main_rev2_sft05

This is a Safe SFT LoRA adapter (REV2 SFT05). It uses Completion-only Training and TOML Refinement Filtering.

Base Model

Qwen/Qwen3-4B-Instruct-2507

Training Data (Mixed 65:35, TOML <= 10%)

• 65%: daichira/structured-hard-sft-4k (Filtered + Refined TOML)
• 35%: u-10bei/structured_data_with_cot_dataset_512_v4 (Filtered + Refined TOML)

TOML Refinement Applied

• Eliminated YAML-like lists, Big Arrays, Log keywords.
• Enforced valid TOML syntax (toml.loads).
• Controlled TOML Ratio to max 10% of total dataset.

Method

• Completion-only: User prompts are masked.
• Marker: `

OUTPUT

`.

• Config: 1 Epoch, Max Seq Length 4096.