Xiaolong-Qwen3-8B

Xiaolong is a small, uncensored, reasoning-focused model finetuned using ORPO and QLoRA on top of Qwen3-8B-abliterated-TIES.

Finetuning Details

• Method: ORPO
• Epochs: 2
• Learning Rate: 5e-6, cosine decay w/ 5% warmup
• Batch Size: 1 x 32 (32 effective)
• Max Grad Norm: 0.3
• LoRA Rank: 64
• Hardware: 1x NVIDIA RTX A6000

Dataset Composition

~9,100 samples. 3,000 used Chain of Thought reasoning.

• nbeerbower/GreatFirewall-DPO
• nbeerbower/Schule-DPO
• nbeerbower/Purpura-DPO
• nbeerbower/Arkhaios-DPO
• jondurbin/truthy-dpo-v0.1
• antiven0m/physical-reasoning-dpo
• flammenai/Date-DPO-NoAsterisks
• flammenai/Prude-Phi3-DPO
• Atsunori/HelpSteer2-DPO (1000 samples)
• jondurbin/gutenberg-dpo-v0.1
• nbeerbower/gutenberg2-dpo
• nbeerbower/gutenberg-moderne-dpo

Chain of Thought

• GeneralReasoning/GeneralThought-430K (1000 samples)
• nvidia/OpenMathReasoning (1000 samples)
• nvidia/OpenCodeReasoning (1000 samples)