curia007/pick3_llm_model

Idaho Pick 3 Transformer Model (idaho-pick3-transformer)

An autoregressive Causal Transformer trained on historical Idaho Pick 3 lottery draw sequences using Apple MLX and PyTorch.

Model Summary

• Architecture: Autoregressive Causal Decoder Transformer
• Vocabulary Size: 17 (Digits 0-9, <PAD>, <BOS>, <EOS>, <DAY>, <NIGHT>, <SEP>, <PRED>)
• Embedding Dimension: 64
• Heads: 4
• Layers: 3
• Max Sequence Length: 128
• Parameters: ~147.5k

Evaluation Metrics

• Validation Cross-Entropy Loss: N/A
• Token Accuracy: N/A
• Exact Draw Match Rate: N/A

Quickstart

Loading with Apple MLX

from idaho.pick3.pick3_llm_ticket_model import Pick3LLM, Pick3Tokenizer

# Load directly from Hugging Face Hub or local path
model = Pick3LLM.from_pretrained("idaho-pick3-transformer")
tokenizer = Pick3Tokenizer()

Loading with PyTorch / Hugging Face Transformers

from idaho.pick3.pick3_hf_model import Pick3ForCausalLM

model = Pick3ForCausalLM.from_pretrained("idaho-pick3-transformer")
model.eval()