Qwen3.5-9b-Sushi-Coder

Merged Unsloth fine-tune based on unsloth/qwen3.5-9b.

Training lineage

• Base model: unsloth/qwen3.5-9b
• Earlier training data used for this model line: open-r1/codeforces-cots
• Continuation training dataset: nohurry/Opus-4.6-Reasoning-3000x-filtered
• Continuation method: LoRA continuation from an adapter-only Unsloth Studio output
• Continuation precision: 16bit LoRA / bf16

Current uploaded files

• merged safetensor model shards
• tokenizer files
• processor and generation config
• chat template

Built with Unsloth and TRL.