Base Model: t5-base
Model Type: Text-to-JSON generation
Language: English
Task: Convert natural language music preferences into Spotify audio feature JSON objects
Fine-tuning Dataset: Custom dataset of prompts to Spotify audio features

Training Configuration

Epochs: 7
Learning Rate: 3e-4
Batch Size: 8 (per device)
Gradient Accumulation Steps: 4
Scheduler: Cosine with warmup
Optimizer: AdamW
Max Length: 256 tokens
Precision: bfloat16

Usage

python

from transformers import T5ForConditionalGeneration, T5Tokenizer
import json

# Load model and tokenizer
model = T5ForConditionalGeneration.from_pretrained("afsagag/t5-spotify-features")
tokenizer = T5Tokenizer.from_pretrained("afsagag/t5-spotify-features")

# Example usage
prompt = "I want energetic dance music with high energy and danceability"
input_text = f"prompt: {prompt}"

# Tokenize and generate
input_ids = tokenizer(input_text, return_tensors="pt", max_length=256, truncation=True).input_ids
outputs = model.generate(
    input_ids, 
    max_length=256, 
    num_beams=4, 
    early_stopping=True,
    do_sample=False
)

# Decode result
result = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(result)

# Parse JSON output
try:
    spotify_features = json.loads(result)
    print("Generated Spotify Features:", spotify_features)
except json.JSONDecodeError:
    print("Generated text is not valid JSON")

Expected Output Format

The model generates JSON objects with Spotify audio features:

json

{
  "danceability": 0.85,
  "energy": 0.90,
  "valence": 0.75,
  "acousticness": 0.15,
  "instrumentalness": 0.05,
  "speechiness": 0.08,
}

Metrics

Per-set Mean Absolute Error: Measures average prediction accuracy across feature sets
Per-set Root Mean Squared Error: Measures prediction variance
Per-feature Correlation: Pearson correlation for individual audio features

Model Files

config.json: Model configuration
pytorch_model.bin: Model weights
tokenizer.json: Tokenizer vocabulary
tokenizer_config.json: Tokenizer configuration
special_tokens_map.json: Special token mappings

Limitations

Model may occasionally generate invalid JSON that requires post-processing
Trained on specific prompt formats starting with "prompt: "
Performance depends on similarity to training data distribution
May not generalize well to very abstract or unusual music descriptions

Training Data

The model was trained on a custom dataset pairing natural language music descriptions with corresponding Spotify audio feature values.

Ethical Considerations

This model generates music preference predictions and should not be used as the sole basis for music recommendation systems without human oversight.