Qwen3.5-35B-A3B-Opus-Reasoning-Distilled-Uncensored

Model Description

This model is fine-tuned from Li101/Qwen3.5-35B-A3B-Uncensored-Aggressive-safetensors with enhanced reasoning capabilities.

Training Details

• Base Model: Li101/Qwen3.5-35B-A3B-Uncensored-Aggressive-safetensors
• Method: bf16 LoRA + response-only (train_on_responses_only)
• LoRA Rank: 16
• Epochs: 2
• Max Sequence Length: 4096
• Learning Rate: 2e-5
• Framework: Unsloth + TRL

Datasets

• nohurry/Opus-4.6-Reasoning-3000x-filtered
• Jackrong/Qwen3.5-reasoning-700x
• Roman1111111/claude-opus-4.6-10000x

Format

The model uses <think>...</think> tags for chain-of-thought reasoning.