distill_opus45_ours_3597

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

  • Data: opus45-ours (3,597 train examples, OpenHands format with <think> reasoning)
  • Best val loss: 0.140 (epoch 2)
  • LR: 2e-5, cosine, 3 epochs, batch=128, bf16
  • Template: Qwen3 native with thinking enabled
  • Tools: execute_bash, str_replace_editor, think, finish, execute_ipython_cell, task_tracker (all with security_risk)
Downloads last month
3
Safetensors
Model size
8B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for sqy201x/distill_opus45_ours_think_3597

Finetuned
Qwen/Qwen3-8B
Finetuned
(1450)
this model