LocalLLaMA

1 readers

1 users here now

Community to discuss about Llama, the family of large language models created by Meta AI.

founded 10 months ago

MODERATORS

communick@poweruser.forum

7B models keep repeating/glitching after certain number of tokens (alien.top)

submitted 10 months ago by GustavoToyota@alien.top to c/localllama@poweruser.forum

4 comments fedilink hide all child comments

I'm using ollama and I have a RTX 3060 TI. Using only 7B models.

I tested with Mistral 7B, Mistral-OpenOrca and Zephyr, they all had the same problem where they kept repeating or speaking randomly after some amount of chatting.

What could it be? Temperature? VRAM? ollama?

you are viewing a single comment's thread
view the rest of the comments

[–] LocoLanguageModel@alien.top 1 points 9 months ago

I just posted this somewhere else but it seems relevant, try KoboldCPP, it has this feature enabled by default:

Context Shifting is a better version of Smart Context that only works for GGUF models. This feature utilizes KV cache shifting to automatically remove old tokens from context and add new ones without requiring any reprocessing. So long as you use no memory/fixed memory and don't use world info, you should be able to avoid almost all reprocessing between consecutive generations even at max context. This does not consume any additional context space, making it superior to SmartContext. Context Shifting is enabled by default, and will override smartcontext if both are enabled. Your outputs may be different with shifting enabled, but both seem equally coherent. To disable Context Shifting, use the flag --noshift.