Table 1.

Hyperparameters used for the Llamol model

Hyperparameters
Parameter/model	Llamol
Number of attention-heads $n_{\mathit{heads}}$	8
Number of decoder-blocks	8
Dropout probability	10%
Activation function	SwiGLU
Positional embeddings	RoPe
Embedding dimension $d_{\mathit{emb}}$	384
FFN hidden dim $d_{\mathit{ffn}}$	1024
Vocabulary size $d_{\mathit{voc}}$	591
Max SMILES length	256