ScaleAI

/

swe-oec-7b

like 1

Follow

Scale AI 346

Hi! Thank you for releasing this model.
I have a question about the details of training procedure. For student-7B stage-1 "SFT on 5k Claude-3.7-Sonnet SWE-agent trajectories", did you directly use the SWE-bench/SWE-agent-LM-7B ckpt or SFT from Qwen/Qwen2.5-Coder-7B-Instruct from scratch?
From the results, I reckon that student-7B is SFTed individually. If my understanding is correct, could you also release the stage-1 SFTed student-7B model? It would be very helpful for my research.
A million thanks for your kind answer.