Modelpublic

meangrinch/Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distill

sourceHugging Faceapache-2.0updated 7mo agoView on Hugging Face

2likes14downloads

Model Card

Qwen3.5-4B-Claude-4.6-Opus-Reasoning-Distill

Fine-tuned reasoning model distilled from Claude 4.6 Opus chain-of-thought traces.

Setting	Value
Dataset	opus4.6reasoning1579xfiltered.jsonl
Method	SFTTrainer
LoRA Rank	rank=16, alpha=16
Epochs	3
Warmup steps	30
Effective batch size	16
Learning rate	1e-4
Scheduler	cosine
Max sequence length	4096