Well done
#2
by
Trilogix1
- opened
I am still testing it, but overall looks good. I can see already your method working.
It didn't improve the 4b so much (but as I said I am still testing).
Can this be done to GPT-oss, more precisely to this model: https://huggingface.co/hokar3361/gpt-oss-coderjs-v0.1 ?
Applying your method will certainly improve the output but I am more curious if it will also improve the ctx length (this model, the gpt-oss is fast but working with ctx over 60k it loops).
Thanks for the good job.
Can you apply your method to: armand0e/gpt-oss-20b-glm-4.6-distill-GGUF ?