datacrystals / AIStoryWriter

LLM story writer with a focus on high-quality long output based on a user provided prompt.
GNU Affero General Public License v3.0
63 stars 17 forks source link

Problems with long Promts #64

Open 72Joe72 opened 2 months ago

72Joe72 commented 2 months ago

Hello, unfortunately I have a problem with the script, apparently there is an error when converting to JSON format if the prompt is too long.

I have attached the main log file and the logs where the error occurred: Main.log

43_SafeGenerateText.SafeGenerateJSON.ScenesToJSON.ChapterByScene.GenerateChapter.Main.json 43_SafeGenerateText.SafeGenerateJSON.ScenesToJSON.ChapterByScene.GenerateChapter.Main.md 44_SafeGenerateJSON.ScenesToJSON.ChapterByScene.GenerateChapter.Main.json 44_SafeGenerateJSON.ScenesToJSON.ChapterByScene.GenerateChapter.Main.md

It doesn't crash but is trapped in an endless loop until I close the console window. I have already tried different models. The error also occurs with the "ExamplePrompts\Example2" prompt. I use Ollama with my own computer, this is the start code:

python Write.py -Seed 12 -Prompt Prompts/8.txt -InitialOutlineModel phi3:14b -ChapterOutlineModel phi3:14b -ChapterS1Model phi3:14b -ChapterS2Model phi3:14b -ChapterS3Model aya:8b -ChapterS4Model llava:13b -ChapterRevisionModel phi3:14b -RevisionModel llama3 -EvalModel llama3 -InfoModel llama3 -NoScrubChapters -Debug

datacrystals commented 2 weeks ago

Ah yeah that might depend on the model you're using, phi seems to have issues, perhaps try Mistral large?