distill_opus45_ours_lr1e4_3524

Qwen3-8B fine-tuned for autonomous coding via OpenHands agent framework.

Training Details

• Base model: Qwen3-8B
• Data: opus45_ours_lr1e4 (3524 training examples)
• Format: OpenHands free-text function calling with <think> reasoning
• Best val loss: 0.1766
• Checkpoint: global_step_56
• LR: 1e-4, cosine schedule
• Epochs: 3, batch=128, bf16

Usage with vLLM

python -m vllm.entrypoints.openai.api_server \
    --model sqy201x/distill_opus45_ours_lr1e4_3524 \
    --max-model-len 32768 \
    --enforce-eager

No --enable-auto-tool-choice needed — model outputs <function=...> as plain text.