Good job
#1
by
Trilogix1
- opened
Hi, your work is interesting and actually works. It is quite expensive in tokens and is not 0 shot but it works. Mean that your method brings it very near to compete with way bigger models and even beat them.
Can you try with Instruct and thinking qwen models?
Qwen3 Coder 30b a3b instruct and Qwen3 thinking 2507 (30b a3b).
Thanks in advance.
Thank you for your interest! Actually the gpt-oss-20b models are experimental (not an official release) as there are some issues in the model since the training is quite expensive; we are still trying to figure this out, apologies for the confusion!
Actually, we have Qwen3 thinking models (but not the newer Instruct/Thinking model, working on those), we recommend you to try those! :)
Trilogix1
changed discussion status to
closed