TFTMBI-180

TFT finetuned on Dipole datasets v9 for Dedicated MD 2025-08-13.

┏━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━┓
┃    ┃ Name                         ┃ Type                            ┃ Params ┃
┡━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━┩
│ 0  │ criterion                    │ QuantileLoss                    │      0 │
│ 1  │ model                        │ PFTemporalFusionTransformerMod… │  5.5 M │
│ 2  │ model.static_vs              │ VariableSelection               │  153 K │
│ 3  │ model.enc_vs                 │ VariableSelection               │  617 K │
│ 4  │ model.dec_vs                 │ VariableSelection               │  462 K │
│ 5  │ model.static_ctxt_vs         │ GatedResidualNetwork            │  263 K │
│ 6  │ model.static_ctxt_enrichment │ GatedResidualNetwork            │  263 K │
│ 7  │ model.lstm_init_hidden       │ GatedResidualNetwork            │  263 K │
│ 8  │ model.lstm_init_cell         │ GatedResidualNetwork            │  263 K │
│ 9  │ model.enc_lstm               │ LSTM                            │  1.1 M │
│ 10 │ model.dec_lstm               │ LSTM                            │  1.1 M │
│ 11 │ model.enc_gate1              │ GatedLinearUnit                 │  131 K │
│ 12 │ model.enc_norm1              │ AddNorm                         │    512 │
│ 13 │ model.static_enrichment      │ GatedResidualNetwork            │  329 K │
│ 14 │ model.attn                   │ InterpretableMultiHeadAttention │  148 K │
│ 15 │ model.attn_gate1             │ GatedLinearUnit                 │  131 K │
│ 16 │ model.attn_norm1             │ AddNorm                         │    512 │
│ 17 │ model.attn_grn               │ GatedResidualNetwork            │  263 K │
│ 18 │ model.attn_gate2             │ GatedLinearUnit                 │  131 K │
│ 19 │ model.attn_norm2             │ AddNorm                         │    512 │
│ 20 │ model.output_layer           │ Linear                          │  1.8 K │
└────┴──────────────────────────────┴─────────────────────────────────┴────────┘                                        Trainable params: 5.5 M
Non-trainable params: 0
Total params: 5.5 M
Total estimated model params size (MB): 22