Upload weight-space autoencoder (encoder + decoder) and configuration
Browse files- README.md +6 -6
- config.yaml +8 -3
- decoder.pt +2 -2
- encoder.pt +2 -2
README.md
CHANGED
|
@@ -27,15 +27,15 @@ It includes both an encoder (compresses weights into latent representations) and
|
|
| 27 |
|
| 28 |
## Training Config
|
| 29 |
|
| 30 |
-
- **Loss Functions**: reconstruction,
|
| 31 |
- **Optimizer**: adamw
|
| 32 |
- **Learning Rate**: 0.0001
|
| 33 |
- **Batch Size**: 32
|
| 34 |
|
| 35 |
## Performance Metrics (Test Set)
|
| 36 |
|
| 37 |
-
- **MSE**: 0.
|
| 38 |
-
- **MAE**: 0.
|
| 39 |
-
- **RMSE**: 0.
|
| 40 |
-
- **Cosine Similarity**: 0.
|
| 41 |
-
- **R² Score**:
|
|
|
|
| 27 |
|
| 28 |
## Training Config
|
| 29 |
|
| 30 |
+
- **Loss Functions**: reconstruction, functional
|
| 31 |
- **Optimizer**: adamw
|
| 32 |
- **Learning Rate**: 0.0001
|
| 33 |
- **Batch Size**: 32
|
| 34 |
|
| 35 |
## Performance Metrics (Test Set)
|
| 36 |
|
| 37 |
+
- **MSE**: 0.122959
|
| 38 |
+
- **MAE**: 0.256139
|
| 39 |
+
- **RMSE**: 0.350655
|
| 40 |
+
- **Cosine Similarity**: 0.0575
|
| 41 |
+
- **R² Score**: 0.0069
|
config.yaml
CHANGED
|
@@ -12,6 +12,7 @@ architecture:
|
|
| 12 |
dropout: 0.025
|
| 13 |
num_heads: 8
|
| 14 |
num_layers: 6
|
|
|
|
| 15 |
encoder:
|
| 16 |
activation: gelu
|
| 17 |
d_model: 512
|
|
@@ -78,7 +79,7 @@ logging:
|
|
| 78 |
verbose: true
|
| 79 |
loss:
|
| 80 |
contrastive:
|
| 81 |
-
enabled:
|
| 82 |
projection_head:
|
| 83 |
hidden_dim: 128
|
| 84 |
input_dim: 256
|
|
@@ -89,12 +90,16 @@ loss:
|
|
| 89 |
benchmark_path: /Users/max/Desktop/muat/model_zoo/configs/autoencoder/benchmark_dataset.json
|
| 90 |
enabled: true
|
| 91 |
test_samples: null
|
| 92 |
-
weight: 0.
|
| 93 |
reconstruction:
|
| 94 |
enabled: true
|
| 95 |
type: mse
|
| 96 |
weight: 0.6
|
| 97 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 98 |
run_log_cleanup: false
|
| 99 |
tokenization:
|
| 100 |
chunk_size: 1
|
|
|
|
| 12 |
dropout: 0.025
|
| 13 |
num_heads: 8
|
| 14 |
num_layers: 6
|
| 15 |
+
num_memory_tokens: 8
|
| 16 |
encoder:
|
| 17 |
activation: gelu
|
| 18 |
d_model: 512
|
|
|
|
| 79 |
verbose: true
|
| 80 |
loss:
|
| 81 |
contrastive:
|
| 82 |
+
enabled: false
|
| 83 |
projection_head:
|
| 84 |
hidden_dim: 128
|
| 85 |
input_dim: 256
|
|
|
|
| 90 |
benchmark_path: /Users/max/Desktop/muat/model_zoo/configs/autoencoder/benchmark_dataset.json
|
| 91 |
enabled: true
|
| 92 |
test_samples: null
|
| 93 |
+
weight: 0.05
|
| 94 |
reconstruction:
|
| 95 |
enabled: true
|
| 96 |
type: mse
|
| 97 |
weight: 0.6
|
| 98 |
+
variance:
|
| 99 |
+
enabled: false
|
| 100 |
+
target_variance: 0.01
|
| 101 |
+
weight: 0.1
|
| 102 |
+
run_dir: /Users/max/Desktop/muat/model_zoo/runs/train-encoder-decoder_config_2025-12-20_12-55-25
|
| 103 |
run_log_cleanup: false
|
| 104 |
tokenization:
|
| 105 |
chunk_size: 1
|
decoder.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2e8076647a75544334db431a6665fce841f654289b2b7d28f838389b84177a6c
|
| 3 |
+
size 82225188
|
encoder.pt
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a92e43fd2ef0647bed913fcabcf216fa2a8333b8044f8aa9611f62a0dc003fc9
|
| 3 |
+
size 126816434
|