v1.6.0 — 34 Models, Llama Support

What's New

Model Expansion: 11 → 34 supported models

New model families:

• Meta Llama — Llama 3.2 1B/3B, Llama 3.1 8B
• Qwen3 VL — Qwen3-VL-2B, Qwen3-VL-4B
• Qwen2.5 Math — Qwen2.5-Math-1.5B, Qwen2.5-Math-7B
• Qwen2.5 Coder — Qwen2.5-Coder-0.5B, Qwen2.5-Coder-3B
• Qwen2.5 VL — Qwen2.5-VL-3B, Qwen2.5-VL-7B
• Large models — Qwen3-4B, Qwen3-8B, Qwen3.5-27B

Registry

• 14 new auto-generated configs with correct scaling parameters
• 40 models tracked in manifest (11 verified, 19 config-ready)
• LoRA auto-enabled for 7B+, batch sizes scaled by model size

Full stats

• 17 algorithms (5 RL + 12 SL)
• 34 listed models (15 verified + 19 config-ready)
• 64 tests passing

Full Changelog: v1.5.0...v1.6.0