Documentation
Model Documentation
Technical Details
The model has been pre-trained on a full corpus of Taco's lyrics from Genius. It has been fine-tuned to handle markup tags, which are treated as single, indivisible tokens. This allows the model to perfectly understand the song structure without wasting token limits on technical characters.
How to Feed the Model
You can write freely or force a specific section using tags:
// Free style (completion)
Wracam ekspresem do Warszawy...
Wracam ekspresem do Warszawy...
// Structural style (tags)
<title>ZAKOCHAŁEM SIĘ POD APTEKĄ</title>
<verse>Był rok dwa tysiące dwunasty...</verse>
<title>ZAKOCHAŁEM SIĘ POD APTEKĄ</title>
<verse>Był rok dwa tysiące dwunasty...</verse>
Temperature (Creativity)
A parameter controlling the randomness of the model. Choose your vibe:
0.01 - 0.50Cool RationalistConsistent, logical, repetitive. Low risk.
0.51 - 1.10Classic StorytellingThe middle ground. Balance between rhyme and reason.
1.11 - 1.70Warsaw Night LifeChaos, abstraction, hallucinations, and strange metaphors.
Token Count
Defines the limit of the generated text. For simplification:1 Token ≈ 1 Letter OR 1 Tag.
API Limit: 2048 tokens