Model hallucination and performance degradation
#30
by
snxtyle
- opened
We hosted the GLM 4.6 using vllm and also we have official api key, but the response we are getting from both the models are totally different, also the number of file changes. So do the model have set of different params that were used or some kind of prompt injection done. Or is their something we don't know cause its quality also degrades after a while making it hallucinate and make unnecessary file changes creating noise in code.