Good job!
Awesome job! v1.3 has been my default model for creative writing and story boarding since the initial release.
Edit:
V2 handles long context tasks better than v1.3. Prior, the v1.3 model would begin showing signs of hallucinating after 20k tokens mark.
Glad to hear!
Can't say I have the same luck. Around 20k context, model hallucinates, makes up random facts out of nowhere, constantly confuses You/I in narrative. I used recommended settings from json, and tried to switch between DM / storywriter / storyteller. I also played around with temperature and minp. No luck at all.
(P.S. Using Q5_K_M GGUF and kv cache quantization Q8_0, context size limit 32768)
Can't say I have the same luck. Around 20k context, model hallucinates, makes up random facts out of nowhere, constantly confuses You/I in narrative. I used recommended settings from json, and tried to switch between DM / storywriter / storyteller. I also played around with temperature and minp. No luck at all.
Are you using the recommended template? It's mistral v7 not mistral v7 tekken
Can't say I have the same luck. Around 20k context, model hallucinates, makes up random facts out of nowhere, constantly confuses You/I in narrative. I used recommended settings from json, and tried to switch between DM / storywriter / storyteller. I also played around with temperature and minp. No luck at all.
Are you using the recommended template? It's mistral v7 not mistral v7 tekken
Yes, straight from recommended .json of model card.