Improve time-series transformer cfg to increase number of attention h…

…ead for XTab

Since XTab uses tokens of much bigger size (192 vs 8), increase the number of heads in multi-head attention for the time-series transformer when used w/ XTab

didactic/config/task/img_tokenizer/model/time-series-transformer.yaml

@@ -30,7 +30,7 @@ _args_:
       encoder_layer:
         _target_: torch.nn.TransformerEncoderLayer
         d_model: ${task.embed_dim}
-        nhead: 2
+        nhead: ${oc.select:task.model.encoder.attention_n_heads,${oc.select:task.model.encoder.encoder_layer.nhead,???}}
         dim_feedforward: ${op.mul:1.5,${task.embed_dim},int}
         dropout: 0.1
         activation: relu