How does this compare to deepseekcoder?
Im curious how this compared to the deepseek-coder 33b model? They say that model can outperform gpt-3.5 and even get close to gpt-4
Oooo man im testing them both right now, same quant and settings and they are very similar. I do want to say that xwincoder is slightly better.
@dillfrescott I did my own testing, and although the code output seems ok, idk the logic doesnt seem very good. The model also refuses to admit that its code is incorrect alot and refuses to fix it. Deepseek coder does a great job troubleshooting its own code. I think i give the win to deepseek coder here. Ive tested the 13b xwincoder, 34b xwin coder, 6.7 deepseek and 33b deepseek, and only the 33b deepseek-coder-instruct model has impressed me at all. Its the closest ive seen to gpt-3.5.
Oh cool!