Note Core MyPO DPO dataset: chosen vs rejected Python code quality pairs.

Note Legacy validation comparison subset for the Phi-3 MyPO line.

joshuasundance/phi3-mini-4k-qlora-python-code-20k-mypo-4k-rfc

Updated Jul 13, 2024 • 1

Note Legacy Phi-3 MyPO adapter experiment.

joshuasundance/phi3-mini-4k-qlora-python-code-20k-mypo-4k-rfc-full

Text Generation • 4B • Updated Jul 14, 2024 • 13

Note Legacy Phi-3 full/merged MyPO model.

joshuasundance/phi3-mini-4k-qlora-python-code-20k-mypo-4k-rfc-pipe

Text Generation • 4B • Updated Jul 14, 2024 • 5 • 1

Note Legacy Phi-3 pipeline-export MyPO model used in early comparisons.

joshuasundance/mypo-training

Updated Apr 23

Note Training scripts, reports, evaluation artifacts, and durable HumanEval+ pipeline source of truth.

joshuasundance/mypo-qwen2.5-coder-1.5b-sft

Text Generation • Updated Apr 23 • 20

Note Qwen 2.5 Coder SFT LoRA adapter for typed Python generation.

joshuasundance/mypo-qwen2.5-coder-1.5b-dpo-v2

Text Generation • Updated Apr 23 • 29

Note Qwen 2.5 Coder DPO v2 LoRA adapter; reproducibility artifact and near-base baseline.

joshuasundance/mypo-qwen2.5-coder-1.5b-dpo-v3

Text Generation • 2B • Updated Apr 23 • 15

Note Qwen 2.5 Coder DPO v3 merged model; strongest in-domain type-hint preference model in the project.