1. When they plot against me with treachery, I confine them
With walls of saber blows and a trench.

2. And a heart bids me farewell there, a heartbroken bidder farewell,
And the shadow of death, dire and truthful.

3. It weighs over twenty raṭls, and likewise
It quivers lightly from the pommel and elegance.

4. In its width, a span, and its length
Five handspans with a separated span.

5. If it did not split the skulls, it would scrape them,
Various scrapings, from palms and wrists.

6. It has a slant after striking as if it were
The murmur of youth between burning embers.

7. So whomsoever it missed, it condemned, and whichever of them
It hit, it left him a seed that did not evolve.

8. As if meeting the blade as it scraped him
Was like meeting a sharp splitting rock.

9. As if when they circled around my side,
They circled around a pillar of the most unyielding character.

10. The wings of birds slip from its swipes,
Scented with a star in the darkest night.

11. So when they saw the look of the bare sword,
They prayed with a burning flame of fire.

12. They fled, and they sneaked away from my severity,
And whoever fears me will win the race.

13. And Aḥmus, the party of God, chased them
Beating whenever the mounts quickened pace.

14. So they surrendered with their hands and cast down their weapons,
And the call to capture spread—shackling is best.

15. I rammed the head of the Turk, and it clouded over for him
Beneath the woven mail coat's dazzling shine.

16. I adorned him while the patterned scabbard draped down
With the gaze of a blue-eyedInitialization vector: vector of parameters to initialize the model's state. Used for generating embeddings.

17. vocab_size: Vocabulary size.
category_vocab_size: Category vocabulary size for knowledge graph.

18. category_num: Number of categories, including root category.
dim: Dimension of the word and category embedding.

19. gru_dim: Dimension of the GRU layers.
projection_dim: Dimension of the MLP projection layer.

20. mlp_dim: Dimension of the MLP layers.
attn_dim: Dimension of the attention layers.

21. num_heads: Number of heads for multi-head attention.
dropout: Dropout rate.

22. max_seq_len: Maximum sequence length.
first_layer: First GRU layer to add position embeddings.