
+  batch_size: 8
+  early_stopping:
+    enabled: true
+    mode: min
+    monitor: val_loss
+    patience: 5
+  epochs: 5
+  learning_rate: 0.0001
+  lr_scheduler:
+    enabled: true
+    factor: 0.5
+    min_lr: 1.0e-06
+    patience: 3
+    type: reduce_on_plateau
+  optimizer: adam
+  weight_decay: 0.0001

decoder.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a13ca74c3d1bffd112b0fc0156a7837c832fa3ad7514a0f4f72fa07cc6d053e2
+size 102545486

encoder.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d46982df3162def7569a6cf378470c8a9513c0b7c64934800135c11b3f85ae6
+size 77277228

tokenizer_config.json

@@ -0,0 +1,8 @@
+{
+  "chunk_size": 1,
+  "max_tokens": 512,
+  "include_metadata": true,
+  "metadata_features": 5,
+  "token_dim": 11,
+  "granularity": "neuron"
+}